Virtual resource searching method and device, model training method and device, equipment and medium

By entering numerical change information into the resource interface and using the resource search model, the numerical trend matching of virtual resources is automatically searched, which solves the problem of low search efficiency of virtual resources and achieves fast, convenient and accurate resource positioning.

CN118915932APending Publication Date: 2024-11-08腾安基金销售(深圳)有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411002287.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the prior art, the search efficiency of virtual resources is low, and users need to manually search in a large number of virtual resources, resulting in cumbersome and inefficient search.

Method used

A virtual resource search method is provided. By entering numerical change information into the resource interface, displaying the trend of resource values ​​changing with time, and using the resource search model to match the input trends, and automatically searching for virtual resources with similar trends. The method includes displaying a resource interface, responding to information input operations, determining the detection box size, acquiring a reference image, performing object detection and matching degree evaluation to train a resource search model.

Benefits of technology

It realizes fast and convenient search of virtual resources, and can find matching resources without knowing the resource name, improving search efficiency and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915932A_ABST
    Figure CN118915932A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses a virtual resource searching method and device, a model training method and device, equipment and a medium, and belongs to the technical field of computers. The method comprises the steps of displaying a resource interface; in response to an information input operation in the resource interface, displaying input numerical value change information in the resource interface; and in response to a resource search operation for the numerical value change information, displaying a target virtual resource in the resource interface. According to the embodiment of the invention, a new virtual resource search mode is realized, the virtual resource which the user wants to check can be quickly searched, and the search of the virtual resource can be realized even if the name of the virtual resource cannot be known, so that the convenience and accuracy of the search of the virtual resource are ensured, and the resource search efficiency is also ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to a virtual resource search method, model training method, device, equipment and medium. Background Art

[0002] With the development of computer technology, there are more and more types of virtual resources. Usually, users can view any virtual resource through a device. When a user wants to view a virtual resource, the user needs to browse multiple virtual resources to find the virtual resource that the user wants to view. However, this virtual resource search method requires the user to manually search among a large number of virtual resources, and the search efficiency of virtual resources is low. Summary of the invention

[0003] The embodiment of the present application provides a virtual resource search method, model training method, device, equipment and medium, which can ensure the convenience of virtual resource search and the efficiency of resource search. The technical solution is as follows:

[0004] In one aspect, a virtual resource search method is provided, the method comprising:

[0005] Display the resource interface;

[0006] In response to the information input operation in the resource interface, the input value change information is displayed in the resource interface, and the value change information is used to indicate the change trend of the resource value of the virtual resource to be searched over time;

[0007] In response to a resource search operation for the value change information, a target virtual resource is displayed in the resource interface, and a similarity between a change trend of the target virtual resource and a change trend represented by the value change information is greater than a threshold.

[0008] On the other hand, a model training method is provided, the method comprising:

[0009] Determining a plurality of detection frame sizes based on bounding frame sizes of curves in a plurality of first sample images, wherein the curves are used to represent a trend of a change in a resource value of a virtual resource over time, and the bounding frame sizes are sizes of bounding frames of the curves in the first sample images;

[0010] Acquire a reference image and a second sample image, wherein a first curve in the reference image is used to represent a change trend of a resource value of a virtual resource to be searched over time, and a similarity between the first curve in the reference image and a second curve in the second sample image is greater than a threshold;

[0011] By means of a resource search model, based on the multiple detection frame sizes, target detection is performed on the reference image and the second sample image respectively to obtain a first local image and a second local image, wherein the first local image is a local image of the reference image including the first curve, and the second local image is a local image of the second sample image including the second curve;

[0012] Determining the matching degree between the first partial image and the second partial image by using the resource search model;

[0013] Based on the matching degree, the resource search model is trained.

[0014] In a possible implementation, before exchanging partial data at the same position in the two candidate sizes to obtain two new candidate sizes, the method further includes:

[0015] In the case of selecting the two candidate sizes, generating an exchange probability, the exchange probability indicating the possibility of exchanging part of the data at the same position in the two candidate sizes;

[0016] The step of exchanging part of the data at the same position in the two candidate sizes to obtain two new candidate sizes includes:

[0017] When the swap probability is greater than a first probability threshold, partial data at the same position in the two candidate sizes are swapped to obtain two new candidate sizes.

[0018] In another possible implementation, before the reverse processing of part of the data in the candidate size to obtain a new candidate size, the method further includes:

[0019] Determine a mutation probability corresponding to the candidate size, the mutation probability indicating a possibility of performing reverse processing on a portion of the data in the candidate size;

[0020] The step of performing reverse processing on part of the data in the candidate size to obtain a new candidate size includes:

[0021] When the mutation probability is greater than the second probability threshold, a portion of the data in the candidate size is processed in reverse order to obtain a new candidate size.

[0022] In another possible implementation, when the mutation probability is greater than a second probability threshold, before reversing the order of some data in the candidate size to obtain a new candidate size, the method further includes:

[0023] In the process of performing multiple iterative updates on the multiple candidate size sets, the second probability threshold is determined based on a third probability threshold, a current number of iterations, and a total number of iterations, and the third probability threshold is the probability threshold when the multiple candidate size sets are updated for the first time.

[0024] In another aspect, a virtual resource search device is provided, the device comprising:

[0025] Display module, used to display the resource interface;

[0026] The display module is further used to respond to the information input operation in the resource interface, and display the input value change information in the resource interface, wherein the value change information is used to indicate the change trend of the resource value of the virtual resource to be searched over time;

[0027] The display module is further used to respond to a resource search operation for the numerical change information, and display a target virtual resource in the resource interface, wherein the similarity between the change trend of the target virtual resource and the change trend represented by the numerical change information is greater than a threshold.

[0028] In a possible implementation, the information input operation is a drawing operation, and the numerical change information is a curve drawn by the drawing operation; the display module is used to respond to the drawing operation in the resource interface and display the curve drawn by the drawing operation in the resource interface.

[0029] In another possible implementation, the numerical change information is a curve included in the first image; the display module is used to display multiple candidate images in the resource interface in response to an image upload operation in the resource interface; and display the first image in the resource interface in response to a confirmation operation on the first image among the multiple candidate images.

[0030] In another possible implementation, the display module is further used to respond to a condition input operation in the resource interface, and display a search condition in the resource interface, wherein the search condition indicates the risk type of the virtual resource to be searched or the time range corresponding to the value change information;

[0031] The display module is used to display the target virtual resource in the resource interface in response to a resource search operation for the numerical value change information and the search conditions; wherein the target virtual resource belongs to the risk type, or the similarity between the change trend of the resource value of the target virtual resource within the time range and the change trend represented by the numerical value change information is greater than the threshold.

[0032] In another possible implementation, the display module is used to respond to a resource search operation for the numerical change information, and display the numerical change information corresponding to the target virtual resource in the resource interface. The numerical change information corresponding to the target virtual resource is used to represent the changing trend of the resource value of the target virtual resource over time.

[0033] In another possible implementation, the numerical change information is a first curve; the display module is used to respond to a resource search operation for the numerical change information, perform target detection on the second image and the change trend images of multiple candidate virtual resources, respectively, to obtain a first local image and multiple second local images, the first local image is a local image of the second image that includes the first curve, the second local image is a local image of the change trend image that includes the second curve, and the second curve is used to represent the change trend of the resource value of the candidate virtual resource over time; determine the matching degree between the first local image and each second local image, and determine the candidate virtual resource corresponding to the second local image with the largest matching degree as the target virtual resource; and display the virtual resource in the resource interface.

[0034] In another possible implementation, the display module is used to, in response to a resource search operation for the value change information, perform target detection on the second image and the multiple change trend images respectively through a resource search model to obtain the first local image and the multiple second local images;

[0035] The display module is used to determine the matching degree between the first partial image and each second partial image through the resource search model, and determine the candidate virtual resource corresponding to the second partial image with the greatest matching degree as the target virtual resource.

[0036] In another aspect, a model training device is provided, the device comprising:

[0037] a determination module, configured to determine a plurality of detection frame sizes based on bounding frame sizes of curves in a plurality of first sample images, wherein the curves are used to represent a trend of a change in a resource value of a virtual resource over time, and the bounding frame sizes are sizes of bounding frames of the curves in the first sample images;

[0038] an acquisition module, used to acquire a reference image and a second sample image, wherein a first curve in the reference image is used to represent a change trend of a resource value of a virtual resource to be searched over time, and a similarity between the first curve in the reference image and the second curve in the second sample image is greater than a threshold;

[0039] a detection module, configured to perform target detection on the reference image and the second sample image respectively based on the multiple detection frame sizes through a resource search model to obtain a first local image and a second local image, wherein the first local image is a local image of the reference image including the first curve, and the second local image is a local image of the second sample image including the second curve;

[0040] The determination module is further configured to determine a matching degree between the first partial image and the second partial image through the resource search model;

[0041] A training module is used to train the resource search model based on the matching degree.

[0042] In a possible implementation, the determination module is used to cluster multiple bounding box sizes to obtain multiple cluster clusters, each cluster cluster includes at least one bounding box size; generate multiple candidate size sets based on the central sizes of the multiple cluster clusters, each candidate size set includes at least one candidate size; for each candidate size set, based on the fitness of the candidate sizes in the candidate size set, update the candidate size set to obtain a new candidate size set, the fitness is used to indicate the degree of matching between the candidate size and the multiple bounding box sizes; determine the candidate size with the largest fitness in each new candidate size set as the detection box size.

[0043] In another possible implementation, the determination module is configured to perform multiple random perturbations on the center size of each cluster to obtain multiple candidate sizes; and combine the multiple candidate sizes and the center size into the candidate size set.

[0044] In another possible implementation, the determination module is used to filter out a candidate size subset from the candidate size set based on the fitness of the candidate sizes in the candidate size set, wherein the candidate size subset includes at least one candidate size; generate a new candidate size based on the candidate sizes in the candidate size subset; and form the new candidate size set by combining the candidate sizes in the candidate size subset and the new candidate size.

[0045] In another possible implementation, the determination module is used to determine a first sum value, where the first sum value is the fitness sum value of the candidate sizes in the candidate size set; generate a random number from a numerical range consisting of 0 and the first sum value; start from any candidate size in the candidate size set, traverse the candidate sizes in the candidate size set; when the second sum value when traversing to the i-th candidate size is less than the random number and the third sum value when traversing to the i+1-th candidate size is greater than the random number, determine the i+1-th candidate size as the candidate size in the candidate size subset, the second sum value is the fitness sum value of the i candidate sizes that have been traversed, and the third sum value is the fitness sum value of the i+1 candidate sizes that have been traversed, and i is an integer greater than 0.

[0046] In another possible implementation, the determination module is configured to randomly select two candidate sizes from the candidate size subset, swap part of the data at the same position in the two candidate sizes, and obtain two new candidate sizes; or, for any candidate size in the candidate size subset, reversely process part of the data in the candidate size to obtain a new candidate size.

[0047] In another possible implementation, the determination module is further configured to generate an exchange probability when the two candidate sizes are selected, wherein the exchange probability indicates a possibility of exchanging partial data at the same position in the two candidate sizes;

[0048] The determination module is used to exchange part of the data at the same position in the two candidate sizes to obtain two new candidate sizes when the exchange probability is greater than a first probability threshold.

[0049] In another possible implementation, the determination module is further configured to determine a variation probability corresponding to the candidate size, wherein the variation probability indicates a possibility of performing reverse processing on a portion of the data in the candidate size;

[0050] The determination module is used to perform reverse processing on part of the data in the candidate size to obtain a new candidate size when the mutation probability is greater than a second probability threshold.

[0051] In another possible implementation, the determination module is further used to determine the second probability threshold based on a third probability threshold, a current number of iterations, and a total number of iterations during multiple iterative updates of the multiple candidate size sets, wherein the third probability threshold is the probability threshold when the multiple candidate size sets are updated for the first time.

[0052] On the other hand, a computer device is provided, which includes a processor and a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the virtual resource search method or model training method described in the above aspects.

[0053] On the other hand, a computer-readable storage medium is provided, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to implement the operations performed by the virtual resource search method or the model training method as described in the above aspects.

[0054] On the other hand, a computer program product is provided, including a computer program, which, when executed by a processor, implements the operations performed by the virtual resource search method or the model training method described in the above aspects.

[0055] In the solution provided in the embodiment of the present application, in the resource interface, by inputting numerical change information representing the changing trend of the resource numerical value of the virtual resource to be searched over time, virtual resources whose changing trend of the resource numerical value matches the numerical change information can be automatically searched out, thereby realizing a new virtual resource search method, which can quickly search out the virtual resources that the user wants to view, and can realize the search of virtual resources even when the name of the virtual resource is unknown, thereby ensuring the convenience and accuracy of the virtual resource search, and also ensuring the efficiency of the resource search. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0057] Figure 1 It is a structural diagram of an implementation environment provided by an embodiment of the present application;

[0058] Figure 2 is a flow chart of a virtual resource search method provided by an embodiment of the present application;

[0059] Figure 3 is a flow chart of another virtual resource search method provided by an embodiment of the present application;

[0060] Figure 4 is a flow chart of another virtual resource search method provided by an embodiment of the present application;

[0061] Figure 5is a flow chart of another virtual resource search method provided by an embodiment of the present application;

[0062] Figure 6 is a schematic diagram of a resource interface provided in an embodiment of the present application;

[0063] Figure 7 is a flow chart of a model training method provided in an embodiment of the present application;

[0064] Figure 8 is a schematic diagram of mapping detection frame parameters back to the original image provided by an embodiment of the present application;

[0065] Fig. 9 It is a structural schematic diagram of a feature extraction sub-model provided in an embodiment of the present application;

[0066] Fig.10 It is a structural schematic diagram of a convolutional layer provided in an embodiment of the present application;

[0067] Fig.11 It is a structural schematic diagram of a feature extraction sub-model provided in an embodiment of the present application;

[0068] Fig.12 It is a flow chart of training a resource search model provided by an embodiment of the present application;

[0069] Fig.13 is a flow chart of a model training method provided in an embodiment of the present application;

[0070] Fig.14 is a flow chart for determining the size of a detection frame provided in an embodiment of the present application;

[0071] Fig.15 It is a structural diagram of a virtual resource search device provided in an embodiment of the present application;

[0072] Fig.16 It is a structural schematic diagram of a model training device provided in an embodiment of the present application;

[0073] Fig.17 is a schematic diagram of the structure of a terminal provided in an embodiment of the present application;

[0074] Fig.18 It is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0075] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0076] The terms "first", "second", "third", etc. used in this application may be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, a first sum value may be referred to as a second sum value, and similarly, a second sum value may be referred to as a first sum value without departing from the scope of this application.

[0077] The terms "at least one", "a plurality of", "each", and "any" used in this application include one, two or more than two, a plurality of includes two or more than two, each refers to each of the corresponding plurality, and any refers to any one of the plurality. For example, the plurality of candidate images include three candidate images, and each refers to each candidate image of the three candidate images, and any refers to any one of the three candidate images, which may be the first candidate image, or the second candidate image, or the third candidate image.

[0078] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions. For example, the data change information and virtual resources involved in this application are obtained with full authorization.

[0079] The virtual resource search method provided in the embodiment of the present application is executed by a computer device. Optionally, the computer device is a terminal or a server. Optionally, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal is a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, an intelligent voice interaction device, a smart home appliance, and a car terminal, etc., but is not limited to this.

[0080] In some embodiments, the computer device is provided as a terminal. Figure 1 is a schematic diagram of an implementation environment provided by an embodiment of the present application. Figure 1 , the implementation environment includes a terminal 101 and a server 102. The terminal 101 and the server 102 are connected via a wireless or wired network.

[0081] Terminal 101 is used to display the input numerical change information. Through the connection with server 102, a search request is sent to server 102. The search request carries the numerical change information. Server 102 is used to respond to the search request, search for the target virtual resources matching the numerical change information, and send the target virtual resources to terminal 101. Terminal 101 is used to receive the target virtual resources sent by server 102 and display the target virtual resources.

[0082] In some embodiments, the terminal 101 is installed with an application provided by the server 102, and the terminal 101 can realize functions such as viewing virtual resources and searching virtual resources through the application. Optionally, the application is an application in the operating system of the terminal 101, or an application provided by a third party. For example, the application is a virtual resource application, and the virtual resource application has functions such as viewing virtual resources and searching virtual resources. Of course, the virtual resource application can also have other functions, such as shopping functions, navigation functions, game functions, etc.

[0083] Terminal 101 is used to log in to the application based on an account, display the resource interface through the application, and can display the input value change information in the resource interface. The application sends a search request to server 102, and server 102 is used to respond to the search request, search for the target virtual resources that match the value change information, and send the target virtual resources to terminal 101. Terminal 101 is used to receive the target virtual resources sent by server 102 through the application and display the target virtual resources.

[0084] In some embodiments, a resource search model is deployed in the server 102, and the server 102 provides services to the application based on the resource search model, so as to implement the search of virtual resources based on the resource search model.

[0085] In some embodiments, the resource search model is trained by the server 102, or by other servers. In the case where the resource search model is trained by the server 102, the server 102 trains the resource search model in advance, and then saves the trained resource search model, so as to implement the search for virtual resources based on the trained resource search model. In the case where the resource search model is trained by other servers, after the other servers train the resource search model, they can send the trained resource search model to the server 102, and then the server 102 receives and saves the trained resource search model, so as to implement the search for virtual resources based on the trained resource search model.

[0086] Figure 2 is a flowchart of a virtual resource search method provided by an embodiment of the present application, taking the method executed by a computer device as an example. Figure 2 As shown, the method includes:

[0087] 201. The computer device displays a resource interface.

[0088] In an embodiment of the present application, the resource interface is used to display any virtual resource and can also initiate a virtual resource search, that is, a resource search is initiated in the resource display interface so that the search results are displayed in the resource interface, and the search results indicate the searched virtual resources or indicate that no virtual resources are searched.

[0089] Among them, a virtual resource is a virtual resource whose resource value changes over time. The virtual resource is any type of virtual resource. For example, the virtual resource is a virtual resource in the financial field, such as a stock or a fund. For example, the resource value of the virtual resource is the price of the stock or the price of the fund. For another example, the virtual resource is a virtual item, and the resource value of the virtual resource is the price of the virtual item.

[0090] 202. The computer device responds to the information input operation in the resource interface and displays the input value change information in the resource interface, where the value change information is used to indicate the change trend of the resource value of the virtual resource to be searched over time.

[0091] In an embodiment of the present application, a virtual resource search can be initiated in the resource interface, and arbitrary numerical change information can be entered in the resource interface. The numerical change information indicates the changing trend of the resource value of the virtual resource that the user wants to search for over time, thereby facilitating the user's search for virtual resources.

[0092] The information input operation is used to input a value change information for indicating the change trend of the virtual resource to be searched in the resource interface. The value change information can be expressed in any form, for example, the value change information is expressed in the form of a curve or other forms. The resource value of the virtual resource is used to indicate the total value or unit value of the virtual resource. For example, if the virtual resource is a fund, the resource value of the virtual resource is the price of the fund.

[0093] In the embodiment of the present application, the information input operation can be any type of operation, and the value change information can be input in any manner. For example, in response to the information input operation, a curve can be drawn in the resource interface, any image including the curve can be uploaded, or any image can be selected from multiple images provided by the resource interface, wherein the image provided by the resource interface includes the curve, and the drawn curve, the curve included in the uploaded image, and the curve included in the selected image are equivalent to the value change information.

[0094] 203. In response to the resource search operation for the value change information, the computer device displays the target virtual resource in the resource interface, and the similarity between the change trend of the target virtual resource and the change trend represented by the value change information is greater than a threshold.

[0095] In an embodiment of the present application, when numerical change information is input in a resource interface, a resource search operation can be triggered based on the numerical change information, so that virtual resources can be searched based on the numerical change information, so as to search for target virtual resources that are sufficiently similar to the numerical change information, and display the searched target virtual resources in the resource interface, thereby realizing a convenient resource search method.

[0096] The threshold is an arbitrary value, for example, the threshold is 0.8 or 0.9, etc. If the similarity between the change trend of the resource value of the target virtual resource over time and the value change information is greater than the threshold, it means that the target virtual resource is sufficiently similar to the value change information, that is, the target virtual resource is the virtual resource that the user wants to search for.

[0097] In the solution provided in the embodiment of the present application, in the resource interface, by inputting numerical change information representing the changing trend of the resource numerical value of the virtual resource to be searched over time, virtual resources whose changing trend of the resource numerical value matches the numerical change information can be automatically searched out, thereby realizing a new virtual resource search method, which can quickly search out the virtual resources that the user wants to view, and can realize the search of virtual resources even when the name of the virtual resource is unknown, thereby ensuring the convenience and accuracy of the virtual resource search, and also ensuring the efficiency of the resource search.

[0098] exist Figure 2 On the basis of the illustrated embodiment, the embodiment of the present application can also adopt a drawing method to draw a curve in the resource interface so as to search for virtual resources based on the drawn curve. The specific process is detailed in the following embodiment.

[0099] Figure 3 is a flowchart of another virtual resource search method provided by an embodiment of the present application, taking the method executed by a computer device as an example. Figure 3 As shown, the method includes:

[0100] 301. The computer device displays a resource interface.

[0101] In a possible implementation, taking the computer device as a terminal, an application is installed on the terminal; then step 301 includes: the terminal displays a resource interface through the application.

[0102] In the embodiment of the present application, the resource interface is an interface provided by the application, and the user can display the resource interface through the terminal to initiate a search for virtual resources in the resource interface and display the searched virtual resources in the resource interface.

[0103] The application is any application, for example, the application is a fund application or a stock application.

[0104] In a possible implementation, the resource interface includes an information input area and a search result display area, wherein the information input area is used to input value change information, and the search result display area is used to display search results.

[0105] 302. The computer device responds to the drawing operation in the resource interface and displays a curve drawn by the drawing operation in the resource interface, where the curve drawn by the drawing operation is used to represent a trend of a resource value of the virtual resource to be searched changing over time.

[0106] In an embodiment of the present application, a drawing operation can be triggered in the resource interface, and numerical change information can be input in the resource interface by drawing. An arbitrary curve can be drawn in the resource interface so that a virtual resource search can be performed subsequently according to the drawn curve, thereby achieving the convenience of information input and ensuring the convenience of resource search.

[0107] In the embodiment of the present application, the drawing operation in the resource interface is equivalent to the information input operation in the resource interface, and the curve drawn by the drawing operation is the numerical value change information input in the resource interface.

[0108] In one possible implementation, the drawing operation refers to pressing a certain position in the resource interface and sliding in the resource interface while keeping the pressure on. When the resource interface is no longer pressed, the drawing operation ends, and the sliding track is the curve drawn by the drawing operation.

[0109] For example, the user presses a certain position in the resource interface through the terminal and keeps pressing the resource interface to slide in the resource interface. The sliding trajectory is the curve drawn by the drawing operation. When the sliding ends, it is equivalent to the end of the drawing operation, and the drawn curve is displayed in the resource interface.

[0110] In one possible implementation, step 302 includes: responding to a drawing operation in a resource interface, determining multiple points drawn by the drawing operation, connecting the multiple points drawn, and smoothing the connected lines to obtain a curve drawn by the drawing operation, and displaying the curve in the resource interface.

[0111] In an embodiment of the present application, a user can trigger a drawing operation in a resource interface through a terminal so as to draw multiple points in the resource interface, and then connect the multiple points into a line according to the positions of the multiple points, and smooth the connected line to ensure that the curve drawn by the drawing operation is smooth, improve the accuracy of the curve, and ensure the display effect of the curve, so that virtual resources can be searched based on the curve later. In addition, the method of drawing curves is enriched, and the curve can be automatically generated only by drawing multiple points, thereby ensuring the efficiency of drawing curves and the convenience of inputting numerical change information.

[0112] In one possible implementation, step 302 includes: responding to a drawing operation in a resource interface, determining multiple line segments drawn by the drawing operation, connecting the multiple line segments, and smoothing the connected lines to obtain a curve drawn by the drawing operation, and displaying the curve in the resource interface.

[0113] In an embodiment of the present application, a user can trigger a drawing operation in a resource interface through a terminal so as to draw multiple line segments in the resource interface, and then connect the multiple line segments into a line according to the positions of the multiple line segments, and smooth the connected line to ensure that the curve drawn by the drawing operation is smooth, improve the accuracy of the curve, and ensure the display effect of the curve, so that virtual resources can be searched based on the curve later. In addition, the method of drawing curves is enriched, and the curve can be automatically generated only by drawing multiple line segments, thereby ensuring the efficiency of drawing curves and the convenience of inputting numerical change information.

[0114] In a possible implementation, step 302 includes: in response to a drawing operation in the resource interface, displaying an image corresponding to the drawing operation in the resource interface, where the image corresponding to the drawing operation includes a curve drawn by the drawing operation.

[0115] In an embodiment of the present application, when a drawing operation is triggered in a resource interface, the curve drawn by the drawing operation will be displayed in real time in the resource interface, and an image will be generated based on the drawn curve, which includes the drawn curve, so that the image including the drawn curve can be used for subsequent resource search.

[0116] Optionally, the resource interface includes a drawing area, and the drawing area is used to trigger a drawing operation to draw a curve. Then step 302 includes: in response to the drawing operation in the drawing area, based on the curve drawn by the drawing operation, generating an image in real time, and displaying the image in the drawing area. The image includes the content included in the drawing area.

[0117] In an embodiment of the present application, during the process of triggering a drawing operation in a drawing area, an image including the curve drawn by the drawing operation is generated in real time as the drawing operation changes, so as to update the image displayed in the drawing area in real time to ensure that the displayed image matches the curve drawn by the drawing operation and to ensure the accuracy of the displayed image.

[0118] For example, a curve is drawn in the drawing area through a drawing operation, and the image displayed in the drawing area is blank except for the drawn curve.

[0119] It should be noted that the embodiment of the present application is explained using a curve drawn with numerical value change information as an example. In another embodiment, there is no need to execute the above step 302, but other methods are adopted to respond to the information input operation in the resource interface. The input numerical value change information is displayed in the resource interface. The numerical value change information is used to indicate the changing trend of the resource value of the virtual resource to be searched over time.

[0120] 303. In response to the resource search operation for the value change information, the computer device displays the target virtual resource in the resource interface, and the similarity between the change trend of the target virtual resource and the change trend represented by the value change information is greater than a threshold.

[0121] In a possible implementation, step 303 includes: in response to the drawing operation being completed, displaying the target virtual resource in the resource interface.

[0122] In an embodiment of the present application, the end of the drawing operation is equivalent to detecting a resource search operation for the currently drawn curve, and the target virtual resource is displayed in the resource interface. In this way, the user does not need to trigger other operations, and the convenience and real-time nature of the resource search can be guaranteed.

[0123] In a possible implementation, a search option is displayed in the resource interface, and step 303 includes: in response to a triggering operation on the search option, displaying the target virtual resource in the resource interface.

[0124] In an embodiment of the present application, the triggering operation of the search option is equivalent to detecting a resource search operation for the currently drawn curve, and the target virtual resources are displayed in the resource interface. In this way, the user can independently choose whether to search for virtual resources according to the drawn curve, which can meet the needs of the user and ensure the convenience and accuracy of searching resources.

[0125] In one possible implementation, step 303 includes: in response to a resource search operation for numerical change information, displaying the numerical change information corresponding to the target virtual resource in the resource interface, wherein the numerical change information corresponding to the target virtual resource is used to represent the changing trend of the resource value of the target virtual resource over time.

[0126] In an embodiment of the present application, in response to a resource search operation for numerical change information, when a target virtual resource is searched for whose change trend is sufficiently similar to the change trend indicated by the numerical change information, the numerical change information corresponding to the target virtual resource is displayed, so that the user can view the displayed numerical change information through the resource interface to obtain the changing trend of the resource value of the target virtual resource over time, thereby ensuring the display effect of the search results.

[0127] Optionally, the value change information corresponding to the target virtual resource includes a change trend curve corresponding to the target virtual resource.

[0128] In the embodiment of the present application, each virtual resource corresponds to a change trend curve, so as to represent the change trend of the resource value of the corresponding virtual resource over time in the form of a curve. Since the value change information input in the resource interface is a drawn curve, the drawn curve is matched with the change trend curve corresponding to each virtual resource, so as to find the target virtual resource whose change trend curve is sufficiently similar to the drawn curve, so as to ensure the accuracy of the searched target virtual resource.

[0129] Optionally, when the resource interface displays a change trend curve corresponding to the target virtual resource, the resource interface also displays a resource identifier of the target virtual resource and a value change curve corresponding to the target type.

[0130] The resource identifier of the target virtual resource can be represented in any form, for example, the resource identifier of the target virtual resource is the name or code of the target virtual resource. The target type is the resource type to which the target virtual resource belongs. For example, if the target virtual resource is an industrial type virtual resource, the target type is an industrial type. The value change curve corresponding to the target type is used to represent the average change trend of the virtual resources belonging to the target type, that is, the average change trend of the resource values ​​of all virtual resources of the target type over time.

[0131] In an embodiment of the present application, when the target virtual resource is searched, the resource interface will not only display the change trend curve corresponding to the target virtual resource, but also display the resource identifier of the target virtual resource and the numerical change curve corresponding to the target type, so as to enrich the displayed information and ensure the display effect of the search results.

[0132] In one possible implementation, taking the numerical change information input in the resource interface as the first curve as an example; the step 303 includes: in response to the resource search operation for the numerical change information, performing target detection on the second image and the change trend images of multiple candidate virtual resources respectively, to obtain a first local image and multiple second local images, the first local image is a local image of the second image including the first curve, the second local image is a local image of the change trend image including the second curve, and the second curve is used to represent the change trend of the resource value of the candidate virtual resource over time; determining the matching degree between the first local image and each second local image, and determining the candidate virtual resource corresponding to the second local image with the largest matching degree as the target virtual resource; and displaying the target virtual resource in the resource interface.

[0133] In an embodiment of the present application, a curve is drawn in the resource interface through a drawing operation triggered in the resource interface, and an image is generated based on the first curve drawn by the drawing operation, that is, a second image is generated. The change trend image corresponding to each virtual resource includes a curve that can represent the change trend of the resource value of the virtual resource over time. Since the image may contain other content in addition to the curve, target detection is performed on the second image including the first curve and the change trend images of multiple candidate virtual resources to detect a local image containing the curve, and then the obtained local images are matched to ensure the accuracy of the determined target virtual resources, to avoid interference from content other than the curve in the change trend image, and to ensure the accuracy of resource search.

[0134] The second image is an image including the first curve, and the first curve is equivalent to the curve drawn by the drawing operation. Optionally, the resource interface includes a drawing area, and when the first curve is drawn in the drawing area, the second image is generated based on the content of the drawing area.

[0135] Optionally, the process of performing target detection on the second image includes: performing feature extraction on the second image to obtain image features, detecting the image features based on multiple detection frame sizes to obtain target parameters, the target parameters indicating the position of the first curve in the second image, and extracting a first local image from the second image based on the target parameters.

[0136] In an embodiment of the present application, the image features of the second image are utilized in combination with multiple detection frame sizes to detect target parameters for representing the position of the first curve from the image features, and then a local image including the first curve is extracted from the second image according to the target parameters to ensure the accuracy of the determined local image.

[0137] Optionally, the target parameters include center coordinates and size, the center coordinates are center coordinates of a bounding box of the first curve in the second image, and the size is the size of the bounding box.

[0138] Optionally, the process of detecting image features based on the detection frame size includes: performing feature extraction on the second image to obtain multi-scale image features, dividing the image features of each scale into multiple grid feature units for the image features of the scale, traversing each grid feature unit, and based on the detection frame size corresponding to the scale, determining the detection frame indicated by the detection frame size from the traversed grid feature units, detecting features contained in the detection frame to obtain detection frame parameters and probabilities corresponding to the detection frame, and obtaining multiple detection frame parameters and corresponding probabilities after traversing multiple grid feature units included in the image features of the scale, and determining the target parameters based on the multiple detection frame parameters and corresponding probabilities.

[0139] The detection frame parameters corresponding to the detection frame indicate the center position and size of the detection frame, for example, the detection frame parameters include the center coordinates of the detection frame, the height and width of the detection frame. The probability corresponding to the detection frame indicates the possibility that the detection frame includes the first curve.

[0140] In an embodiment of the present application, for image features at each scale, a traversal method is adopted to traverse each grid feature unit in the image features so as to determine the probability that the detection frame in each grid feature unit includes the first curve, and then, based on multiple detection frame parameters and corresponding probabilities, the detection frame parameters with probabilities greater than the first threshold are fused to obtain target parameters to determine the position of the first curve in the second image.

[0141] In the embodiment of the present application, each scale corresponds to one or more detection frame sizes. For example, feature extraction is performed on the second image to obtain image features of 3 scales, and each scale corresponds to 3 detection frame sizes, that is, there are 9 detection frame sizes in total.

[0142] Optionally, the process of determining a detection frame from the traversed grid feature unit includes: for the traversed grid feature unit, taking the center point of the grid feature unit as the center, and determining a detection frame in the grid feature unit based on a detection frame size corresponding to the scale.

[0143] For example, the size of an image feature at a certain scale is 416×416, and the size of the divided grid feature unit is 13×13. The sizes of the three detection frames corresponding to this scale are 5×5, 3×9, and 9×3, respectively. Then, with the center point of the grid feature unit as the center and these three detection frame sizes, three detection frames can be determined from the grid feature unit.

[0144] Optionally, the process of determining the target parameters includes: detecting features contained in the detection box, obtaining detection box parameters, probabilities and confidences corresponding to the detection box, obtaining multiple detection box parameters and corresponding probabilities after traversing the grid feature units included in the image features of the scale, and determining the target parameters based on the multiple detection box parameters, corresponding probabilities and confidences.

[0145] The confidence is used to indicate the possibility that the detection box includes the target, and the probability indicates the possibility that the target included in the detection box is a curve. The features included in the detection box are local features in the grid feature unit. Optionally, the feature included in the detection box is detected by decoding the feature included in the detection box.

[0146] In an embodiment of the present application, for image features of each scale, a traversal method is adopted to traverse each grid feature unit in the image features so as to determine the probability that the detection frame in each grid feature unit includes the first curve, and then, based on multiple detection frame parameters, corresponding probabilities and confidences, detection frame parameters with confidences less than a confidence threshold are screened out from the multiple detection frame parameters, and based on the remaining detection frame parameters and the corresponding probabilities, the detection frame parameters with probabilities greater than the first threshold are fused to obtain target parameters to determine the position of the first curve in the second image.

[0147] It should be noted that the above is only an example of target detection on the second image. The process of target detection on the change trend image of the candidate virtual resource is the same as the process of target detection on the second image, which will not be repeated here.

[0148] Optionally, the process of determining the degree of matching between the first partial image and each second partial image includes: determining the degree of matching between the first partial image and the second partial image as the degree of similarity between an image feature of the first partial image and an image feature of the second partial image.

[0149] In the embodiment of the present application, any method can be adopted to determine the similarity between the image features of the first partial image and the image features of the second partial image. For example, the Euclidean distance method can be adopted to determine the similarity between the image features of the first partial image and the image features of the second partial image.

[0150] Optionally, the above process of determining the target virtual resource is executed by a server. Taking a computer device as a terminal as an example, the terminal sends a resource search request to the server in response to a resource search operation for numerical change information. The resource search request carries numerical change information. The server performs target detection on the second image and the change trend image of multiple candidate virtual resources in the above manner, respectively, to obtain a first local image and multiple second local images. The first local image is a local image of the second image that includes a first curve, and the second local image is a local image of the change trend image that includes a second curve. The second curve is used to represent the change trend of the resource value of the candidate virtual resource over time; determine the matching degree between the first local image and each second local image, determine the candidate virtual resource corresponding to the second local image with the largest matching degree as the target virtual resource, send the target virtual resource to the terminal, the terminal receives the target virtual resource, and displays the target virtual resource in the resource interface.

[0151] Optionally, before performing target detection on the second image, the data points in the drawn curve are normalized, that is, the method also includes: determining the maximum and minimum values ​​of each data point in the drawn first curve, determining a first difference, the first difference being the difference between the maximum and minimum values; for each data point in the first curve, determining a second difference, the second difference being the difference between the data point and the minimum value, and determining the ratio of the second difference to the first difference as the updated data point of the data point.

[0152] In an embodiment of the present application, the resource interface includes a drawing area. After drawing the first curve in the drawing area, the coordinates of each point on the first curve can be determined, and the coordinates of each point are used as the data point, so that each data point on the first curve can be normalized in the above manner to ensure the accuracy of the first curve after normalization.

[0153] Optionally, the process of normalizing each data point in the first curve satisfies the following relationship:

[0154]

[0155] Among them, X max It is used to represent the maximum value among the data points in the first curve, X min It is used to represent the minimum value of each data point in the first curve. X is used to represent any data point in the first curve. norm It is used to represent the data point after normalization of data point X, where data point X norm The value is between 0 and 1.

[0156] Optionally, the drawn curve is smoothed before performing target detection on the second image, that is, the method further includes: smoothing the drawn first curve to obtain a processed first curve.

[0157] In the embodiment of the present application, bilateral filtering can be used to smooth the curve to reduce random noise in the first curve, making the first curve smoother and ensuring the accuracy of the first curve.

[0158] Optionally, the following relationship can be adopted to smooth the first curve:

[0159]

[0160] Among them, N is used to represent the total number of weights, ω i is used to represent the i-th weight; x is used to represent the position of any pixel in the second image, i Used to indicate the position of neighboring pixels of pixel x, |xx i |Used to represent pixel x and pixel xi The distance between s It is used to represent the standard deviation in the spatial domain, to control the weight of spatial proximity, and to determine the width of the Gaussian kernel function in the spatial domain. G(·) is used to represent the Gaussian kernel function, which is used to calculate the weighting of the spatial domain and the intensity domain. I(x) is used to represent the pixel value of the pixel point x. I(x i ) is used to represent pixel x i The pixel value of |I(x)-I(x i ) is used to represent the absolute value of the difference in pixel values ​​and is used for Gaussian weight calculation in the intensity domain; σ r It is used to represent the standard deviation in the intensity domain, to control the similarity of pixel intensities, and to determine the width of the Gaussian kernel function in the intensity domain; B(x, k) is used to represent the updated pixel value of the pixel point x in the second image.

[0161] In an embodiment of the present application, the second image is processed using bilateral filtering to smooth the first curve drawn in the second image, reduce random noise in the first image, and retain edge information of the first curve, thereby improving the recognizability of the first curve.

[0162] In the solution provided in the embodiment of the present application, in the resource interface, by inputting numerical change information representing the changing trend of the resource numerical value of the virtual resource to be searched over time, virtual resources whose changing trend of the resource numerical value matches the numerical change information can be automatically searched out, thereby realizing a new virtual resource search method, which can quickly search out the virtual resources that the user wants to view, and can realize the search of virtual resources even when the name of the virtual resource is unknown, thereby ensuring the convenience and accuracy of the virtual resource search, and also ensuring the efficiency of the resource search.

[0163] exist Figure 2 On the basis of the illustrated embodiment, the embodiment of the present application can also adopt the image uploading method to input an image including a curve in the resource interface so as to search for virtual resources based on the input image. The specific process is detailed in the following embodiment.

[0164] Figure 4 is a flowchart of another virtual resource search method provided by an embodiment of the present application, taking the method executed by a computer device as an example. Figure 4 As shown, the method includes:

[0165] 401. The computer device displays a resource interface.

[0166] The step 401 is the same as the above step 301 and will not be described again.

[0167] 402. The computer device displays a plurality of candidate images in the resource interface in response to an image upload operation in the resource interface.

[0168] In the embodiment of the present application, the resource interface has the function of searching for virtual resources, and at the same time, has the function of uploading images, and can upload any image in the resource interface so as to subsequently search for virtual resources based on the uploaded image.

[0169] The multiple candidate images are arbitrary images, for example, the multiple candidate images are images including different curves. For example, the multiple candidate images are locally stored images, such as the multiple candidate images are images in the terminal's photo album, and there may be images that do not include curves among the multiple candidate images.

[0170] In an embodiment of the present application, the multiple candidate images are images in the terminal's photo album, and may also be multiple images including curves provided by a resource interface. When multiple candidate images are provided by the resource interface, in response to an image upload operation in the resource interface, the multiple images including curves provided can be displayed on the resource interface for user selection.

[0171] In a possible implementation, an image upload option is displayed in the resource interface, and step 402 includes: in response to a click operation on the image upload option, displaying a plurality of candidate images in the resource interface.

[0172] The image upload option is any type of option, for example, the image upload option is a button or a slide bar.

[0173] 403. In response to a confirmation operation on a first image among multiple candidate images, the computer device displays the first image in a resource interface, and the value change information is a curve included in the first image, and the curve is used to represent a change trend of the resource value of the virtual resource to be searched over time.

[0174] In an embodiment of the present application, the resource interface has the function of searching for virtual resources, and at the same time, has the function of uploading images. By triggering the image upload operation in the resource interface, any image can be selected from multiple candidate images displayed and uploaded to the resource interface so that it can be displayed in the resource interface for users to view. This enriches the way of inputting numerical change information in the resource interface, ensures the convenience of inputting numerical change information, and thus ensures the convenience of searching for virtual resources. In addition, it enables users to know what kind of image will be used for subsequent virtual resource searches, so as to enhance user experience.

[0175] The plurality of candidate images include a first image, and the first image includes a curve. The images other than the first image in the plurality of candidate images may include a curve or may not include a curve.

[0176] In a possible implementation, the confirmation operation on the first image can be any type of operation, for example, the confirmation operation on the first image is a click operation or a drag operation, etc. Alternatively, when the resource interface displays multiple candidate images, a confirmation option is displayed in the resource interface, and the confirmation operation on the first image is a triggering operation on the confirmation option when the first image is selected.

[0177] In a possible implementation, the first image is an image captured by a camera.

[0178] In an embodiment of the present application, any curve can be photographed to obtain a first image, and then the photographed image can be uploaded to the resource interface and displayed, so that virtual resources can be searched for later based on the photographed image. This realizes a new virtual resource search method and improves the convenience of virtual resource search.

[0179] For example, a user shoots an arbitrary curve through a terminal to obtain a first image, and the terminal stores the first image. When the terminal displays a resource interface, the terminal uploads the image to upload the first highlighted curve to the resource interface for display.

[0180] It should be noted that the embodiment of the present application is described by taking the example of inputting an image in the resource interface by uploading an image. In another embodiment, there is no need to execute the above steps 402-403. Instead, other methods are adopted to respond to the information input operation in the resource interface. The input value change information is displayed in the resource interface. The value change information is used to indicate the changing trend of the resource value of the virtual resource to be searched over time.

[0181] 404. The computer device displays the target virtual resource in the resource interface in response to the resource search operation for the first image, and the similarity between the change trend of the target virtual resource and the change trend represented by the numerical change information is greater than a threshold.

[0182] In the embodiment of the present application, the resource search operation for the first image is a resource search operation for the curve included in the first image, which is also equivalent to a resource search operation for numerical value change information.

[0183] In an embodiment of the present application, when a first image is input in a resource interface, a resource search operation can be triggered for the first image so that virtual resources can be searched based on a curve included in the first image, so as to search for target virtual resources whose change trend has a similarity greater than a threshold with the change trend represented by the curve included in the first image, and display the searched target virtual resources in the resource interface, thereby realizing a convenient resource search method.

[0184] In a possible implementation, step 404 includes: when the duration for which the first image is displayed on the resource interface reaches a target duration, displaying the target virtual resource in the resource interface.

[0185] The target duration is any duration, for example, the target duration is 0.5 seconds or 1 second.

[0186] In an embodiment of the present application, when the duration of displaying the first image on the resource interface reaches the target duration, which is equivalent to detecting a resource search operation for the currently drawn curve, the target virtual resource is displayed in the resource interface. In this way, the user does not need to trigger other operations, thereby ensuring the convenience and real-time nature of resource search.

[0187] In a possible implementation, a search option is displayed in the resource interface, and step 404 includes: in response to a triggering operation on the search option, displaying the target virtual resource in the resource interface.

[0188] In an embodiment of the present application, the triggering operation of the search option is equivalent to detecting a resource search operation for the first image, and the target virtual resources are displayed in the resource interface. In this way, the user can independently choose whether to search for virtual resources according to the drawn curve, which can meet the needs of the user and ensure the convenience and accuracy of searching resources.

[0189] In one possible implementation, step 404 includes: in response to a resource search operation for the first image, performing target detection on the first image and the change trend images of multiple candidate virtual resources, respectively, to obtain a third local image and multiple second local images, the third local image being a local image including the first curve in the first image, the second local image being a local image including the second curve in the change trend image, and the second curve being used to represent the change trend of the resource value of the candidate virtual resource over time; determining the degree of match between the third local image and each second local image, and determining the candidate virtual resource corresponding to the second local image with the largest degree of match as the target virtual resource; and displaying the virtual resource in the resource interface.

[0190] In an embodiment of the present application, the first image uploaded in the resource interface includes a curve, which is equivalent to the changing trend of the resource value of the virtual resource to be searched over time. The changing trend image corresponding to each virtual resource includes a curve that can represent the changing trend of the resource value of the virtual resource over time. Since the image may contain other content in addition to the curve, target detection is performed on the uploaded first image and the changing trend images of multiple candidate virtual resources to detect the local image containing the curve, and then the obtained local images are matched to ensure the accuracy of the determined target virtual resource, to avoid interference from content other than the curve in the changing trend image, and to ensure the accuracy of the resource search.

[0191] It should be noted that the above process of performing target detection on the first image and the process of determining the matching degree between the third partial image and the second partial image are similar to the above step 303 and will not be described in detail here.

[0192] It should be noted that the embodiment of the present application is explained by taking the curve included in the first image displayed on the resource interface as the numerical change information as an example. In another embodiment, there is no need to execute the above step 404, but other methods are adopted. In response to the resource search operation for the numerical change information, the target virtual resources are displayed in the resource interface, and the similarity between the change trend of the target virtual resources and the change trend represented by the numerical change information is greater than a threshold.

[0193] In the solution provided in the embodiment of the present application, in the resource interface, by inputting numerical change information representing the changing trend of the resource numerical value of the virtual resource to be searched over time, virtual resources whose changing trend of the resource numerical value matches the numerical change information can be automatically searched out, thereby realizing a new virtual resource search method, which can quickly search out the virtual resources that the user wants to view, and can realize the search of virtual resources even when the name of the virtual resource is unknown, thereby ensuring the convenience and accuracy of the virtual resource search, and also ensuring the efficiency of the resource search.

[0194] exist Figure 2 On the basis of the illustrated embodiment, the embodiment of the present application can also input search conditions in the resource interface so as to search for virtual resources in combination with the search conditions. The specific process is detailed in the following embodiment.

[0195] Figure 5 is a flowchart of another virtual resource search method provided by an embodiment of the present application, taking the method executed by a computer device as an example. Figure 5 As shown, the method includes:

[0196] 501. The computer device displays a resource interface.

[0197] 502. The computer device responds to the information input operation in the resource interface and displays the input value change information in the resource interface. The value change information is used to indicate the change trend of the resource value of the virtual resource to be searched over time.

[0198] The steps 501-502 are similar to the above steps 201-202 and will not be described again.

[0199] 503. The computer device responds to the condition input operation in the resource interface and displays the search condition in the resource interface, where the search condition indicates the time range corresponding to the risk type or value change information of the virtual resource to be searched.

[0200] In an embodiment of the present application, search conditions can also be entered in the resource interface to specify the conditions that need to be met by the virtual resources to be searched, so that the virtual resource search can be subsequently performed in combination with the input numerical change information and the search conditions. This not only meets the various needs of users, but also can filter virtual resources through search conditions, and then determine whether the filtered virtual resources are similar to the change trends represented by the numerical change information. There is no need to determine whether the change trends of all virtual resources are similar to the change trends represented by the numerical change information, which can reduce the amount of search calculations and ensure search efficiency.

[0201] Among them, the risk type indicates the risk of the virtual resource, which can reflect the magnitude of the change in the resource value of the virtual resource. The larger the change in the resource value of the virtual resource, that is, the greater the fluctuation in the resource value of the virtual resource, the higher the risk of the virtual resource; the smaller the change in the resource value of the virtual resource, that is, the greater the fluctuation in the resource value of the virtual resource, the lower the risk of the virtual resource. For example, risk types include low risk, medium risk and high risk. The time range corresponding to the value change information refers to the time range of the change trend represented by the value change information. Combining the value change information and the corresponding time range can reflect the change trend of the resource value of the virtual resource to be searched over time within the time range.

[0202] In a possible implementation, the resource interface includes a condition input area, and a method of inputting search conditions in the resource interface includes: in response to a condition belonging to an operation in the condition input area, displaying the input search conditions in the condition input area.

[0203] The condition input area is used to input any search condition.

[0204] Optionally, the condition input area includes a condition input box, and the condition input box is used to input the search condition. For example, the condition input box is a text input box, and the user can input any risk type or time range in the text input box through the terminal.

[0205] Optionally, the condition input area includes a condition input option, and in response to a triggering operation on the condition input option, the input search condition is displayed on the resource interface.

[0206] In the embodiment of the present application, the condition input option is used to input the search condition. For example, the condition input option includes multiple risk type options. The user selects any risk type option in the resource interface through the terminal, and the risk type corresponding to the selected risk type option can be used as the search condition. For another example, the condition input option is represented in the form of a slider, and the user adjusts the point on the slider through the terminal to adjust the input risk type or time range.

[0207] It should be noted that the embodiment of the present application is described by taking the example of executing step 502 first and then executing step 503. In another embodiment, step 503 can also be executed first and then step 502. The present application does not limit this.

[0208] 504. The computer device responds to the resource search operation for the numerical change information and the search conditions, and displays the target virtual resource in the resource interface. The target virtual resource belongs to the risk type, or the similarity between the change trend of the resource value of the target virtual resource within the time range and the change trend represented by the numerical change information is greater than a threshold.

[0209] In an embodiment of the present application, the resource interface is capable of inputting not only numerical change information but also search conditions, so as to combine the input numerical change information and search conditions to perform a virtual resource search, so as to search for virtual resources that match the numerical change information and search conditions, thereby meeting the various needs of users and ensuring the convenience and success rate of virtual resource searches.

[0210] In an embodiment of the present application, when the search conditions include risk type, the changing trend of the resource value of the searched virtual resource over time is similar to the changing trend represented by the value change information, and the searched virtual resource belongs to the risk type; when the search conditions include time range, the changing trend of the resource value of the searched virtual resource within the time range is similar to the changing trend represented by the value change information; when the search conditions include risk type and time range, the changing trend of the resource value of the searched virtual resource within the time range is similar to the changing trend represented by the value change information, and the searched virtual resource belongs to the risk type.

[0211] In one possible implementation, step 504 includes: when the search conditions include risk types, in response to a resource search operation for numerical change information and the search conditions, based on the search conditions, screening out a first virtual resource belonging to the risk type from multiple candidate virtual resources, based on the numerical change information, screening out a target virtual resource from multiple first virtual resources, and displaying the target virtual resource in the resource interface.

[0212] In an embodiment of the present application, when the search conditions include risk types, multiple candidate virtual resources can be screened based on the search conditions first, and then combined with the numerical change information, target virtual resources similar to the numerical change information can be determined from the screened virtual resources. There is no need to determine the similarity between each candidate virtual resource and the numerical change information, which can reduce the amount of calculation and ensure the efficiency of determining the target virtual resources.

[0213] Each candidate virtual resource corresponds to a risk type, and based on the search condition and the risk type of each candidate virtual resource, the virtual resource that meets the search condition can be determined. Figure 3 or Figure 4 Similar to the embodiment shown, the target virtual resource can be screened out by using an image including value change information and an image of a change trend of a candidate virtual resource, which will not be described in detail here.

[0214] In one possible implementation, step 504 includes: when the search conditions include a time range corresponding to the numerical change information, in response to a resource search operation for the numerical change information and the search conditions, based on the search conditions, obtaining the numerical change information of multiple candidate virtual resources within the time range, based on the input numerical change information and the numerical change information of each candidate virtual resource within the time range, determining a target virtual resource from the multiple candidate virtual resources, and displaying the target virtual resource in the resource interface.

[0215] In an embodiment of the present application, when the search conditions include a time range corresponding to numerical change information, the numerical change information of each candidate virtual resource within the time range can be obtained according to the input time range, and then matched with the input time change information to ensure the accuracy of the determined target virtual resource.

[0216] The value change information of the candidate virtual resource is used to indicate the change trend of the resource value of the candidate virtual resource over time. The value change information of the candidate virtual resource within a time range is used to indicate the change trend of the resource value of the candidate virtual resource over time within the time range.

[0217] It should be noted that the above process is the same as the above Figure 3 or Figure 4 Similar to the embodiment shown, when the change trend image of the candidate virtual resources includes the numerical change information of the candidate virtual resources, based on the input time range, the target image of each candidate virtual resource is obtained, and the target image of the candidate virtual resource is used to represent the change trend of the resource value of the candidate virtual resource over time within the time range. The target virtual resources are screened out in combination with the image including the input numerical change information and the target image of each candidate virtual resource, which will not be repeated here.

[0218] It should be noted that the embodiment of the present application is explained by taking the example of searching for the target virtual resources in combination with the input numerical change information and the search conditions. In another embodiment, there is no need to execute the above steps 503 and 504, but other methods are adopted. In response to the resource search operation for the numerical change information, the target virtual resources are displayed in the resource interface, and the similarity between the change trend of the target virtual resources and the change trend represented by the numerical change information is greater than a threshold.

[0219] In the solution provided in the embodiment of the present application, in the resource interface, by inputting numerical change information representing the changing trend of the resource numerical value of the virtual resource to be searched over time, virtual resources whose changing trend of the resource numerical value matches the numerical change information can be automatically searched out, thereby realizing a new virtual resource search method, which can quickly search out the virtual resources that the user wants to view, and can realize the search of virtual resources even when the name of the virtual resource is unknown, thereby ensuring the convenience and accuracy of the virtual resource search, and also ensuring the efficiency of the resource search.

[0220] It should be noted that in the above Figures 2 to 5 Based on the embodiment shown, the embodiment of the present application also provides a schematic diagram of a resource interface, such as Figure 6 As shown, the resource interface displays an information input area, a condition input area and a result display area. The information input area is used to input numerical change information, the condition input area is used to input search conditions, and the result display area is used to display the searched virtual resources; the condition input area includes two sliders, one slider is used to input the risk type, and the other slider is used to input the time range. The user can adjust the position of the dots on the sliders through the terminal to adjust the input risk type or time range; taking the information input area as the drawing area as an example, the user draws an arbitrary curve in the drawing area through the terminal. When the drawing is completed, the terminal can search for virtual resources based on the drawn curve, the input risk type and the input time range. In the result display area, the numerical change information of the searched target virtual resources is displayed, such as the numerical change curve of the target virtual resources, the average change trend of the same type, etc.

[0221] The embodiment of the present application provides a resource interface, which allows users to specify search conditions according to their risk tolerance and the search time range of interest, and allows users to directly draw the change trend curve they expect on the resource interface, so as to search for the virtual resources that users want to search for in combination with the search conditions and the drawn curve, and can display the resource search results in the resource interface. This intuitive drawing method not only enables users to accurately express their search expectations, but also provides a clear reference for virtual resource searches, so as to make more accurate virtual resource matching, to ensure that the virtual group members searched are the virtual resources that users want to search for, to ensure the accuracy of the virtual resources searched, and to improve the convenience of searching for virtual resources, thereby improving the user experience.

[0222] In addition, the resource interface can also visualize the searched virtual resources, and can represent the change trend of the searched virtual resources in the form of a curve graph, and display the average change trend of the same type, etc., so that users can know the average change trend of the same type as the searched virtual resources, enriching the content of the search results and ensuring the display effect of the search results.

[0223] In a possible implementation, the embodiment of the present application also provides a feedback mechanism. For example, when the searched target virtual resource is displayed in the resource interface, an evaluation option is also displayed. The user can trigger any evaluation option through the terminal to evaluate the search result. Subsequently, the search algorithm for virtual resources can be optimized based on the evaluation feedback from the user. The evaluation option is any option, such as Figure 6 As shown, the evaluation options include a like option or a dislike option.

[0224] The embodiments of the present application also provide a user feedback mechanism, which allows users to evaluate search results. These feedbacks will be collected by the system and used to optimize the search algorithm to ensure that the search results can better meet the actual needs and changes of users over time.

[0225] It should be noted that in the above Figures 2 to 5 On the basis of the illustrated embodiment, in response to a resource search operation for numerical change information, a resource search model can be used to search for virtual resources that match the numerical change information. That is, the above process of determining the target virtual resources includes: in response to a resource search operation for numerical change information, a resource search model is used to perform target detection on the second image and the change trend images of multiple candidate virtual resources, respectively, to obtain a first local image and multiple second local images; a resource search model is used to determine the degree of match between the first local image and each second local image, and the candidate virtual resource corresponding to the second local image with the largest degree of match is determined as the target virtual resource.

[0226] In an embodiment of the present application, taking drawing a curve in a resource interface as an example, the drawn curve is numerical change information, and the image including the drawn curve is a second image. Through the resource search model, the local image including the curve can be extracted from the second image and the change trend images of multiple candidate virtual resources respectively, and then the local images are matched to obtain the target virtual resources that match the numerical change information to ensure the accuracy of the target virtual resources.

[0227] Among them, the resource search model can be any neural network model, for example, the resource search model is YOLOV3 (You Only Look Once Version 3) or YOLOV4 (You Only Look Once Version 4), CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory), RNN (Recurrent Neural Network), Transformer (a neural network model), Faster R-CNN (Faster Region-based Convolutional Neural Network) or Mask R-CNN (Mask Region-based Convolutional Neural Network).

[0228] In one possible implementation, a plurality of detection frame sizes are provided in the resource search model, and the process of performing target detection on the second image includes: performing feature extraction on the second image through the resource search model to obtain features of the second image, dividing the features of the second image into a plurality of grid feature units, traversing each grid feature unit, taking the center point of the traversed grid feature unit as the center, and based on the detection frame size corresponding to the scale, determining the detection frame indicated by the detection frame size from the traversed grid feature unit, detecting features included in the detection frame, and obtaining the probability of the detection frame.

[0229] It should be noted that the embodiment of the present application is described by taking the example of determining the target virtual resource based on the second image, while in another embodiment, when the numerical change information is input by uploading an image, according to the above method, the resource search model is used to perform target detection on the input first image and multiple change trend images respectively, and then the target virtual resource is determined. The above process is the same and will not be repeated here.

[0230] On the basis of the above-described embodiment, in the embodiment of the present application, before performing a virtual resource search through the resource search model, the resource search model needs to be trained. The training process is detailed in the following embodiment.

[0231] Figure 7 is a flow chart of a model training method provided in an embodiment of the present application, taking the method executed by a computer device as an example, Figure 7 As shown, the method includes:

[0232] 701. The computer device determines multiple detection box sizes based on the bounding box size of the curve in the multiple first sample images, where the curve is used to represent the changing trend of the resource value of the virtual resource over time, and the bounding box size is the size of the bounding box of the curve in the first sample image.

[0233] Among them, different first sample images include different curves, and the curve included in the first sample image is used to represent the changing trend of the resource value of any virtual resource over time. The curves included in different first sample images may be the changing trend of the resource value of the same virtual resource over time in different time ranges, or may be the changing trend of the resource values ​​of different virtual resources over time in the same or different time ranges. The bounding box can be a bounding box of any shape, for example, the bounding box is a rectangular or circular bounding box. For any first sample image, the bounding box in the first sample image can include the curve included in the first sample image. For example, taking the bounding box as a rectangle as an example, the first sample image includes a curve, and the size of the rectangular bounding box that includes the curve is the bounding box size.

[0234] In the embodiment of the present application, since the curves included in the multiple first sample images are different, the bounding box size of the curve in each first sample image is determined, so that multiple detection frame sizes can be determined from the bounding box sizes corresponding to the multiple first sample images, so that the determined detection frame size can be applicable to images containing arbitrary curves, so as to ensure that the local image including the curve can be accurately extracted from the image based on the detection frame size.

[0235] The number of the multiple detection frame sizes is an arbitrary number. For example, based on the bounding box sizes of the curves in the multiple first sample images, 9 detection frame sizes are determined.

[0236] In a possible implementation, the process of acquiring a plurality of first sample images includes: acquiring a plurality of change trend images of virtual resources, and determining the acquired change trend images as first sample images.

[0237] In an embodiment of the present application, the first sample image is a change trend image corresponding to any virtual resource. By acquiring the change trend images of multiple virtual resources, the detection frame size is determined based on the boundary frame size of the curve in the change trend image, so as to ensure the accuracy of the subsequent extraction of the local image containing the curve from the change trend image using the detection frame size.

[0238] In a possible implementation, when multiple first sample images are acquired, any method can be used to determine the size of the bounding box of the curve in the first sample image. For example, each first sample image is manually labeled to determine the size of the bounding box of the curve in each first sample image.

[0239] 702. The computer device obtains a reference image and a second sample image, wherein the first curve in the reference image is used to represent a changing trend of a resource value of the virtual resource to be searched over time, and the similarity between the first curve in the reference image and the second curve in the second sample image is greater than a threshold.

[0240] In the embodiment of the present application, the reference image is equivalent to an image containing numerical change information, which is used as the input of the resource search model, and the second sample image is equivalent to a change trend image corresponding to the virtual resource matching the reference image. The second sample image may be any sample image among the multiple first sample images, or any sample image other than the multiple first sample images.

[0241] In one possible implementation, the method of obtaining the reference image is the same as that described above. Figure 3 or Figure 4 Similar to the embodiment shown, any curve can be drawn in the resource interface, and a reference image can be generated based on the drawn curve; or, the reference image is based on a photographed image.

[0242] 703. The computer device performs target detection on the reference image and the second sample image respectively based on a plurality of detection frame sizes through a resource search model to obtain a first local image and a second local image, wherein the first local image is a local image including the first curve in the reference image, and the second local image is a local image including the second curve in the second sample image.

[0243] In one possible implementation, the resource search model includes a feature extraction sub-model and a detection sub-model, and step 703 includes: extracting features from a reference image through the feature extraction sub-model to obtain image features; detecting image features based on multiple detection frame sizes through the detection sub-model to obtain prediction parameters, the prediction parameters indicating the position of a curve in the reference image; and extracting a first local image from the reference image based on the prediction parameters.

[0244] It should be noted that the process of extracting the first partial image from the reference image through the resource search model is the same as the process of extracting the partial image from the second image in step 303, and will not be repeated here.

[0245] In the embodiment of the present application, the reference image can be convolved into multi-scale image features through the resource search model, and the resource search locates and classifies the curves based on the multi-scale image features. Due to the difference between the image features and the reference image, the parameters extracted from the image features and used to represent the position of the first curve in the reference image cannot directly extract the local image from the reference image. Therefore, Figure 8 As shown, after the image features extracted by the resource search model are gridded, all grid feature units on the gridded image features are traversed, and multiple detection box parameters, corresponding probabilities and confidences are generated in combination with the detection box size. The spatial invariance of the resource search model is used to map the detection box parameters back to the original image, that is, mapped back to the reference image.

[0246] In an embodiment of the present application, a resource search model is used to obtain multiple detection frame parameters, corresponding probabilities and confidences, and the detection frame parameters are decoded to determine the corresponding positions of the detection frame parameters in the reference image. Since the resource search model convolves the reference image into multi-scale image features, the multi-scale image features need to be processed when decoding the detection frame parameters. Subsequently, according to the coordinates of the grid feature units on the gridded image features and the detection frame parameters, the detection frame parameters predicted on the gridded image features are mapped back to the reference image through the following relationship:

[0247] b x =σ(t x +c x )

[0248] b y =σ(t y +c y )

[0249]

[0250] Among them, the size of the detection box on the gridded image feature is represented by c x 、c y 、p w 、p h , c x and c y Used to represent the center coordinates in the detection box parameters, p w and p h Used to represent the width and height of the detection box parameters; t w and t hUsed to represent the displacement of the detection box parameters and the grid feature unit coordinates; t x and t y It is used to represent the fine-tuning coefficient of the center coordinates in the detection frame parameters. The size of the detection frame after mapping back to the reference image is represented by b x , b y , b w , b h , b x and b y Used to represent the center coordinates in the detection box parameters, b w and b h It is used to represent the width and height of the detection box parameters, and σ(·) is used to represent an activation function.

[0251] 704. The computer device determines the matching degree between the first partial image and the second partial image through a resource search model.

[0252] The process of step 704 is similar to the process of determining the matching degree between the first partial image and the second partial image in step 303, and will not be described in detail here.

[0253] 705. The computer device trains the resource search model based on the matching degree.

[0254] In an embodiment of the present application, under real circumstances, the matching degree between the reference image and the second sample image should be 1, then the matching degree obtained based on the resource search model can reflect the accuracy of the resource search model, and the resource search model is trained based on the matching degree to improve the accuracy of the resource search model.

[0255] In one possible implementation, step 705 includes: determining a first loss value based on the matching degree, and training a resource search model based on the first loss value.

[0256] In an embodiment of the present application, a first loss value is determined based on the degree of matching so that the first loss value can reflect the accuracy of the resource search model, and then the resource search model is trained based on the first loss value to improve the accuracy of the resource search model.

[0257] In the solution provided in the embodiment of the present application, since the curves included in the multiple first sample images are different, the bounding box size of the curve in each first sample image is determined, so that multiple detection frame sizes are determined from the bounding box sizes corresponding to the multiple first sample images, so that the determined detection frame size can be applicable to images containing arbitrary curves, so as to ensure that the local image including the curve can be accurately extracted from the image based on the detection frame size in the future, and the determined multiple detection frame sizes are set in the resource search model, and then the resource search model is trained based on the reference image and the second sample image to improve the accuracy of the resource search model.

[0258] It should be noted that the above Figure 7 The embodiment shown is based on the training of the resource search model based on the reference image and the first sample image as an example for explanation, and in another embodiment, after step 701, the resource search model can be iteratively trained according to the above steps 702-705. In the process of iterative training of the resource search model, when the loss value of the current iteration is less than the third threshold value, or when the number of iterations reaches the number threshold value, the training of the resource search model is stopped.

[0259] It should be noted that the above Figure 7 The embodiment shown is to illustrate the training of the resource search model based on the reference image and the second sample image as an example, and in another embodiment, the resource search model can also be trained based on the reference image, the second sample image and multiple third sample images. Wherein, if the similarity between the curve in the third sample image and the first curve in the reference image is lower than the threshold, then according to the above step 703, the fourth partial image containing the curve is extracted from the third sample image, and according to the above step 704, the matching degree between the first partial image and the fourth partial image is determined, and then based on the matching degree between the first partial image and the second partial image, and the matching degree between the first partial image and the fourth partial image, the resource search model is trained.

[0260] In the above Figure 7 On the basis of the embodiment shown, in the embodiment of the present application, the feature extraction submodel in the resource search model is to construct a deep convolutional neural network with excellent performance, fast detection speed and high detection accuracy. The backbone network inside the feature extraction submodel will combine the multi-scale fusion structure to extract discriminative abstract features from the image so that the resource search model can use these features for learning and recognition. Taking the feature extraction submodel as Darknet-53 (a neural network model) as an example, the feature extraction submodel aims to extract discriminative features between different types of targets from the image. The stronger the expression ability of the extracted features, the more conducive to the distinction of targets. The embodiment of the present application adopts a feature extraction method based on deep learning, and automatically extracts image features with good expression ability by constructing a feature extraction submodel, thereby improving the detection effect of the resource search model. At the same time, when constructing the feature extraction submodel, the depth and parameter size of the feature extraction submodel are fully considered to facilitate the trade-off between the detection speed and detection accuracy of the resource search model.

[0261] Darknet-53 is one of the convolutional neural networks that has a good balance between detection speed and detection accuracy. Therefore, it is used as the feature extraction submodel in the resource search model. The structure of the feature extraction submodel is as follows: Fig. 9As shown in the figure, the number on the left outside the box represents the number of repetitions of the residual block, that is, the network uses 1×1 and 3×3 convolutional layers alternately and continuously to extract features from the input image, and scales the size of the feature map through a convolution kernel with a step size of 2. Darknet-53 is a fully convolutional neural network, that is, it abandons the pooling layer and the fully connected layer, and the network contains only two structural components: convolutional layer (Convolutional) and residual block (Residual). The structure of the convolutional layer is shown in the figure. Fig.10 As shown in the figure, the convolutional layer consists of a convolution operation (Conv2D), a batch normalization layer (BN Layer), and a nonlinear function activation layer (Leaky ReLU Layer). Each residual block is stacked with convolutional layers with 1×1 and 3×3 convolution kernel sizes. Among them, the branch connection of the identity mapping (Identity) allows the output of the shallow layer to be used as the input of the deep layer. This structure facilitates the information flow of the deep neural network and can avoid the situation where the gradient gradually disappears in the back propagation as the number of layers increases.

[0262] In the process of extracting features from the input image through the feature extraction sub-model, Darknet-53 will generate image features of multiple scales. The multi-scale image feature fusion structure draws on the idea of ​​the Feature Pyramid Network, which aims to fuse the image features of three different depths in the feature extraction sub-model. The structure after Darknet-53 is combined with the multi-scale feature fusion structure is as follows: Fig.11As shown. In the embodiment of the present application, before the image is input into the feature extraction submodel, it will be uniformly processed into a tensor of 416×416×3. After passing through the continuous residual blocks in Darknet-53, the size of the image features will be further reduced by 32 times, and the extracted image feature tensor is 13×13×1024. That is, the minimum image feature size is 13×13, and the dimension of the image feature is 1024. Then, according to the category to be detected, the convolution group compresses the feature tensor to the corresponding dimension. The calculation formula for the dimension is 3×(5+n), which means that after dividing the image features of any scale into multiple grid units, each grid unit of the feature map corresponds to 3 detection frame sizes, so that the grid units can be used to detect targets using these 3 detection frame sizes. Each grid unit contains 5 detection frame parameters and n categories. The feature tensor extracted in the first branch is 13×13×27. In addition, in order to improve the detection accuracy of small targets, the feature tensor in the first branch is upsampled twice in the 79th and 91st layers through a multi-scale feature fusion structure, and is concatenated with the feature tensors extracted from the 11th and 19th residual blocks, respectively. Therefore, the feature tensors extracted by the second and third branches are 26×26×27 and 52×52×27, respectively. The three different feature tensors extracted by this network structure have achieved good results when applied to large, medium, and small scale target detection.

[0263] On the basis of the above-mentioned embodiment, in the process of training the resource search model, for the input reference image and the second sample image, the real marking frame of the curve in the reference image and the real marking frame of the curve in the second sample image can also be used to train the resource search model. The real marking frame of the curve in the reference image is used to indicate the real position of the curve in the reference image, and the real marking frame of the curve in the second sample image is used to indicate the real position of the curve in the second sample image.

[0264] In a possible implementation, one or more of the following four loss functions can be used to train the resource search model.

[0265] The first loss function indicates the coordinate error of the center point of the marked box, which is composed of the distance difference between the marked box and the center point of the ground truth. The first loss function satisfies the following relationship:

[0266]

[0267] Among them, Loss coord Used to represent the center loss value, S 2 It is used to indicate the number of grid units into which the image features are divided, B is used to indicate the number of detection frame sizes corresponding to the grid units, i is used to indicate the sequence number of the grid units, and j is used to indicate the sequence number of the detection frame sizes corresponding to the grid units. Used to indicate whether the detection frame contains a curve; and Used to represent the center coordinates of the detection box parameters obtained based on the resource search model, and Used to represent the center coordinates of the ground-truth labeled box.

[0268] The second loss function indicates the width and height error of the marked box, which is composed of the width difference and height difference between the marked box and the true marked box. The second loss function satisfies the following relationship:

[0269]

[0270] Among them, Loss wh Used to indicate the width and height loss value of the marking box, S 2 It is used to indicate the number of grid units into which the image features are divided, B is used to indicate the number of detection frame sizes corresponding to the grid units, i is used to indicate the sequence number of the grid units, and j is used to indicate the sequence number of the detection frame sizes corresponding to the grid units. Used to indicate whether the detection frame contains a curve; and Used to represent the width and height of the detection box parameters obtained based on the resource search model. and Used to represent the width and height of the ground truth marker box.

[0271] The third loss function indicates the confidence error, which is calculated by the cross entropy formula to obtain the confidence of the object to be identified contained in the marked box. It is further divided into two sub-parts, one for the confidence of the object to be identified contained in the marked box, and the other for the confidence of the object to be identified not contained in the marked box. A high proportion of background information in the image can easily lead to the confidence of the object to be identified not contained in the marked box being higher than the confidence of the object to be identified contained in the marked box. In order to balance the impact of the two on the model, the balance coefficient λ is introduced in the embodiment of the present application. noobj , the formula part of the object to be identified that is not included in the marked box is scaled to achieve a balance between the two values. The third loss function satisfies the following relationship:

[0272]

[0273] Among them, Loss confidence Used to represent the confidence loss value, S 2It is used to indicate the number of grid units into which the image features are divided, B is used to indicate the number of detection frame sizes corresponding to the grid units, i is used to indicate the sequence number of the grid units, and j is used to indicate the sequence number of the detection frame sizes corresponding to the grid units. Used to indicate whether the detection frame contains a curve; It is used to indicate the confidence level of the detection box parameters obtained based on the resource search model. Used to represent the confidence level based on the true labeled box, λ noobj Used to indicate the balance factor Used to indicate whether the detection box does not contain the curve.

[0274] The fourth loss function indicates the classification error. This part uses the cross entropy calculation formula to obtain the error between the probability of each category to which each object in the marked box belongs and the true category. The fourth loss function satisfies the following relationship:

[0275]

[0276] Among them, Loss class Used to represent the classification loss value, S 2 It is used to indicate the number of grid units into which the image features are divided, and i is used to indicate the sequence number of the grid unit. Used to indicate whether the detection frame contains a curve; It is used to represent the probability of each object belonging to each category within the detection box parameters obtained based on the resource search model. Used to represent the true category corresponding to the true labeled box.

[0277] Based on the four loss functions shown above, the resource search model can also be trained using the sum of the four loss functions. The sum of the four loss functions satisfies the following relationship:

[0278] Loss=Loss coord +Loss wh +Loss confidencce +Loss class

[0279] Among them, it is used to represent the sum of four loss functions, Loss coord Used to indicate the center loss value, Loss wh Used to indicate the width and height loss value of the marker box, Loss confidence Used to represent the confidence loss value, Loss class Used to represent the classification loss value.

[0280] Based on the above-mentioned embodiment, the training process of the resource search model is as follows: Fig.12As shown in the figure, by setting the hyperparameters in the resource search model training process, the weight parameters of each layer of the resource search model are fine-tuned. First, the input image is isomorphized and scaled to a uniform size; then the weight parameters of the resource search model are initialized and the fitting loss function is constructed; finally, the hyperparameters of the resource search model are set and training tuning techniques are added.

[0281] In the above Figure 7 On the basis of the illustrated embodiment, the embodiment of the present application can adopt a clustering approach and, in combination with the adaptability of the size, determine multiple detection box sizes based on multiple bounding box sizes. The specific process is detailed in the following embodiment.

[0282] Fig.13 is a flow chart of a model training method provided in an embodiment of the present application, taking the method executed by a computer device as an example, Fig.13 As shown, the method includes:

[0283] 1301. A computer device clusters a plurality of bounding box sizes to obtain a plurality of clusters, each of which includes at least one bounding box size.

[0284] In an embodiment of the present application, for the bounding box sizes of curves in multiple first sample images, a clustering approach is adopted to divide the multiple bounding box sizes into multiple cluster clusters according to the similarity between the multiple bounding box sizes, so as to reflect the diversity of the bounding box sizes of curves in different sample images.

[0285] The number of the plurality of clusters can be any number, for example, the number of the clusters is 9.

[0286] In a possible implementation, step 1301 includes: randomly selecting a preset number of bounding box sizes from multiple bounding box sizes, and using the selected bounding box sizes as center sizes; for each bounding box size other than the center size among the multiple bounding box sizes, based on the similarity between the bounding box size and each center size, assigning the bounding box size to the center size with the greatest similarity, and obtaining a preset number of first clusters; in the case where a preset number of i-th clusters are currently obtained, determining the median of the sizes included in each i-th cluster as a new center size, and obtaining a preset number of new center sizes; for each bounding box size other than the new center size among the multiple bounding box sizes, based on the similarity between the bounding box size and each new center size, assigning the bounding box size to the new center size with the greatest similarity, and obtaining a preset number of i+1-th clusters.

[0287] The preset number is the number of clusters to be clustered, and the preset number is an arbitrary number, for example, the preset number is 9. i is an integer greater than 0. For the median value of the sizes included in the i-th cluster, the sizes included in the i-th cluster are sorted from large to small or from small to large, and the size in the middle position is determined as the median value of the sizes included in the i-th cluster.

[0288] In an embodiment of the present application, for multiple bounding box sizes, a preset number of bounding box sizes are first randomly selected from the multiple bounding box sizes as the center size, and the multiple bounding box sizes are clustered according to the similarity between the bounding box size and the center size. Then, the median of the size included in the cluster is used as the new center size, and the multiple bounding box sizes are clustered again. The above process is repeated to obtain the final multiple clusters, thereby ensuring the accuracy of the obtained clusters.

[0289] In an embodiment of the present application, K-Medians (a median-based clustering algorithm) clusters multiple bounding box sizes and takes the median rather than the average value in the cluster to update the center of the cluster, thereby avoiding the adverse effect of the outlier values ​​in the cluster on the center of the cluster to ensure clustering accuracy and the accuracy of the subsequently determined detection box size.

[0290] Optionally, the process of determining the similarity between the bounding box size and the center size includes: determining the similarity between the bounding box size and the center size based on a feature of the bounding box size and a feature of the center size.

[0291] Optionally, the similarity between the bounding box size and the center size satisfies the following relationship:

[0292] d(i,j)=|x i -x j |+|y i -y j |

[0293] Wherein, d(i, j) is used to indicate the similarity between the bounding box size and the center size, i is used to indicate the ordinal number of the bounding box size, and j is used to indicate the ordinal number of the center size in multiple bounding box sizes; x i The feature used to represent the size of the bounding box is the x-axis coordinate in the feature space, x j The x-axis coordinate of the feature used to represent the center size in the feature space; i The Y-axis coordinate of the feature used to represent the size of the bounding box in the feature space, y i The Y-coordinate of the feature used to represent the center size in feature space.

[0294] In the embodiment of the present application, the Manhattan distance is adopted to determine the similarity between the bounding box size and the center size.

[0295] Optionally, in the above process of updating a preset number of clusters, for each k-th cluster, when the difference between the median of the size included in the k-th cluster and the center size of the k-th cluster is less than a second threshold, stop updating the cluster, where k is an integer greater than 1.

[0296] In the embodiment of the present application, when the updating of the clustering cluster is stopped, the current clustering cluster is used as the final clustering cluster, so that the following step 1302 is performed using the current clustering cluster.

[0297] In the embodiment of the present application, the second threshold is an arbitrary value. If the difference between the median of the size included in the kth cluster and the center size of the kth cluster is less than the second threshold, it means that the current cluster is accurate enough and there is no need to update the cluster.

[0298] 1302. The computer device generates a plurality of candidate size sets based on the center sizes of a plurality of clusters, each candidate size set including at least one candidate size.

[0299] In the embodiment of the present application, when multiple clusters are determined, a candidate size set is generated based on the center size of each cluster, thereby obtaining multiple candidate size sets. The candidate sizes included in each candidate size set include at least the center size of one cluster.

[0300] In a possible implementation, step 1302 includes: for each cluster center size, performing multiple random perturbations on the center size to obtain multiple candidate sizes; and combining the multiple candidate sizes and the center size into a candidate size set.

[0301] In an embodiment of the present application, for the center size of any cluster, the center size is randomly perturbed multiple times to generate candidate sizes that are different from the center size, and the candidate sizes and the center size are combined into a candidate size set, so as to ensure the accuracy of the candidate sizes in the candidate size set while ensuring the diversity of the candidate sizes in the candidate size set, thereby ensuring the accuracy of the detection frame size determined subsequently.

[0302] In the embodiment of the present application, according to the above method, for the center size of each cluster, multiple candidate size sets can be obtained.

[0303] Among them, randomly perturbing the center size refers to fine-tuning the center size to obtain a size that is different from the center size, and taking the obtained size as a candidate size. Each time the center size is randomly perturbed once, a candidate size can be obtained, and the center size is randomly perturbed multiple times to obtain multiple candidate sizes.

[0304] Optionally, the method of randomly perturbing the center size includes: randomly determining a perturbation amplitude from a perturbation amplitude range, and adding the center size to the determined perturbation amplitude to obtain a candidate size.

[0305] For example, the center size includes the height and width of the bounding box. The high perturbation amplitude or the wide perturbation amplitude is randomly determined from the perturbation amplitude range, and the height of the center size is merged with the high perturbation amplitude to obtain a new size as a candidate size; or, the width of the center size is merged with the wide perturbation amplitude to obtain a new size as a candidate size.

[0306] 1303. The computer device updates each candidate size set based on the fitness of the candidate sizes in the candidate size set to obtain a new candidate size set, where the fitness is used to indicate the degree of matching between the candidate size and the sizes of the multiple bounding boxes.

[0307] In an embodiment of the present application, after obtaining multiple candidate size sets, the candidate size sets can be updated in combination with the fitness of the candidate sizes in the candidate size sets to improve the accuracy of the sizes included in the candidate size sets, so that a more accurate detection frame size can be determined subsequently.

[0308] In one possible implementation, the process of determining the fitness of the candidate size includes: determining the matching degree between the candidate size and each bounding box size, and determining the maximum matching degree corresponding to the candidate size as the fitness of the candidate size; or, determining the average value of multiple matching degrees corresponding to the candidate size as the fitness of the candidate size.

[0309] In the embodiment of the present application, the matching degree between the candidate size and the bounding box size can reflect the overlap between the bounding box indicated by the candidate size and the bounding box indicated by the bounding box size. The larger the overlapping area between the bounding box indicated by the candidate size and the bounding box indicated by the bounding box size, the greater the matching degree between the candidate size and the bounding box size.

[0310] For example, in any image, with the same pixel as the center, based on the candidate size and the bounding box size, a bounding box size corresponding to the candidate size and a bounding box corresponding to the bounding box size are drawn in the image, and the matching degree between the candidate size and the bounding box size is equivalent to the intersection and union ratio of the two bounding boxes drawn in the image.

[0311] Optionally, the matching degree between the candidate size and any bounding box size satisfies the following relationship:

[0312]

[0313] Among them, Fitness is used to indicate the matching degree between the candidate size and any bounding box size. Used to indicate the width of the candidate size. Used to indicate the height of the candidate size; indiv annot,w Used to indicate the width of the bounding box size, indiv annot,h The height used to represent the size of the bounding box.

[0314] In the embodiment of the present application, the intersection over union (IoU) method is adopted to determine the matching degree between the candidate size and the bounding box size to ensure the accuracy of the matching degree.

[0315] In one possible implementation, step 1303 includes the following steps 1-3.

[0316] Step 1: based on the fitness of the candidate sizes in the candidate size set, a candidate size subset is screened out from the candidate size set, where the candidate size subset includes at least one candidate size.

[0317] In an embodiment of the present application, based on the fitness of the candidate sizes in the candidate size set, a candidate size subset is screened out from the candidate size set, so that the candidate size subset includes at least one candidate size, so as to screen out a candidate size with a sufficiently high fitness, and the candidate sizes in the candidate size subset maintain diversity.

[0318] In a possible implementation, step 1 includes: determining a first sum value, the first sum value being the fitness sum value of the candidate sizes in the candidate size set; generating a random number from a numerical range consisting of 0 and the first sum value; starting from any candidate size in the candidate size set, traversing the candidate sizes in the candidate size set; when the second sum value when traversing to the i-th candidate size is less than the random number and the third sum value when traversing to the i+1-th candidate size is greater than the random number, determining the i+1-th candidate size as the candidate size in the candidate size subset, the second sum value is the fitness sum value of the i candidate sizes that have been traversed, and the third sum value is the fitness sum value of the i+1 candidate sizes that have been traversed, and i is an integer greater than 0.

[0319] The random number is not less than 0 and not greater than the first sum value. As for the method of traversing the candidate sizes in the candidate size set, a random traversal method can be adopted for traversal.

[0320] In an embodiment of the present application, after determining the first sum value, a random number is generated from a numerical range consisting of 0 and the first sum value, so that a candidate size is determined from the candidate size set in the above manner, and the determined candidate size is used as a candidate size in the candidate size subset; thereafter, a random number is generated again from a numerical range consisting of 0 and the first sum value, so that another candidate size is determined from the candidate size set in the above manner, and the determined candidate size is used as a candidate size in the candidate size subset, and the above process is repeated to filter out the candidate size subset from the candidate size set.

[0321] In an embodiment of the present application, candidate sizes are screened from a candidate size set in the above manner to ensure that the candidate sizes in the candidate size subset have higher fitness, and the candidate sizes in the candidate size subset are different from each other, so that the candidate sizes in the candidate size subset have diversity to ensure the accuracy of the detection frame size determined subsequently.

[0322] Step 2: Generate a new candidate size based on the candidate sizes in the candidate size subset.

[0323] In the embodiment of the present application, when a candidate size subset is obtained, the candidate sizes can be processed based on the candidate sizes in the candidate size subset to generate new candidate sizes to enrich the diversity of the candidate sizes.

[0324] In one possible implementation, step 2 includes: randomly selecting two candidate sizes from the candidate size subset, exchanging part of the data at the same position in the two candidate sizes, and obtaining two new candidate sizes; or, for any candidate size in the candidate size subset, reversing part of the data in the candidate size to obtain a new candidate size.

[0325] In the embodiment of the present application, in the process of generating a new candidate size, part of the data at the same position in the two candidate sizes is exchanged to achieve cross processing of the two candidate sizes; part of the data in the candidate size is reversed to change the arrangement order of this part of the data, so as to achieve variation processing of the candidate size. By adopting methods such as crossing or mutating the candidate sizes to generate new candidate sizes, the obtained candidate sizes are enriched as much as possible, the diversity of the new candidate sizes is guaranteed, so as to ensure the diversity of the candidate sizes in the subsequent candidate size set, and thus ensure the accuracy of the subsequently determined detection frame size.

[0326] Optionally, the process of obtaining a new candidate size by means of data exchange includes: when two candidate sizes are selected, generating an exchange probability, the exchange probability indicating the possibility of exchanging partial data at the same position in the two candidate sizes; when the exchange probability is greater than a first probability threshold, exchanging partial data at the same position in the two candidate sizes to obtain two new candidate sizes.

[0327] In an embodiment of the present application, an exchange probability, which is also equivalent to a crossover probability, is randomly generated for two randomly selected candidate sizes. When the exchange probability is greater than a first probability threshold, partial data at the same position in the two candidate sizes are exchanged; and when the exchange probability is less than the first probability threshold, partial data in the two candidate sizes are no longer exchanged.

[0328] In an embodiment of the present application, an interchange probability is randomly generated for two randomly selected candidate sizes. Based on the interchange probability and a first probability threshold, when the interchange probability is greater than the first probability threshold, part of the data at the same position in the two candidate sizes is interchanged to ensure the randomness of the generated new candidate size, thereby ensuring the diversity of the candidate sizes subsequently obtained.

[0329] Optionally, the process of exchanging partial data at the same position in two candidate sizes includes: generating an exchange interval for the two candidate sizes, and exchanging data in the exchange feature interval in the two candidate sizes to obtain two new candidate sizes.

[0330] The generated swap interval is an interval corresponding to the partial data to be swapped. Optionally, based on the data intervals of the two candidate sizes, a partial interval is randomly determined from the data interval as the swap interval.

[0331] For example, the two candidate sizes are represented as "abcdef" and "123456", and the generated exchange interval is [3,5]. That is, the data from the 3rd to the 5th bit of the two candidate sizes are exchanged, that is, the "cde" and "345" of the two candidate sizes are exchanged, and the two new candidate sizes are represented as "ab345f" and "12cde6" respectively.

[0332] Optionally, taking the two candidate sizes as a first candidate size and a second candidate size as an example, the process of exchanging partial data at the same position in the two candidate sizes satisfies the following relationship:

[0333]

[0334] in, Used to indicate the first candidate size The data in the interval [x1, x2] are swapped. Used to represent the second candidate size The data of the interval [x1, x2] are interchanged; the first candidate size after the interchange is called the third candidate size, and the second candidate size after the interchange is called the fourth candidate size; Used to represent the third candidate size The data in the interval [x1, x2] are swapped. Used to represent the fourth candidate size Swap the data in the interval [x1, x2].

[0335] It should be noted that the embodiment of the present application only takes the example of exchanging partial data at the same position in any two candidate sizes in the candidate size subset, and in another embodiment, partial data at the same position in every two candidate sizes in the candidate size subset can be exchanged in the above manner, which will not be repeated here.

[0336] Optionally, the process of obtaining a new candidate size by reverse processing includes: determining a mutation probability corresponding to the candidate size, the mutation probability indicating the possibility of reverse processing of part of the data in the candidate size; when the mutation probability is greater than a second probability threshold, reverse processing of part of the data in the candidate size to obtain a new candidate size.

[0337] In an embodiment of the present application, for any candidate size in the candidate size subset, the mutation probability corresponding to the candidate size is determined, and when the mutation probability is greater than the second probability threshold, part of the data in the candidate size is reversed; and when the mutation probability is less than the second probability threshold, part of the data in the candidate size is no longer reversed.

[0338] In an embodiment of the present application, for any candidate size in the candidate size subset, the mutation probability corresponding to the candidate size is determined. When the mutation probability is greater than the second probability threshold, part of the data in the candidate size is processed in reverse order to ensure the randomness of the generated new candidate size, thereby ensuring the diversity of the candidate sizes obtained subsequently.

[0339] Optionally, the process of performing reverse processing on part of the data in the candidate size includes: generating a variation interval for the candidate size, and performing reverse processing on the data in the variation interval in the candidate size to obtain a new candidate size.

[0340] The generated variation interval is an interval corresponding to the part of the data to be processed in reverse order. Optionally, based on the data interval of the candidate size, a part of the interval is randomly determined from the data interval as the variation interval.

[0341] For example, the candidate sizes are represented as "abcdef", and the generated variation interval is [3,5]. That is, the data from the 3rd to the 5th bit in the candidate size is processed in reverse order, that is, the "cde" in the candidate size is processed in reverse order, and the new candidate size is represented as "abedcf".

[0342] Optionally, the data in the variation interval of the candidate size is processed in reverse order to satisfy the following relationship:

[0343]

[0344] in, Used to represent any candidate size in the candidate size subset. Used to represent candidate sizes Data in the variation interval [y1, y2]; Reverse(·) is used to represent the reverse processing function, Used to indicate the new candidate size obtained.

[0345] It should be noted that the embodiment of the present application is described by taking the reverse processing of part of the data in any candidate size in the candidate size subset as an example, while in another embodiment, for each candidate size in the candidate size subset, the reverse processing is performed on part of the data in each candidate size subset in accordance with the above method.

[0346] Step 3: The candidate sizes in the candidate size subset and the new candidate sizes are combined into a new candidate size set.

[0347] In an embodiment of the present application, for any candidate size set, based on the fitness of the candidate sizes in the candidate size set, candidate sizes with sufficiently high fitness and maintaining diversity are screened out to form a candidate size subset, and then new candidate sizes are generated based on the candidate sizes in the candidate size subset, and the candidate sizes in the candidate size subset and the new candidate sizes are formed into a new candidate size set, thereby realizing an update of the candidate size set. In this way, the candidate sizes in the updated candidate size set have sufficiently high fitness and maintain diversity, and can have a variety of different candidate sizes to improve the accuracy of the candidate size set.

[0348] It should be noted that the above description is only based on the example of updating any candidate size set. In another embodiment, each candidate size set can be updated according to the above steps 1 to 3.

[0349] It should be noted that the present application only takes updating any candidate size set once as an example for explanation. In another embodiment, after updating multiple candidate size sets once according to the above steps 1 to 3, multiple new candidate size sets can be updated again according to the above steps 1 to 3 to achieve multiple iterative updates of multiple candidate size sets.

[0350] In one possible implementation, during multiple iterative updates of multiple candidate size sets, the second probability threshold will change with the number of iterations, that is, the process of determining the second probability threshold includes: during multiple iterative updates of multiple candidate size sets, based on a third probability threshold, the current number of iterations and the total number of iterations, determining the second probability threshold, the third probability threshold being the probability threshold when the multiple candidate size sets are updated for the first time.

[0351] The third probability threshold is an arbitrary value, for example, the third probability threshold is 0.3, and the third probability threshold is an initial probability threshold used when updating multiple candidate size sets. The total number of iterations is an arbitrary value, for example, the total number of iterations is 10.

[0352] In an embodiment of the present application, during multiple iterative updates of multiple candidate size sets, the second probability threshold corresponding to the mutation probability will change with the number of iterations, so that during the process of updating the candidate size set, the local optimal solution can be jumped out, so that the global optimal candidate size can be subsequently determined from the updated candidate size set as the final detection frame size to ensure the accuracy of the subsequent detection frame size.

[0353] Optionally, in the process of performing multiple iterative updates on multiple candidate size sets, the third probability thresholds corresponding to different candidate size sets may be different, and the second probability thresholds used by different candidate size sets in the process of performing multiple iterative updates may be different.

[0354] Optionally, in the process of performing multiple iterations to update multiple candidate size sets, the second probability threshold used by any candidate size set in any iteration satisfies the following relationship:

[0355]

[0356] Among them, p is used to represent the sequence number of multiple candidate size sets, q is used to represent the current number of iterations, and MutationRate pq It is used to indicate the second probability threshold used by the p-th candidate size set in the q-th update, MutationRate p1It is used to represent the third probability threshold used by the p-th candidate size set during the first update, and n is used to represent the total number of iterations.

[0357] In an embodiment of the present application, a dynamic adjustment method is adopted for the second probability threshold. According to the current number of iterations of the current candidate size set and the total number of iterations, the second probability threshold changes dynamically with the number of iterations, and the second probability threshold changes within the range of 1-3 times the initial mutation rate (i.e., the third probability threshold), so that in the process of updating the candidate size set, the local optimal solution can be jumped out, so that the global optimal candidate size can be subsequently determined from the updated candidate size set as the final detection frame size, so as to ensure the accuracy of the subsequent detection frame size.

[0358] It should be noted that the above steps 1 to 3 are to first select a candidate size subset from the candidate size set, and then generate a new candidate size for the candidate size in the candidate size subset, and then form a new candidate size set by the new candidate size and the candidate size in the candidate size subset. In another embodiment, after step 1302, other methods can be adopted to update the candidate size set to obtain a new candidate size set. In a possible implementation, the process of obtaining a new candidate size set includes: for each candidate size set, first generate a new candidate size according to the above step 2, that is, based on the candidate size in the candidate size set, and then according to the above step 3, each candidate inner size set and the new candidate size are combined into an alternative size set, and then according to the above step 1, based on the fitness of the candidate size in the alternative size set, a candidate size subset is selected from the alternative size set as a new candidate size set.

[0359] 1304. The computer device determines the candidate size with the greatest fitness in each new candidate size set as the detection frame size.

[0360] In an embodiment of the present application, among multiple new candidate size sets obtained in the above manner, the candidate size with the largest fitness in each new candidate size set is determined as the final detection frame size, so that multiple detection frame sizes can be obtained to ensure the accuracy of the detection frame size.

[0361] In the solution provided by the embodiment of the present application, multiple bounding box sizes are first clustered, and then multiple center sizes that can represent multiple bounding box sizes are screened out based on the clustering clusters, and a candidate size set is generated based on the multiple center sizes, so that the candidate size set is updated in combination with the fitness of the candidate sizes in the candidate size set, so as to improve the diversity of the candidate sizes in the candidate size set while ensuring that the fitness of the candidate sizes in the candidate size set is sufficiently high, and then the candidate size with the largest fitness in each new candidate size set is determined as the final detection frame size, so that the detection frame size can represent multiple bounding box sizes, ensuring that the matching degree of the detection frame size and the multiple bounding box sizes is sufficiently high, and avoiding that the multiple detection frame sizes are too similar, and ensuring the diversity of the multiple detection frame sizes, and then ensuring the accuracy of the detection frame size. In addition, it is ensured that the detection frame size can be applicable to a variety of different change trend images, and then the detection frame size is set in the resource search model, so that the resource search model uses the detection frame size to search for virtual resources, which can ensure the accuracy of resource search.

[0362] In the embodiment of the present application, the K-Medians algorithm is adopted to avoid the adverse effects of outliers in multiple bounding box sizes. After that, a multi-population microhabitat genetic method based on feature selection is adopted. The multi-population microhabitat genetic method has the advantages of large detection space, large population diversity, and high survey efficiency, thereby improving the efficiency of determining the detection frame size, ensuring the matching degree of the detection frame size with multiple bounding box sizes, avoiding the situation where the determined detection frame size is the optimal solution in the local bounding box size, and can further optimize the candidate size set, thereby ensuring the accuracy of the determined detection frame size.

[0363] In the above Figure 7 Based on the embodiment shown in FIG. 1 , the embodiment of the present application further provides a flowchart for determining the size of a detection frame. When multiple candidate size sets are obtained according to step 1302, the multiple candidate size sets are iteratively updated, such as Fig.14As shown, in the process of iteratively updating multiple candidate size sets, the fitness of each candidate size in each candidate size set is determined, and it is determined whether the iteration termination condition is currently met, and the iteration termination condition refers to that the number of iterations reaches the total number of iterations; if the iteration termination condition is not currently met, a candidate size subset is selected from the candidate size set according to the above step 1, and then, according to the method provided in the above step 2, the sizes in the candidate size subset are crossover and mutation processed, and then, in a niche manner, the candidate sizes in the candidate size subset and the new candidate sizes are formed into a new candidate size set according to the above step 3; the new candidate size set is migrated, and only the new candidate size set is retained, and the original candidate size set is deleted; then, the fitness of each candidate size in the new candidate size set is recalculated, and the above process is repeated. In the process of iteratively updating multiple candidate size sets, if the iteration termination condition is currently met, multiple detection box sizes are screened out according to the above step 1304.

[0364] It should be noted that the above Figure 7 In the embodiment shown, when multiple clusters are obtained, multiple candidate size sets are generated based on the center sizes of the multiple clusters. In another embodiment, there is no need to execute the above step 1302. Instead, when multiple clusters are obtained, for each cluster, a hybrid filtering method can be used to determine the importance of each bounding box size in the cluster, and the bounding box sizes in the cluster whose importance is greater than a preset threshold are combined into a candidate size set, thereby obtaining multiple candidate size sets. Thereafter, the above steps 1303-1304 are executed based on the multiple candidate size sets.

[0365] Among them, the importance is used to indicate the contribution of the bounding box size to the resource search model. The preset threshold is an arbitrary value.

[0366] In an embodiment of the present application, the importance of each bounding box size in each cluster is determined, and then based on the importance of the bounding box size, the bounding box size in each cluster is screened, and the importance of the bounding box size is used as the probability of the bounding box size being selected, so as to screen out the bounding box sizes whose importance is greater than a preset threshold from the cluster to form a set of candidate sizes.

[0367] It should be noted that the above-mentioned multiple optional embodiments can be combined arbitrarily, and this application will not go into details.

[0368] In an embodiment of the present application, in the process of training the resource search model, the K-Medians clustering algorithm is combined with the niche genetic algorithm to optimize the detection frame size, and the resource search model is improved to recognize the curve used to represent the change trend of virtual resources. In addition, a feature extraction sub-model is constructed using the Darknet-53 backbone network and the multi-scale feature fusion structure. The feature extraction sub-model is used to automatically extract the key features in the change trend image, reduce the interference of human factors, improve the detection speed and accuracy of the model, and perform target detection and positioning through the resource search model. Through an intuitive interactive drawing tool, the user is allowed to draw the change trend curve, and then the system first normalizes the drawn curve to standardize the data range, and applies bilateral filtering to smooth the curve to reduce noise and enhance the curve characteristics. Based on the resource search model, virtual resources that match the curve input by the user can be searched, and the search results are displayed in the resource interface.

[0369] The uniqueness of the embodiment of the present application lies in that the user can interactively draw a curve to represent the changing trend of virtual resources, which can better reflect the user's graphical understanding of virtual resources, and combine various technical means such as deep learning and dynamic time warping algorithm to achieve more efficient and accurate virtual resource search, so as to help users efficiently identify and analyze the curve representing the changing trend of virtual resources, thereby ensuring the accuracy of virtual resource search.

[0370] In the embodiment of the present application, by providing an intuitive interactive drawing tool, the user is allowed to draw any curve for representing the change trend of virtual resources, so as to more accurately understand the needs of the user, thereby providing a more personalized virtual resource search, and improving the intuitiveness and flexibility of virtual resource search. The drawn curve is first normalized to standardize the data range, and bilateral filtering is applied to smooth the curve to reduce noise and enhance the curve characteristics, ensuring that the data input by different users has a unified scale in the resource search model, improving the consistency of the data and the comparability between different data sets, and effectively removing the random noise in the curve, while retaining important change trend information, improving the quality of the data, and providing a clearer signal for subsequent analysis. In addition, the K-Medians clustering algorithm and the microhabitat genetic algorithm are combined to optimize the detection frame size, reduce the influence of outliers, and improve the search ability of the resource search model. The dynamically adjusted mutation rate coefficient enables the microhabitat genetic algorithm to jump out of the local optimal solution during the optimization process, and improves the ability of global optimization. Through the grid traversal mapping method of the resource search model, the model prediction results are accurately mapped back to the original image, and the accurate positioning of the target is achieved. Through the automated intelligent search process, virtual resources that match the drawn curve can be quickly searched, significantly improving the search efficiency of virtual resources. In addition, a user feedback mechanism is introduced, allowing users to intuitively see the search results and optimize the resource search model based on feedback, thereby improving the user experience. The accuracy of virtual resource search is improved through clustering analysis and other technologies. By integrating modern data security technologies, the security of user data is ensured, while the reliability and security of the system are improved. The computing efficiency is improved through improved algorithms and network structure design, and a large number of resource search requests can be quickly processed to provide users with real-time search results. Through multi-scale feature fusion, the resource search model can better understand and detect changing trend images of different sizes, enhancing the generalization ability of the resource search model.

[0371] Fig.15 is a schematic diagram of the structure of a virtual resource search device provided in an embodiment of the present application, such as Fig.15 As shown, the device comprises:

[0372] Display module 1501, used to display the resource interface;

[0373] The display module 1501 is also used to respond to the information input operation in the resource interface, and display the input value change information in the resource interface, where the value change information is used to indicate the change trend of the resource value of the virtual resource to be searched over time;

[0374] The display module 1501 is also used to respond to the resource search operation for the value change information, and display the target virtual resource in the resource interface, wherein the similarity between the change trend of the target virtual resource and the change trend represented by the value change information is greater than a threshold.

[0375] In a possible implementation, the information input operation is a drawing operation, and the value change information is a curve drawn by the drawing operation; the display module 1501 is used to respond to the drawing operation in the resource interface and display the curve drawn by the drawing operation in the resource interface.

[0376] In another possible implementation, the numerical change information is a curve included in the first image; the display module 1501 is used to respond to the image upload operation in the resource interface, and display multiple candidate images in the resource interface; in response to the confirmation operation of the first image among the multiple candidate images, the first image is displayed in the resource interface.

[0377] In another possible implementation, the display module 1501 is further used to respond to a condition input operation in the resource interface, and display a search condition in the resource interface, where the search condition indicates a time range corresponding to a risk type or value change information of a virtual resource to be searched;

[0378] Display module 1501 is used to display the target virtual resource in the resource interface in response to the resource search operation for the numerical change information and the search conditions; wherein the target virtual resource belongs to the risk type, or the similarity between the change trend of the resource numerical value of the target virtual resource within the time range and the change trend represented by the numerical change information is greater than a threshold.

[0379] In another possible implementation, display module 1501 is used to respond to a resource search operation for numerical change information and display the numerical change information corresponding to the target virtual resource in the resource interface. The numerical change information corresponding to the target virtual resource is used to indicate the changing trend of the resource value of the target virtual resource over time.

[0380] In another possible implementation, the numerical change information is a first curve; the display module 1501 is used to respond to the resource search operation for the numerical change information, perform target detection on the second image and the change trend images of multiple candidate virtual resources, respectively, to obtain a first local image and multiple second local images, the first local image is a local image of the second image including the first curve, the second local image is a local image of the change trend image including the second curve, and the second curve is used to represent the change trend of the resource value of the candidate virtual resource over time; determine the matching degree between the first local image and each second local image, and determine the candidate virtual resource corresponding to the second local image with the largest matching degree as the target virtual resource; display the virtual resource in the resource interface.

[0381] In another possible implementation, the display module 1501 is used to respond to the resource search operation for the value change information, perform target detection on the second image and the multiple change trend images respectively through the resource search model, and obtain the first partial image and the multiple second partial images;

[0382] The display module 1501 is used to determine the matching degree between the first partial image and each second partial image through a resource search model, and determine the candidate virtual resource corresponding to the second partial image with the greatest matching degree as the target virtual resource.

[0383] It should be noted that the virtual resource search device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the virtual resource search device provided in the above embodiment and the virtual resource search method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0384] Fig.16 is a schematic diagram of the structure of a model training device provided in an embodiment of the present application, such as Fig.16 As shown, the device comprises:

[0385] A determination module 1601 is used to determine a plurality of detection frame sizes based on bounding frame sizes of curves in a plurality of first sample images, the curves being used to represent a trend of a change in a resource value of a virtual resource over time, and the bounding frame size being a size of a bounding frame of the curve in the first sample image;

[0386] An acquisition module 1602 is used to acquire a reference image and a second sample image, wherein a first curve in the reference image is used to represent a trend of a change in a resource value of a virtual resource to be searched over time, and a similarity between the first curve in the reference image and a second curve in the second sample image is greater than a threshold;

[0387] The detection module 1603 is used to perform target detection on the reference image and the second sample image respectively based on a plurality of detection frame sizes through a resource search model to obtain a first partial image and a second partial image, wherein the first partial image is a partial image including the first curve in the reference image, and the second partial image is a partial image including the second curve in the second sample image;

[0388] The determination module 1601 is further used to determine the matching degree between the first partial image and the second partial image through a resource search model;

[0389] The training module 1604 is used to train the resource search model based on the matching degree.

[0390] In a possible implementation, the determination module 1601 is used to cluster multiple bounding box sizes to obtain multiple cluster clusters, each cluster cluster includes at least one bounding box size; based on the central sizes of the multiple cluster clusters, generate multiple candidate size sets, each candidate size set includes at least one candidate size; for each candidate size set, based on the fitness of the candidate size in the candidate size set, update the candidate size set to obtain a new candidate size set, the fitness is used to indicate the degree of matching between the candidate size and the multiple bounding box sizes; determine the candidate size with the largest fitness in each new candidate size set as the detection box size.

[0391] In another possible implementation, the determination module 1601 is used to perform multiple random perturbations on the center size of each cluster to obtain multiple candidate sizes; and combine the multiple candidate sizes and the center size into a candidate size set.

[0392] In another possible implementation, the determination module 1601 is used to filter out a candidate size subset from the candidate size set based on the fitness of the candidate sizes in the candidate size set, where the candidate size subset includes at least one candidate size; generate a new candidate size based on the candidate sizes in the candidate size subset; and form a new candidate size set by combining the candidate sizes in the candidate size subset and the new candidate size.

[0393] In another possible implementation, the determination module 1601 is used to determine a first sum value, where the first sum value is the fitness sum value of the candidate sizes in the candidate size set; generate a random number from a numerical range consisting of 0 and the first sum value; start from any candidate size in the candidate size set, traverse the candidate sizes in the candidate size set; when the second sum value when traversing to the i-th candidate size is less than the random number and the third sum value when traversing to the i+1-th candidate size is greater than the random number, determine the i+1-th candidate size as the candidate size in the candidate size subset, the second sum value is the fitness sum value of the i candidate sizes that have been traversed, and the third sum value is the fitness sum value of the i+1 candidate sizes that have been traversed, where i is an integer greater than 0.

[0394] In another possible implementation, the determination module 1601 is used to randomly select two candidate sizes from the candidate size subset, swap part of the data at the same position in the two candidate sizes, and obtain two new candidate sizes; or, for any candidate size in the candidate size subset, reversely process part of the data in the candidate size to obtain a new candidate size.

[0395] In another possible implementation, the determination module 1601 is further configured to generate an exchange probability when two candidate sizes are selected, where the exchange probability indicates a possibility of exchanging partial data at the same position in the two candidate sizes;

[0396] The determination module 1601 is used to exchange part of the data at the same position in two candidate sizes to obtain two new candidate sizes when the exchange probability is greater than a first probability threshold.

[0397] In another possible implementation, the determination module 1601 is further configured to determine a mutation probability corresponding to the candidate size, where the mutation probability indicates a possibility of performing reverse processing on a portion of the data in the candidate size;

[0398] The determination module 1601 is used to perform reverse processing on part of the data in the candidate size to obtain a new candidate size when the mutation probability is greater than a second probability threshold.

[0399] In another possible implementation, the determination module 1601 is also used to determine a second probability threshold based on a third probability threshold, a current number of iterations, and a total number of iterations during multiple iterative updates of multiple candidate size sets, where the third probability threshold is a probability threshold when the multiple candidate size sets are updated for the first time.

[0400] It should be noted that the model training device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the model training device provided in the above embodiment belongs to the same concept as the model training method embodiment. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0401] An embodiment of the present application also provides a computer device, which includes a processor and a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the virtual resource search method or model training method of the above-mentioned embodiment.

[0402] Optionally, the computer device is provided as a terminal. Fig.17 The structure block diagram of a terminal 1700 provided by an exemplary embodiment of the present application is shown. The terminal 1700 includes: a processor 1701 and a memory 1702 .

[0403] The processor 1701 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1701 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 1701 may also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 1701 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1701 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0404] The memory 1702 may include one or more computer-readable storage media, which may be non-transitory. The memory 1702 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1702 is used to store at least one computer program, which is used to be executed by the processor 1701 to implement the virtual resource search method or model training method provided in the method embodiment of the present application.

[0405] In some embodiments, the terminal 1700 may further optionally include: a peripheral device interface 1703 and at least one peripheral device. The processor 1701, the memory 1702 and the peripheral device interface 1703 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 1703 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 1704, a display screen 1705, a camera assembly 1706, an audio circuit 1707 and a power supply 1708.

[0406] The peripheral device interface 1703 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 1701 and the memory 1702. In some embodiments, the processor 1701, the memory 1702, and the peripheral device interface 1703 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1701, the memory 1702, and the peripheral device interface 1703 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0407] The radio frequency circuit 1704 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1704 communicates with the communication network and other communication devices through electromagnetic signals. The radio frequency circuit 1704 converts the electrical signal into an electromagnetic signal for transmission, or converts the received electromagnetic signal into an electrical signal. Optionally, the radio frequency circuit 1704 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. The radio frequency circuit 1704 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes, but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1704 may also include circuits related to NFC (Near Field Communication), which is not limited in this application.

[0408] The display screen 1705 is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1705 is a touch display screen, the display screen 1705 also has the ability to collect touch signals on the surface or above the surface of the display screen 1705. The touch signal can be input to the processor 1701 as a control signal for processing. At this time, the display screen 1705 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 1705 can be one, set on the front panel of the terminal 1700; in other embodiments, the display screen 1705 can be at least two, respectively set on different surfaces of the terminal 1700 or in a folding design; in other embodiments, the display screen 1705 can be a flexible display screen, set on a curved surface or a folding surface of the terminal 1700. Even, the display screen 1705 can also be set to a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 1705 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0409] The camera assembly 1706 is used to capture images or videos. Optionally, the camera assembly 1706 includes a front camera and a rear camera. The front camera is arranged on the front panel of the terminal, and the rear camera is arranged on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize the panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 1706 may also include a flash. The flash can be a monochrome temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0410] The audio circuit 1707 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 1701 for processing, or input them into the radio frequency circuit 1704 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there may be multiple microphones, which are respectively arranged at different parts of the terminal 1700. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 1701 or the radio frequency circuit 1704 into sound waves. The speaker may be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio circuit 1707 may also include a headphone jack.

[0411] The power supply 1708 is used to power various components in the terminal 1700. The power supply 1708 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 1708 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery charged through a wired line, and a wireless rechargeable battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0412] Those skilled in the art will understand that Fig.17 The structure shown in the figure does not constitute a limitation on the terminal 1700, and the terminal 1700 may include more or less components than those shown in the figure, or combine certain components, or adopt a different component arrangement.

[0413] Optionally, the computer device is provided as a server. Fig.18 This is a schematic diagram of the structure of a server provided in an embodiment of the present application. The server 1800 may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 1801 and one or more memories 1802, wherein at least one computer program is stored in the memory 1802, and at least one computer program is loaded and executed by the processor 1801 to implement the methods provided in the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the server may also include other components for implementing device functions, which will not be described in detail here.

[0414] An embodiment of the present application also provides a computer-readable storage medium, which stores at least one computer program. The at least one computer program is loaded and executed by a processor to implement the operations performed by the virtual resource search method or model training method of the above-mentioned embodiment.

[0415] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the operations performed by the virtual resource search method or model training method of the above-mentioned embodiment.

[0416] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0417] The above description is only an optional embodiment of the embodiments of the present application and is not intended to limit the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A virtual resource search method, characterized in that: The method comprises: Display the resource interface; In response to the information input operation in the resource interface, the input value change information is displayed in the resource interface, and the value change information is used to indicate the change trend of the resource value of the virtual resource to be searched over time; In response to a resource search operation for the value change information, a target virtual resource is displayed in the resource interface, and a similarity between a change trend of the target virtual resource and a change trend represented by the value change information is greater than a threshold.

2. The method according to claim 1, characterized in that The information input operation is a drawing operation, and the value change information is a curve drawn by the drawing operation; In response to the information input operation in the resource interface, displaying the input value change information in the resource interface includes: In response to the drawing operation in the resource interface, a curve drawn by the drawing operation is displayed in the resource interface.

3. The method according to claim 1, characterized in that: The numerical value change information is a curve included in the first image; the response to the information input operation in the resource interface, displaying the input numerical value change information in the resource interface, includes: In response to an image upload operation in the resource interface, displaying a plurality of candidate images in the resource interface; In response to a confirmation operation on the first image among the plurality of candidate images, the first image is displayed in the resource interface.

4. The method according to claim 1, characterized in that: In response to the resource search operation for the value change information, before displaying the target virtual resource in the resource interface, the method further includes: In response to a condition input operation in the resource interface, a search condition is displayed in the resource interface, the search condition indicating the risk type of the virtual resource to be searched or the time range corresponding to the value change information; In response to the resource search operation for the value change information, displaying the target virtual resource in the resource interface includes: In response to a resource search operation for the value change information and the search condition, displaying the target virtual resource in the resource interface; The target virtual resource belongs to the risk type, or the similarity between the change trend of the resource value of the target virtual resource within the time range and the change trend represented by the value change information is greater than the threshold.

5. The method according to claim 1, characterized in that In response to the resource search operation for the value change information, displaying the target virtual resource in the resource interface includes: In response to a resource search operation for the value change information, the value change information corresponding to the target virtual resource is displayed in the resource interface, and the value change information corresponding to the target virtual resource is used to indicate a changing trend of the resource value of the target virtual resource over time.

6. The method according to claim 1, characterized in that The value change information is a first curve; and in response to the resource search operation for the value change information, displaying the target virtual resource in the resource interface includes: In response to a resource search operation for the value change information, target detection is performed on the second image and the change trend images of the plurality of candidate virtual resources respectively to obtain a first partial image and a plurality of second partial images, wherein the first partial image is a partial image of the second image including the first curve, and the second partial image is a partial image of the change trend image including the second curve, and the second curve is used to represent a change trend of the resource value of the candidate virtual resource over time; Determining a matching degree between the first partial image and each second partial image, and determining a candidate virtual resource corresponding to the second partial image with the greatest matching degree as the target virtual resource; In the resource interface, the virtual resources are displayed.

7. The method according to claim 6, characterized in that In response to the resource search operation for the value change information, target detection is performed on the second image and the change trend images of multiple candidate virtual resources to obtain a first partial image and multiple second partial images, including: In response to a resource search operation for the value change information, target detection is performed on the second image and the plurality of change trend images respectively through a resource search model to obtain the first partial image and the plurality of second partial images; The determining of the matching degree between the first partial image and each second partial image, and determining the candidate virtual resource corresponding to the second partial image with the greatest matching degree as the target virtual resource includes: The matching degree between the first partial image and each second partial image is determined by the resource search model, and the candidate virtual resource corresponding to the second partial image with the greatest matching degree is determined as the target virtual resource.

8. A model training method, characterized in that: The method comprises: Determining a plurality of detection frame sizes based on bounding frame sizes of curves in a plurality of first sample images, wherein the curves are used to represent a trend of a change in a resource value of a virtual resource over time, and the bounding frame sizes are sizes of bounding frames of the curves in the first sample images; Acquire a reference image and a second sample image, wherein a first curve in the reference image is used to represent a change trend of a resource value of a virtual resource to be searched over time, and a similarity between the first curve in the reference image and a second curve in the second sample image is greater than a threshold; By means of a resource search model, based on the multiple detection frame sizes, target detection is performed on the reference image and the second sample image respectively to obtain a first local image and a second local image, wherein the first local image is a local image of the reference image including the first curve, and the second local image is a local image of the second sample image including the second curve; Determining the matching degree between the first partial image and the second partial image by using the resource search model; Based on the matching degree, the resource search model is trained.

9. The method according to claim 8, characterized in that The determining of a plurality of detection frame sizes based on the bounding frame sizes of the curves in the plurality of first sample images comprises: Clustering the multiple bounding box sizes to obtain multiple clusters, each cluster including at least one bounding box size; Based on the center sizes of the multiple clusters, generating multiple candidate size sets, each candidate size set including at least one candidate size; For each candidate size set, based on the fitness of the candidate sizes in the candidate size set, the candidate size set is updated to obtain a new candidate size set, wherein the fitness is used to indicate the degree of matching between the candidate size and the multiple bounding box sizes; The candidate size with the greatest fitness in each new candidate size set is determined as the detection frame size.

10. The method according to claim 9, characterized in that The generating a plurality of candidate size sets based on the center sizes of the plurality of clusters comprises: For each cluster center size, performing multiple random perturbations on the center size to obtain multiple candidate sizes; The plurality of candidate sizes and the center size are combined into the candidate size set.

11. The method according to claim 9, characterized in that The updating of the candidate size set based on the fitness of the candidate size in the candidate size set to obtain a new candidate size set includes: Based on the fitness of the candidate sizes in the candidate size set, a candidate size subset is screened out from the candidate size set, where the candidate size subset includes at least one candidate size; generating a new candidate size based on the candidate sizes in the candidate size subset; The candidate sizes in the candidate size subset and the new candidate size are combined into the new candidate size set.

12. The method according to claim 11, characterized in that The step of selecting a candidate size subset from the candidate size set based on the fitness of the candidate sizes in the candidate size set includes: Determine a first sum value, where the first sum value is a fitness sum value of a candidate size in the candidate size set; Generate a random number from a numerical range consisting of 0 and the first sum value; Starting from any candidate size in the candidate size set, traversing the candidate sizes in the candidate size set; When the second sum value when traversing to the i-th candidate size is less than the random number and the third sum value when traversing to the i+1-th candidate size is greater than the random number, the i+1-th candidate size is determined as the candidate size in the candidate size subset, the second sum value is the fitness sum value of the traversed i candidate sizes, the third sum value is the fitness sum value of the traversed i+1 candidate sizes, and i is an integer greater than 0.

13. The method according to claim 11, characterized in that The generating a new candidate size based on the candidate size in the candidate size subset includes: Randomly select two candidate sizes from the candidate size subset, and exchange part of the data at the same position in the two candidate sizes to obtain two new candidate sizes; or For any candidate size in the candidate size subset, a portion of data in the candidate size is processed in reverse order to obtain a new candidate size.

14. A virtual resource search device, characterized in that: The device comprises: Display module, used to display the resource interface; The display module is further used to respond to the information input operation in the resource interface, and display the input value change information in the resource interface, wherein the value change information is used to indicate the change trend of the resource value of the virtual resource to be searched over time; The display module is further used to respond to a resource search operation for the numerical change information, and display a target virtual resource in the resource interface, wherein the similarity between the change trend of the target virtual resource and the change trend represented by the numerical change information is greater than a threshold.

15. A model training device, characterized in that: The device comprises: a determination module, configured to determine a plurality of detection frame sizes based on bounding frame sizes of curves in a plurality of first sample images, wherein the curves are used to represent a trend of a change in a resource value of a virtual resource over time, and the bounding frame sizes are sizes of bounding frames of the curves in the first sample images; an acquisition module, used to acquire a reference image and a second sample image, wherein a first curve in the reference image is used to represent a change trend of a resource value of a virtual resource to be searched over time, and a similarity between the first curve in the reference image and a second curve in the second sample image is greater than a threshold; a detection module, configured to perform target detection on the reference image and the second sample image respectively based on the multiple detection frame sizes through a resource search model to obtain a first local image and a second local image, wherein the first local image is a local image of the reference image including the first curve, and the second local image is a local image of the second sample image including the second curve; The determination module is further configured to determine a matching degree between the first partial image and the second partial image through the resource search model; A training module is used to train the resource search model based on the matching degree.

16. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the virtual resource search method as described in any one of claims 1 to 7; or, to implement the operations performed by the model training method as described in any one of claims 8 to 13.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor to implement the operations performed by the virtual resource search method as described in any one of claims 1 to 7; or to implement the operations performed by the model training method as described in any one of claims 8 to 13.

18. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the operations performed by the virtual resource search method as described in any one of claims 1 to 7; or, it implements the operations performed by the model training method as described in any one of claims 8 to 13.