Target image recognition method and its device, equipment, medium and product

Abstract

The present application discloses a target image recognition method and its device, equipment, medium, and product, the method comprising: obtaining an object picture to be identified as to whether it contains a target image; calling a first composite recognition model to perform image classification recognition on the object picture, obtaining a binary classification judgment result representing whether the object picture contains the target image, the first composite recognition model is configured so that its positive judgment result has a higher credibility than its negative judgment result; calling a second composite recognition model to perform image classification recognition on the object picture, obtaining a binary classification judgment result representing whether the object picture contains the target image, the second composite recognition model is configured so that its negative judgment result has a higher credibility than its positive judgment result; when the binary classification judgment results are the same, the corresponding judgment result is used as the recognition result of the object picture. The present application can obtain a more credible target image recognition result.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to a target image recognition method and its corresponding device, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] Image recognition using artificial neural network models has become a mainstream technology. For application scenarios such as e-commerce platforms, a large number of item images or other images are generated every day. For various application needs, including automatic classification of goods, compliance detection of goods, target product search, etc., it is generally necessary to detect and identify the target objects in the item images uploaded by users, and then perform further processing according to different target objects and needs.

[0003] In the prior art, there are already numerous methods that can achieve target image recognition. It is well known that neural network models rely on massive training data for training to acquire semantic understanding capabilities, but this ability is highly probabilistic, that is, any neural network model, even if it is trained with sufficient training samples, will ultimately not be able to achieve 100% accuracy in identifying relevant target images, and can only identify target icons corresponding to the training samples. This is a natural feature of artificial neural networks. Therefore, it is generally sufficient to train the artificial neural network until the loss function converges and the correct recognition result reaches a certain probability before it can be put into use.

[0004] For application scenarios such as e-commerce platforms, if the pictures of items uploaded by merchants include items that are not for sale, such as knives and swords, the existing neural network models sometimes find it difficult to accurately identify these products from the pictures due to their small size, reflectiveness, and variety of forms. Actual measurements show a high rate of false positives and missed positives, which can cause great trouble for e-commerce platforms. It is even more unrealistic to rely entirely on manual screening.

[0005] Therefore, how to ensure accurate recognition of target images from object pictures and make the recognition results more reliable has become a technical problem that needs to be solved in this field. Summary of the invention

[0006] The primary purpose of the present application is to solve at least one of the above problems and to provide a target image recognition method and its corresponding device, computer equipment, computer-readable storage medium, and computer program product.

[0007] In order to meet the various objectives of this application, this application adopts the following technical solutions:

[0008] A target image recognition method provided for one of the purposes of this application comprises the following steps:

[0009] Get the image of the object to be identified to see if it contains the target image;

[0010] Calling the first composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image, wherein the first composite recognition model is configured such that its positive judgment result has a higher credibility than its negative judgment result;

[0011] Calling the second composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image, wherein the second composite recognition model is configured such that its negative judgment result has a higher credibility than its positive judgment result;

[0012] When the two-classification judgment results are the same, the corresponding judgment results are used as the recognition results of the object image.

[0013] In a further embodiment, the first compound recognition model and the second compound recognition model both include multiple sub-models that use different neural network structures to identify whether the object image contains the target image and obtain corresponding binary classification judgment results; in the first compound recognition model, when all sub-models obtain positive judgment results, they are triggered to output positive judgment results, and when any sub-model obtains negative judgment results, they are triggered to output negative judgment results; in the second compound recognition model, when all sub-models obtain negative judgment results, they are triggered to output negative judgment results, and when any sub-model obtains positive judgment results, it is triggered to output positive judgment results.

[0014] In a specific embodiment, the sub-models of the first compound recognition model and the second compound recognition model include a block recognition sub-model that performs image classification and recognition on a local image of the object picture to obtain a corresponding binary classification judgment result, and at least one full-image recognition sub-model that performs image classification and recognition on the full-frame image of the object picture to obtain a corresponding binary classification judgment result.

[0015] In a preferred embodiment, in the first composite recognition model or the second composite recognition model, each sub-model is implemented based on a different neural network model; and, in the first composite recognition model and the second composite recognition model, the sub-models used by each other are at least partially different.

[0016] In a further embodiment, calling the first composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image includes the following steps:

[0017] Transmitting the object image to the first preliminary screening full-image recognition sub-model of the first composite recognition model;

[0018] The first preliminary screening full-image recognition sub-model recognizes the full-image of the item picture and determines a binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the operation of the first block recognition sub-model is triggered;

[0019] The first block recognition sub-model recognizes the local image of the item picture and determines a binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the operation of the first fine screening full-image recognition sub-model is triggered;

[0020] The first fine screening full-image recognition sub-model recognizes the full image of the object picture and determines a binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the first composite recognition model is triggered to output a positive judgment result.

[0021] In a further embodiment, calling the second composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image includes the following steps:

[0022] Transmitting the object image to the second preliminary screening full-image recognition sub-model of the second composite recognition model;

[0023] The second preliminary screening full-image recognition sub-model recognizes the full-image of the item picture and determines a binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second block recognition sub-model is triggered to run;

[0024] The second block recognition sub-model recognizes the local image of the item picture and determines the binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second fine screening full-image recognition sub-model is triggered to run;

[0025] The second fine screening full-image recognition sub-model recognizes the full image of the object picture and determines a binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second composite recognition model is triggered to output a negative judgment result.

[0026] A target image recognition device provided to meet one of the purposes of the present application includes: an image acquisition module, a first recognition module, a second recognition module, and a result fusion module, wherein the image acquisition module is used to acquire an object image to be identified as to whether it contains a target image; the first recognition module is used to call a first composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result representing whether the object image contains the target image, and the first composite recognition model is configured so that its positive judgment result has a higher credibility than its negative judgment result; the second recognition module is used to call a second composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result representing whether the object image contains the target image, and the second composite recognition model is configured so that its negative judgment result has a higher credibility than its positive judgment result; the result fusion module is used to use the corresponding judgment result as the recognition result of the object image when the binary classification judgment results are the same.

[0027] In a further embodiment, the first compound recognition model and the second compound recognition model both include multiple sub-models that use different neural network structures to identify whether the object image contains the target image and obtain corresponding binary classification judgment results; in the first compound recognition model, when all sub-models obtain positive judgment results, they are triggered to output positive judgment results, and when any sub-model obtains negative judgment results, they are triggered to output negative judgment results; in the second compound recognition model, when all sub-models obtain negative judgment results, they are triggered to output negative judgment results, and when any sub-model obtains positive judgment results, it is triggered to output positive judgment results.

[0028] In a specific embodiment, the sub-models of the first compound recognition model and the second compound recognition model include a block recognition sub-model that performs image classification and recognition on a local image of the object picture to obtain a corresponding binary classification judgment result, and at least one full-image recognition sub-model that performs image classification and recognition on the full-frame image of the object picture to obtain a corresponding binary classification judgment result.

[0029] In a preferred embodiment, in the first composite recognition model or the second composite recognition model, each sub-model is implemented based on a different neural network model; and, in the first composite recognition model and the second composite recognition model, the sub-models used by each other are at least partially different.

[0030] In a further embodiment, the first recognition module includes: a first parameter transmission starter module, which is used to pass the object picture to the first preliminary screening full-image recognition sub-model of the first composite recognition model; a first preliminary screening recognition sub-module, which is used to call the first preliminary screening full-image recognition sub-model to recognize the full-image of the object picture and determine the binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the operation of the first block recognition sub-model is triggered; a first block recognition sub-module, which is used to call the first block recognition sub-model to recognize the local image of the object picture and determine the binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the operation of the first fine screening full-image recognition sub-model is triggered; the first fine screening recognition sub-module is used to call the first fine screening full-image recognition sub-model to recognize the full-image of the object picture and determine the binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the first composite recognition model is triggered to output a positive judgment result.

[0031] In a further embodiment, the second composite recognition model is called to perform image classification and recognition on the object picture to obtain a binary classification judgment result indicating whether the object picture contains the target image, including the following steps: a second parameter transmission initiator module is used to transmit the object picture to the second preliminary screening full-image recognition sub-model of the second composite recognition model; a second preliminary screening recognition sub-module is used to call the second preliminary screening full-image recognition sub-model to perform recognition based on the full-image of the object picture to determine the binary classification judgment result. When the binary classification judgment result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the operation of the second block recognition sub-model is triggered. ; The second block recognition submodule is used to call the second block recognition submodel to perform recognition based on the local image of the object picture, and judge the binary classification judgment result. When the binary classification judgment result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second fine screening full-image recognition submodel is triggered to run; the second fine screening recognition submodule is used to call the second fine screening full-image recognition submodel to perform recognition based on the full image of the object picture, and judge the binary classification judgment result. When the binary classification judgment result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second composite recognition model is triggered to output a negative judgment result.

[0032] A computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the target image recognition method described in the present application.

[0033] A computer-readable storage medium is provided to meet another purpose of the present application, which stores a computer program implemented according to the target image recognition method in the form of computer-readable instructions. When the computer program is called and executed by a computer, the steps included in the method are executed.

[0034] A computer program product provided to meet another purpose of the present application includes a computer program / instruction, which, when executed by a processor, implements the steps of the method described in any embodiment of the present application.

[0035] Compared with the prior art, the advantages of this application are as follows:

[0036] The present application utilizes two image recognition models to respectively identify the target image of the object picture, wherein the first composite recognition model has high credibility for the positive judgment result obtained by the recognition judgment, and the second composite recognition model has high credibility for the negative judgment result obtained by the recognition judgment. When the judgment results of the two recognition models are the same, that is, both positive and negative, the corresponding recognition judgment result is confirmed, otherwise the result is not confirmed and can be left for manual recognition. When both recognition models obtain the same judgment result, although the credibility of the judgment result of one of the recognition models is low, since the judgment result of the other recognition model is highly credible, the two identical results combined are more credible than a single result, and therefore the same judgment result is trusted. When the two models respectively identify different judgment results, that is, one positive and one negative, it means that either the two recognition models both output untrustworthy judgment results and thus doubt the image content of the object picture, or the two recognition models both output credible different results and thus conflict, therefore, they can be temporarily excluded, such as left for manual recognition processing. It can be seen that the present application combines the XOR logic to realize a more credible network architecture. The judgment results obtained by identifying the target image of the object picture are definitely more credible than relying on a single recognition model, and can be directly used in the e-commerce platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0038] Figure 1 A flowchart of a typical embodiment of the target image recognition method of the present application;

[0039] Figure 2 This is a schematic diagram of the workflow of the first composite recognition model in the embodiment of the present application;

[0040] Figure 3 This is a schematic diagram of the network structure of the first composite recognition model in the embodiment of the present application;

[0041] Figure 4This is a schematic diagram of the workflow of the second composite recognition model in the embodiment of the present application;

[0042] Figure 5 This is a schematic diagram of the network structure of the second composite recognition model in the embodiment of the present application;

[0043] Figure 6 This is a principle block diagram of the target image recognition device of the present application;

[0044] Figure 7 A schematic diagram of the structure of a computer device used in this application. DETAILED DESCRIPTION

[0045] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0046] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0047] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0048] It will be understood by those skilled in the art that the "client", "terminal" and "terminal device" used herein include both devices with wireless signal receivers, which are devices with only wireless signal receivers without transmission capabilities, and devices with receiving and transmitting hardware, which are devices with receiving and transmitting hardware capable of two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices such as personal computers, tablet computers, which have single-line displays or multi-line displays or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service, personal communication system), which can combine voice, data processing, fax and / or data communication capabilities; PDA (Personal Digital Assistant, personal digital assistant), which may include a radio frequency receiver, pager, Internet / intranet access, web browser, notepad, calendar and / or GPS (Global Positioning System, global positioning system) receiver; conventional laptop and / or palmtop computers or other devices, which have and / or include a conventional laptop and / or palmtop computer or other device with and / or including a radio frequency receiver. The "client", "terminal" and "terminal device" used herein may be portable, transportable, installed in a vehicle (air, sea and / or land), or suitable for and / or configured to run locally, and / or in a distributed form, at any other location on the earth and / or in space. The "client", "terminal" and "terminal device" used herein may also be a communication terminal, an Internet terminal, a music / video playing terminal, for example, a PDA, a MID (Mobile Internet Device) and / or a mobile phone with a music / video playing function, or a smart TV, a set-top box and other devices.

[0049] The hardware referred to by the names such as "server", "client", and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer. It is a hardware device with the necessary components revealed by the von Neumann principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit calls the program stored in the external memory into the internal memory for execution, executes the instructions in the program, and interacts with the input and output devices to complete specific functions.

[0050] It should be pointed out that the concept of "server" referred to in this application can also be extended to the case of server clusters. According to the network deployment principle understood by those skilled in the art, the servers should be logically divided. In physical space, these servers can be independent of each other but can be called through interfaces, or integrated into a physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility, and should not use it to restrict the implementation of the network deployment method of this application.

[0051] Unless expressly specified, one or more technical features of the present application can be deployed on a server for implementation and accessed by a client through a remote call to obtain an online service interface provided by the server, or can be directly deployed and run on a client for access.

[0052] The neural network models referenced or may be referenced in this application, unless expressly specified, can be deployed on a remote server and remotely called on the client, or can be deployed and directly called on a client with sufficient device capabilities. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.

[0053] Unless explicitly specified, the various data involved in this application can be stored remotely on a server or on a local terminal device, as long as it is suitable for being called by the technical solution of this application.

[0054] Those skilled in the art should be aware that, although the various methods of the present application are described based on the same concept and thus present commonality to each other, unless otherwise specified, these methods can be independently executed. Similarly, for each embodiment disclosed in the present application, they are all proposed based on the same inventive concept, therefore, concepts with the same expression, and concepts that are appropriately changed for convenience despite different expressions, should be understood as equivalent.

[0055] Unless the mutually exclusive relationship between the embodiments to be disclosed in this application is explicitly stated, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct a new embodiment, as long as such combination does not deviate from the creative spirit of this application and can meet the needs of the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0056] A target image recognition method of the present application can be programmed as a computer program product and deployed in a client or server for execution. The method can be executed by accessing an interface opened after the computer program product is running and performing human-computer interaction with the process of the computer program product through a graphical user interface.

[0057] See also Figure 1 In a typical embodiment, the target image recognition method of the present application comprises the following steps:

[0058] Step S1100: Obtain an image of an object to be identified to determine whether it contains a target image:

[0059] The item picture generally refers to a picture containing an item image and used to show the appearance and shape of the item. In an exemplary application scenario used to assist in the present invention, the item picture may be a product picture on an e-commerce platform. The items shown in the product picture are generally sellable products. The implementation of the present invention is to identify non-sellable products from the product picture, such as knives, swords, etc., and therefore it is necessary to identify the corresponding target image in the product picture.

[0060] In this application scenario, the target image is usually determined by the customized items of the e-commerce platform, that is, non-saleable goods. Those skilled in the art can understand that by pre-training the image recognition model, it can learn the ability to recognize the target image from the item picture. Therefore, a large number of product pictures of various appearance shapes containing the non-saleable goods can be used to train the image recognition model, so that such an image recognition model can learn the ability to recognize non-saleable goods from the product pictures. In other application scenarios, an image recognition model for recognizing the target image from the item picture can also be prepared according to this principle.

[0061] Since image recognition models are relatively mature technologies, there are a variety of image recognition models available in the prior art that can be used for reference. They are generally based on CNN convolutional neural networks, including but not limited to ViT, Resnet, HTC, etc. Different image recognition models based on neural networks may have different convolutional structures, resulting in different recognition methods. However, in terms of function and purpose, as long as the recognition model has been pre-trained to a convergence state for the target object and put into production, it can recognize the target image from the object image. Of course, the recognition capabilities of various models present different recognition accuracy rates depending on the model structure and training samples, which is reflected in the different credibility of the recognition results of each recognition model.

[0062] In the application scenario of the e-commerce platform, if it is necessary to obtain item images, one implementation method is to receive input from users of the e-commerce platform, especially input from warehouse management users of merchant instances when configuring product information, and use product images in the product information as the item images. Of course, it is also possible to identify images posted by consumers in the comment area and use the images as item images. In another method, the server of the e-commerce platform can batch process the product images in the e-commerce platform database in the background, and use these product images as the item images to identify the target image.

[0063] Step S1200: calling the first composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image. The first composite recognition model is configured such that its positive judgment result has a higher credibility than its negative judgment result:

[0064] The first composite recognition model is essentially a network structure formed by connecting two or more image recognition sub-models that realize the target image recognition capability. The image recognition sub-model is trained in advance with training samples containing the target image, and acquires the capability to recognize the target image from the object image. The specific training method of each sub-model and the data set required for training are flexibly determined by those skilled in the art. Relying on the existing recognition capabilities of each image recognition sub-module, the first composite recognition model acquires the capability to recognize the target image from the object image accordingly.

[0065] The sub-models adopted by the first composite recognition model are all for obtaining the judgment result of whether the object picture contains the target image. Therefore, these sub-models can use binary classifiers to realize classification judgment, so as to directly obtain the corresponding binary classification judgment result, that is, to characterize whether the object picture contains or does not contain the target image. When it is specifically executed, it will be judged by the probability output by the binary classifier, that is, when the classification corresponding to the target image is included, the target image is included in the object picture; when the classification corresponding to the target image is not included, the target image is not included in the object picture. It can be understood that the binary classification judgment result is binarized, or a positive judgment result is determined, that is, the object picture contains the target image; or a negative judgment result is determined, that is, the object picture does not contain the target image. Of course, the present application does not exclude that individual sub-models use multiple classifiers to judge the classification results and output multiple judgment results. In this regard, those skilled in the art can understand that as long as the binary classification judgment result can eventually lead to the output of a binary classification judgment result, it can still be used in the present application.

[0066] In the first composite recognition model, multiple sub-models are connected in series. Specifically, after one of the sub-models executes the previous judgment on whether the object image contains the target image, a corresponding binary classification judgment result is obtained. If the judgment result indicates that the current object image does not contain the target image, that is, a negative judgment result is obtained, it can be directly determined that the target image does not exist in the object image, and thus it is also used as the binary classification judgment result output by the first composite recognition model for recognition, and a negative judgment result is output. If the binary classification judgment result of the sub-model executed first indicates that the current object image contains the target image, that is, a positive judgment result is obtained, then the binary classification judgment result is not directly output to the first composite recognition model, but the next sub-model is continued to be called to identify the target image of the object image. Similar to the sub-model executed first, after the sub-model executed later obtains the binary classification judgment result by identifying the target image of the object image, if the binary classification judgment result is a negative judgment result, the negative judgment result is directly output to the first composite recognition model. If the binary classification judgment result is a positive judgment result, the next sub-model can be continued to be called to continue to identify the object image. The process is similar to that of the current sub-model. If the current sub-model is the last-level sub-model in the first composite recognition model and a positive classification result is obtained, then the positive classification result is output as the binary classification judgment result of the first composite recognition model, which means that the object image contains the target image after the first composite recognition model has identified and judged it.

[0067] At this point, it can be understood that in the first composite recognition model, when any one of the sub-models obtains a negative judgment result after identifying the object image, the judgment process of the first composite recognition model is directly terminated, and a binary classification judgment result indicating that the object image does not contain the target image is directly output. However, when the previously executed sub-model obtains a positive judgment result after identifying the object image, it is necessary to be reconfirmed by the next sub-model, and so on, layer by layer, until all sub-models confirm that the recognition of the object image has obtained a positive judgment result, and then the positive judgment result is output for the recognition process of the first composite recognition model.

[0068] It can be seen that once the first composite recognition model outputs a positive judgment result, this result has high credibility and has been confirmed by multiple sub-models for many times. On the contrary, if the first composite recognition model outputs a negative judgment result, this result may be output after the individual sub-models recognize it, which is relatively hasty and has low credibility. In other words, among the two-class judgment results output by the first composite recognition model, if a positive judgment result is output, this result has high credibility, and if a negative judgment result is output, this result has low credibility. The credibility of the positive judgment result identified by the first composite recognition model is higher than the credibility of the negative judgment result output by it. In practice, by matching different pre-trained sub-models, the probability that the target image is actually contained in the object image can be controlled to be as high as 99.9% when the first composite recognition model outputs a positive judgment result. However, when the first composite recognition model outputs a negative judgment result, its statistical probability is not as optimistic as the former, so it is relatively untrustworthy.

[0069] The various sub-models adopted by the first composite recognition model generally adopt image recognition models that are different in recognition methods and / or network structures, so as to achieve the integration of multiple different technical capabilities and improve the recognition ability of the first composite recognition model by integrating different technical capabilities. For example, some recommended neural network image recognition models, including HTC (Hybrid Task Cascade for Instance Segmentation), mask-rcnn, Resnet, ViT and their upgraded and evolved versions, are all mature image recognition models and can be used as image recognition models of this application.

[0070] The image recognition model adopted by the sub-model can be for the full-frame image of the object picture, or for the partial image of the object picture, depending on the representation learning of the image recognition model in the training stage. The recognition of different formats of object pictures also determines the difference in the recognition methods adopted by the image recognition model. In this application, according to the different formats of the object pictures targeted by the image recognition model, it can be divided into a category for whole-image recognition for the full-frame portrait, called the full-image recognition sub-model; and another category for block recognition for the partial image, called the block recognition sub-model. The former is better at extracting features through the contour of the target image in the object picture to achieve recognition, and the latter is better at extracting features through the details of the target image in the object picture to achieve recognition. Taking these two categories into consideration can enrich the multi-view and multi-level image recognition capabilities of the first composite recognition model. Of course, in alternative embodiments, multiple sub-models of each type may be selected and used together. For example, for the full-image recognition sub-model, a primary screening full-image recognition sub-model may be used to perform an initial full-image range recognition of the target image on the object image to achieve primary screening; and another fine screening full-image recognition sub-model may be used to perform a secondary full-image range recognition of the target image on the object image to achieve fine screening.

[0071] At this point, it should be understood that after the first composite recognition model of the present application performs image classification and recognition on the object picture, a binary classification judgment result can be obtained. If the binary classification judgment result is a positive judgment result, it is reliable; if it is a negative judgment result, it is relatively unreliable.

[0072] Step S1300: calling the second composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image. The second composite recognition model is configured such that its negative judgment result has a higher credibility than its positive judgment result:

[0073] The second composite recognition model is essentially a network structure formed by connecting two or more image recognition sub-models that realize the target image recognition capability. The image recognition sub-model is trained in advance with training samples containing the target image, and acquires the capability to recognize the target image from the object picture. The specific training method of each sub-model and the data set required for training are flexibly determined by those skilled in the art. Relying on the existing recognition capabilities of each image recognition sub-module, the second composite recognition model acquires the capability to recognize the target image from the object picture accordingly.

[0074] The sub-models adopted by the second composite recognition model are all for obtaining the judgment result of whether the object picture contains the target image. Therefore, these sub-models can use binary classifiers to realize classification judgment, so as to directly obtain the corresponding binary classification judgment result, that is, to characterize whether the object picture contains or does not contain the target image. When it is specifically executed, it will be judged by the probability output by the binary classifier, that is, when the classification corresponding to the target image is included, a higher classification probability is obtained, that is, the object picture contains the target image; when the classification corresponding to the target image is not included, a higher classification probability is obtained, that is, the object picture does not contain the target image. It can be understood that the binary classification judgment result is binarized, or a positive judgment result is determined, that is, the object picture contains the target image; or a negative judgment result is determined, that is, the object picture does not contain the target image. Of course, the present application does not exclude that individual sub-models use multiple classifiers to judge the classification results and output multiple judgment results. In this regard, those skilled in the art can understand that as long as the binary classification judgment result can eventually lead to the output of a binary classification, it can still be used in the present application.

[0075] In the second composite recognition model, multiple sub-models are connected in series. Specifically, after one of the sub-models executes the previous judgment on whether the object image contains the target image, a corresponding binary classification judgment result is obtained. If the judgment result indicates that the current object image contains the target image, that is, a positive judgment result is obtained, it can be directly determined that the target image exists in the object image, and thus the binary classification judgment result output by the second composite recognition model is also output as a positive judgment result. If the binary classification judgment result of the sub-model executed first indicates that the current object image does not contain the target image, that is, a negative judgment result is obtained, then the binary classification judgment result is not directly output to the second composite recognition model, but the next sub-model is continued to be called to recognize the target image of the object image. Similar to the sub-model executed first, after the sub-model executed later obtains the binary classification judgment result by recognizing the target image of the object image, if the binary classification judgment result is a positive judgment result, the positive judgment result is directly output to the second composite recognition model. If the binary classification judgment result is a negative judgment result, the next sub-model can be continued to be called to continue recognizing the object image. The process is similar to that of the current sub-model. If the current sub-model is the last sub-model in the second composite recognition model and a negative classification result is obtained, then the negative classification result is output as the binary classification judgment result of the second composite recognition model, which means that the object image does not contain the target image after the second composite recognition model recognizes and judges it.

[0076] At this point, it can be understood that in the second composite recognition model, when any one of the sub-models obtains a positive judgment result after identifying the object image, the judgment process of the second composite recognition model is directly ended, and a binary classification judgment result indicating that the object image contains the target image is directly output. However, when a negative judgment result is obtained after the previously executed sub-model identifies the object image, it is necessary to be reconfirmed by the next sub-model, and so on, layer by layer, until all sub-models confirm that the recognition of the object image has obtained a negative judgment result, and then a negative judgment result is output for the recognition process of the second composite recognition model.

[0077] It can be seen that once the second composite recognition model outputs a negative judgment result, this result is highly credible and has been confirmed multiple times by multiple sub-models. On the contrary, if the second composite recognition model outputs a positive judgment result, this result may be output after the individual sub-models recognize it, which is relatively hasty and has low credibility. In other words, in the binary classification judgment result output by the second composite recognition model, if a negative judgment result is output, this result is highly credible, and if a positive judgment result is output, this result is low credible. The credibility of the second composite recognition model identifying the negative judgment result is higher than the credibility of its outputting the positive judgment result. In practice, by matching different pre-trained sub-models, the probability that the target image is not actually contained in the object image can be controlled to be as high as 99.9% when the second composite recognition model outputs a negative judgment result, but when the second composite recognition model outputs a positive judgment result, its statistical probability is not as optimistic as the former, so it is relatively untrustworthy.

[0078] The various sub-models adopted by the second composite recognition model generally adopt image recognition models that are different in recognition methods and / or network structures, so as to achieve the integration of multiple different technical capabilities and improve the recognition ability of the second composite recognition model by integrating different technical capabilities. For example, some recommended neural network image recognition models, including HTC (Hybrid Task Cascade for Instance Segmentation), mask-rcnn, Resnet, ViT and its upgraded and evolved versions, are all mature image recognition models and can be used as image recognition models of this application.

[0079] The image recognition model adopted by the sub-model can be for the full-frame image of the object picture, or for the partial image of the object picture, depending on the representation learning of the image recognition model in the training stage. The recognition of different formats of object pictures also determines the difference in the recognition methods adopted by the image recognition model. In this application, according to the different formats of the object pictures targeted by the image recognition model, it can be divided into a category for whole-image recognition for the full-frame portrait, called the full-image recognition sub-model; and another category for block recognition for the partial image, called the block recognition sub-model. The former is better at extracting features through the outline of the target image in the object picture to achieve recognition, and the latter is better at extracting features through the details of the target image in the object picture to achieve recognition. Taking these two categories into consideration can enrich the multi-view and multi-level image recognition capabilities of the second composite recognition model. Of course, in alternative embodiments, multiple sub-models of each type may be selected and used together. For example, for the full-image recognition sub-model, a primary screening full-image recognition sub-model may be used to perform an initial full-image range recognition of the target image on the object image to achieve primary screening; and another fine screening full-image recognition sub-model may be used to perform a secondary full-image range recognition of the target image on the object image to achieve fine screening.

[0080] At this point, it should be understood that after the second composite recognition model of the present application performs image classification and recognition on the object picture, a binary classification judgment result can be obtained. If the binary classification judgment result is a negative judgment result, it is reliable; if it is a positive judgment result, it is relatively unreliable.

[0081] Step S1400: When the two-classification judgment results are the same, the corresponding judgment results are used as the recognition results of the object image:

[0082] The first composite recognition model and the second composite recognition model both output the binary classification judgment results obtained after they have recognized the item image. Considering that the credibility of the different binary classification judgment results output by each composite recognition model is different, the two binary classification judgment results can be jointly judged here. First, several situations where the two composite recognition models output different classification judgments are combined are examined:

[0083] First, the first compound recognition model outputs a positive judgment result, and the second compound recognition model outputs a negative judgment result: In this case, since the positive judgment result output by the first compound recognition model is highly credible, the negative judgment result output by the second compound recognition model is also highly credible. Therefore, they are both yes and no, and conflict with each other, so such results should be excluded.

[0084] Second, the first composite recognition model outputs a negative judgment result, and the second composite recognition model outputs a positive judgment result: In this case, since the negative judgment result with low credibility output by the first composite recognition model is unreliable, and the positive judgment result with low credibility output by the second composite recognition model is also unreliable, the right and wrong are unclear and conflicting, so such results should be excluded.

[0085] Third, the first composite recognition model outputs a positive judgment result, and the second composite recognition model outputs a positive judgment result: In this case, since the positive judgment result output by the first composite recognition model is highly credible, although the positive judgment result output by the second composite recognition model is of low credibility, the recognition of the positive judgment result by the first composite model has been confirmed layer by layer, therefore, the judgment result of the first composite recognition model should be trusted, and the positive judgment result should be adopted as the final recognition result.

[0086] Third, the first composite recognition model outputs a negative judgment result, and the second composite recognition model outputs a negative judgment result: In this case, since the negative judgment result output by the second composite recognition model is highly credible, although the negative judgment result output by the first composite recognition model is of low credibility, the recognition of the negative judgment result by the second composite model is confirmed layer by layer, therefore, the judgment result of the second composite recognition model should be trusted, and the negative judgment result should be adopted as the final recognition result.

[0087] After analysis, it can be understood that it is actually equivalent to performing a similarity or OR operation on the binary classification judgment results input by the first composite recognition model and the second composite recognition model respectively, that is, when the outputs of both are the same logical positive value or negative value, it is judged that the logic is established, and the corresponding recognition result is accepted; when the outputs of both are mutually logical positive values ​​or negative values, it is judged that the logic is not established, and the corresponding recognition result is not accepted.

[0088] So far, it can be understood through this embodiment that the present application uses two image recognition models to respectively identify the target image of the object picture, wherein the first composite recognition model has a high credibility for the positive judgment result of the recognition judgment, and the second composite recognition model has a high credibility for the negative judgment result of the recognition judgment. When the judgment results of the two recognition models are the same, that is, the same positive and the same negative, the corresponding recognition judgment result is confirmed, otherwise the result is not confirmed and can be left for manual recognition. When both recognition models obtain the same judgment result, although the credibility of the judgment result of one of the recognition models is low, because the judgment result of the other recognition model is highly credible, the two identical results are more credible than a single result, so the same judgment result is trusted. When the two models identify different judgment results respectively, that is, one positive and one negative, it means that either the two recognition models both output untrustworthy judgment results and thus doubt the image content of the object picture, or the two recognition models both output credible different results and thus conflict, therefore, they can be temporarily excluded, such as left for manual recognition processing. It can be seen that the present application combines the XOR logic to realize a more credible network architecture. The judgment results obtained by identifying the target image of the object picture are definitely more credible than relying on a single recognition model, and can be directly used in the e-commerce platform.

[0089] It should be noted that in the present application, step S1200 and step S1300 have no logical sequence relationship, and therefore can be executed simultaneously. The execution order of the two should not be limited by the way of writing in the present application, and those skilled in the art should understand this.

[0090] For a more in-depth example, see Figure 2 Step S1200, calling the first composite recognition model to perform image classification and recognition on the object image, and obtaining a binary classification judgment result indicating whether the object image contains the target image, comprises the following steps:

[0091] Step S1210: Transmit the item image to the first preliminary screening full-image recognition sub-model of the first composite recognition model:

[0092] See also Figure 3 The network structure used by the first composite recognition model shown in the figure, according to which the object image is first passed into the first preliminary screening full-image recognition sub-model for target image detection. The first preliminary screening full-image recognition sub-model uses an image recognition model implemented by Mask-RCNN, whose basic network is ResNeXt101, and has achieved good results in practical applications. Of course, other excellent models in the field of target detection can also be selected to replace it.

[0093] Step S1220: the first preliminary screening full-image recognition sub-model recognizes the full-image of the item picture and determines a binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result. Otherwise, the operation of the first block recognition sub-model is triggered:

[0094] The first preliminary screening full-image recognition sub-model is pre-trained to convergence and performs detection based on the full-image of the object image. Of course, according to the needs of the sub-model, the image data of the object image can be pre-processed first, such as scaling to a fixed size, image standardization, etc., which should be understood by those skilled in the art.

[0095] After the first preliminary screening full-image recognition sub-model performs deep semantic understanding on the item image and extracts the image feature information, it will classify it according to the image feature information, thereby outputting a binary classification judgment result. When the judgment result outputs a negative judgment result, it directly triggers the first composite recognition module to output a negative judgment result, and ends the recognition work of the first composite recognition model. Otherwise, if the first preliminary screening full-image recognition sub-module outputs a positive classification result, the next sub-model in the first composite recognition model, namely the first block recognition sub-model, is called and run, in order to confirm the positive judgment result.

[0096] Step S1230: the first block recognition sub-model recognizes the local image of the item picture and determines the binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result. Otherwise, the operation of the first fine screening full-image recognition sub-model is triggered:

[0097] The first block recognition sub-model, in this embodiment, is an image recognition model directly implemented by ResNeXt101. Similarly, it has been pre-trained to a convergence state and can serve the purpose of this application. In practice, it has also achieved better results. Similarly, it can also be replaced by other excellent similar models.

[0098] The first block recognition sub-model is suitable for performing deep semantic extraction based on the local image of the object picture to obtain image feature information corresponding to the local image of the object picture. To this end, the object picture can be processed such as cropping, splicing, scaling to a fixed size, and image standardization, in order to highlight the detailed features of the target image therein, so that these detailed features can be more easily captured by the first block recognition sub-model.

[0099] After the first block recognition submodule obtains the image feature information, it can perform a binary classification judgment based on the image feature information to obtain the corresponding binary classification judgment result. Similar to the previous submodel, when the judgment result outputs a negative judgment result, it directly triggers the first composite recognition module to output a negative judgment result, and ends the recognition work of the first composite recognition model. Otherwise, if the first block recognition submodule outputs a positive classification result, the next submodel in the first composite recognition model, namely the first fine screening full-image recognition submodel, is triggered to call and run, in order to confirm the positive judgment result.

[0100] Step S1240: the first fine screening full-image recognition sub-model recognizes the full-image of the item picture and determines a binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result; otherwise, the first composite recognition model is triggered to output a positive judgment result:

[0101] The first fine-screening full-image recognition sub-model, in this embodiment, is an image recognition model implemented using the ViT (Vision Transformer) model. Similarly, it has been pre-trained to a convergence state and can serve the purpose of this application. It has also achieved better results in practice. Similarly, it can also be replaced by other excellent similar models.

[0102] Based on the first fine-screening full-image recognition sub-model of ViT, image feature information of the object image is extracted at the full-frame level. Of course, the object image can also be pre-formatted by the same method, including scaling to a fixed size, image standardization, etc.

[0103] Based on the extraction of the corresponding image feature information of the object image, the first fine-screening full-image recognition sub-model can obtain the corresponding binary classification judgment result. Figure 3 In the architecture shown, the first fine-screening full-image recognition sub-model is already the last-level sub-model in the first composite recognition model, so it is responsible for making the final judgment on the process of the first composite recognition model recognizing the object image. In this regard, if the binary classification judgment result obtained by the first fine-screening full-image recognition sub-model is a positive judgment result, the first composite recognition model is directly triggered to output a positive judgment result; if the binary classification judgment result is a negative judgment result, the first composite recognition model is directly triggered to output a negative judgment result, thereby ending the process of the first composite recognition model recognizing the target image of the object image.

[0104] It can be seen from this embodiment that three image recognition sub-models are used in the first composite recognition model, which are implemented based on three network structures, namely Mask-RCNN, ResNeXt101, and ViT. The three models have their own advantages. There are sub-models suitable for recognition based on the full image of the object picture, and there are sub-models suitable for recognition based on the partial image of the object picture. On this basis, the first composite recognition model directly releases the negative judgment results output by each sub-model, and repeatedly confirms the positive judgment results output by each sub-model, thereby ensuring that the positive judgment results finally determined by the first composite recognition model have high credibility, while its negative judgment results have lower credibility, that is, ensuring that the positive judgment results output by the first composite recognition model are reliable. Once this positive judgment result is output for an object picture, in the absence of a conflicting counterexample with the same high credibility, it can be directly determined that the object picture contains the target image.

[0105] See also Figure 4 In a further embodiment, step S1300, calling the second composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image, includes the following steps:

[0106] Step S1310: Transmit the item image to the second preliminary screening full-image recognition sub-model of the second composite recognition model:

[0107] See also Figure 5 The network structure used by the second composite recognition model shown in the figure, according to which the object image is first passed into the second preliminary screening full-image recognition sub-model for target image detection, and the second preliminary screening full-image recognition sub-model uses the image recognition model implemented by HTC (Hybrid Task Cascade for Instance Segmentation), which has achieved good results in practical applications. Of course, other excellent models in the field of target detection can also be selected to replace it.

[0108] Step S1320: the second preliminary screening full-image recognition sub-model recognizes the full-image of the item picture and determines a binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second block recognition sub-model is triggered to run:

[0109] The second preliminary screening full-image recognition sub-model is pre-trained to convergence and performs detection based on the full-image of the object image. Of course, according to the needs of the sub-model, the image data of the object image can be pre-processed first, such as scaling to a fixed size, image standardization, etc., which should be understood by those skilled in the art.

[0110] After the second preliminary screening full-image recognition sub-model performs deep semantic understanding on the item image and extracts the image feature information, it will classify it according to the image feature information, thereby outputting a binary classification judgment result. When the judgment result outputs a positive judgment result, it directly triggers the second composite recognition module to output a positive judgment result, and ends the recognition work of the second composite recognition model. Otherwise, if the second preliminary screening full-image recognition sub-module outputs a negative classification result, the next sub-model in the second composite recognition model, namely the second block recognition sub-model, is triggered to call and run, in order to confirm the negative judgment result.

[0111] Step S1330: the second block recognition sub-model recognizes the local image of the item picture and determines the binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result. Otherwise, the second fine screening full-image recognition sub-model is triggered to run:

[0112] The second block recognition sub-model, in this embodiment, is an image recognition model directly implemented using ResNet50. Similarly, it has been pre-trained to a convergence state and can serve the purpose of this application. In practice, it has also achieved better results. Similarly, it can also be replaced by other excellent similar models.

[0113] The second block recognition sub-model is suitable for performing deep semantic extraction based on the local image of the object picture to obtain image feature information corresponding to the local image of the object picture. To this end, the object picture can be processed such as cropping, splicing, scaling to a fixed size, and image standardization, in order to highlight the detailed features of the target image therein, so that these detailed features can be more easily captured by the second block recognition sub-model.

[0114] In an alternative embodiment, the second block recognition sub-model may perform cropping on the object image according to the following strategy:

[0115] 1) With the upper left corner of the item image as the vertex, cut a 300*300 pixel (or other values ​​such as 224*224 in actual use) block to the right and downward. The portion less than 300 pixels is filled with pure white or scaled to 300*300 after cutting.

[0116] 2) With the upper right corner of the item image as the vertex, cut a 300*300 pixel (or other values ​​such as 224*224 in actual use) block to the left and bottom. The portion less than 300 pixels is filled with pure white or scaled to 300*300 after cutting.

[0117] 3) With the lower left corner of the item image as the vertex, cut a 300*300 pixel (or other values ​​such as 224*224 in actual use) block to the right and upward. The portion less than 300 pixels is filled with pure white or scaled to 300*300 after cutting.

[0118] 4) With the lower right corner of the item image as the vertex, cut a 300*300 pixel (or other values ​​such as 224*224 in actual use) block to the left and top. The portion less than 300 pixels is filled with pure white or scaled to 300*300 after cutting.

[0119] 5) With the center point of the object image as the vertex, cut out blocks of 300*300 pixels (or other values ​​such as 224*224 in actual use) upward, downward, left, and right. For the part less than 300, fill it with pure white or scale it to 300*300 after cutting.

[0120] 6) Crop the original image into 9 small images in a 3*3 nine-square grid.

[0121] According to the above cropping method, a total of 14 small images can be obtained. At this point, these small images can be spliced ​​with features and then input into the second block recognition sub-model for target image recognition as required by the second block recognition sub-model.

[0122] After the second block recognition submodule obtains the image feature information, it can perform a binary classification judgment based on the image feature information to obtain the corresponding binary classification judgment result. Similar to the previous submodel, when the judgment result outputs a positive judgment result, it directly triggers the second composite recognition module to output a positive judgment result, and ends the recognition work of the second composite recognition model. Otherwise, if the second block recognition submodule outputs a negative classification result, the next submodel in the second composite recognition model, namely the second fine screening full-image recognition submodel, is triggered to call and run, in order to confirm the positive judgment result.

[0123] Step S1340: the second fine screening full-image recognition sub-model recognizes the full-image of the item picture and determines a binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result; otherwise, the second composite recognition model is triggered to output a negative judgment result:

[0124] The second fine-screening full-image recognition sub-model, in this embodiment, is an image recognition model implemented using the ViT (Vision Transformer) model. Similarly, it has been pre-trained to a convergence state and can serve the purpose of this application. It has also achieved better results in practice. Similarly, it can also be replaced by other excellent similar models.

[0125] Based on the second fine-screening full-image recognition sub-model of ViT, the image feature information of the object image is extracted at the full-frame level. Of course, the object image can also be pre-formatted by the same method, including scaling to a fixed size, image standardization, etc.

[0126] Based on the extraction of the corresponding image feature information of the object image, the second fine-screening full-image recognition sub-model can obtain the corresponding binary classification judgment results. Figure 5 In the architecture shown, the second fine-screening full-image recognition sub-model is already the last-level sub-model in the second composite recognition model, and therefore, it is responsible for making the final decision on the process of the second composite recognition model recognizing the object image. In this regard, if the binary classification judgment result obtained by the second fine-screening full-image recognition sub-model is a negative judgment result, the second composite recognition model is directly triggered to output a negative judgment result; if the binary classification judgment result is a positive judgment result, the second composite recognition model is directly triggered to output a positive judgment result, thereby ending the process of the second composite recognition model recognizing the target image of the object image.

[0127] It can be seen from this embodiment that three image recognition sub-models are adopted in the second composite recognition model, which are implemented based on three network structures, namely HTC, ResNet50 and ViT. The three models have their own advantages. There are sub-models suitable for recognition based on the full image of the object picture and sub-models suitable for recognition based on the partial image of the object picture. On this basis, the second composite recognition model directly releases the positive judgment results output by each sub-model, and repeatedly confirms the negative judgment results output by each sub-model, thereby ensuring that the negative judgment results finally determined by the second composite recognition model have high credibility, while its positive judgment results have lower credibility, that is, ensuring that the negative judgment results output by the second composite recognition model are reliable. Once this negative judgment result is output for an object picture, in the absence of a conflicting counterexample with the same high credibility, it can be directly determined that the object picture does not contain the target image.

[0128] See also Figure 6, a target image recognition device provided to meet one of the purposes of the present application is a functional embodiment of the target image recognition method of the present application, the device includes: an image acquisition module 1100, a first recognition module 1200, a second recognition module 1300, and a result fusion module 1400, wherein the image acquisition module 1100 is used to obtain an object image to be identified whether it contains a target image; the first recognition module 1200 is used to call the first composite recognition model to perform image classification recognition on the object image to obtain a binary classification judgment result representing whether the object image contains the target image, and the first composite recognition model is configured so that its positive judgment result has a higher credibility than its negative judgment result; the second recognition module 1300 is used to call the second composite recognition model to perform image classification recognition on the object image to obtain a binary classification judgment result representing whether the object image contains the target image, and the second composite recognition model is configured so that its negative judgment result has a higher credibility than its positive judgment result; the result fusion module 1400 is used to use the corresponding judgment result as the recognition result of the object image when the binary classification judgment results are the same.

[0129] In a further embodiment, the first compound recognition model and the second compound recognition model both include multiple sub-models that use different neural network structures to identify whether the object image contains the target image and obtain corresponding binary classification judgment results; in the first compound recognition model, when all sub-models obtain positive judgment results, they are triggered to output positive judgment results, and when any sub-model obtains negative judgment results, they are triggered to output negative judgment results; in the second compound recognition model, when all sub-models obtain negative judgment results, they are triggered to output negative judgment results, and when any sub-model obtains positive judgment results, it is triggered to output positive judgment results.

[0130] In a specific embodiment, the sub-models of the first compound recognition model and the second compound recognition model include a block recognition sub-model that performs image classification and recognition on a local image of the object picture to obtain a corresponding binary classification judgment result, and at least one full-image recognition sub-model that performs image classification and recognition on the full-frame image of the object picture to obtain a corresponding binary classification judgment result.

[0131] In a preferred embodiment, in the first composite recognition model or the second composite recognition model, each sub-model is implemented based on a different neural network model; and, in the first composite recognition model and the second composite recognition model, the sub-models used by each other are at least partially different.

[0132] In a further embodiment, the first recognition module 1200 includes: a first parameter transmission starter module, which is used to transmit the object image to the first preliminary screening full-image recognition sub-model of the first composite recognition model; a first preliminary screening recognition sub-module, which is used to call the first preliminary screening full-image recognition sub-model to recognize the full-image of the object image and determine the binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the operation of the first block recognition sub-model is triggered; the first block recognition sub-module is used to call the first block recognition sub-model Identify according to the partial image of the object picture and judge the binary classification result. When the binary classification result is a negative judgment result, trigger the first composite recognition model to output a negative judgment result, otherwise trigger the operation of the first fine screening full-image recognition sub-model; the first fine screening recognition sub-module is used to call the first fine screening full-image recognition sub-model to identify according to the full image of the object picture and judge the binary classification result. When the binary classification result is a negative judgment result, trigger the first composite recognition model to output a negative judgment result, otherwise trigger the first composite recognition model to output a positive judgment result.

[0133] In a further embodiment, the second composite recognition model is called to perform image classification and recognition on the object picture to obtain a binary classification judgment result indicating whether the object picture contains the target image, including the following steps: a second parameter transmission initiator module is used to transmit the object picture to the second preliminary screening full-image recognition sub-model of the second composite recognition model; a second preliminary screening recognition sub-module is used to call the second preliminary screening full-image recognition sub-model to perform recognition based on the full-image of the object picture to determine the binary classification judgment result. When the binary classification judgment result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the operation of the second block recognition sub-model is triggered. ; The second block recognition submodule is used to call the second block recognition submodel to perform recognition based on the local image of the object picture, and judge the binary classification judgment result. When the binary classification judgment result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second fine screening full-image recognition submodel is triggered to run; the second fine screening recognition submodule is used to call the second fine screening full-image recognition submodel to perform recognition based on the full image of the object picture, and judge the binary classification judgment result. When the binary classification judgment result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second composite recognition model is triggered to output a negative judgment result.

[0134] In order to solve the above technical problems, the present application also provides a computer device. Figure 7As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement a target image recognition method. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the target image recognition method of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0135] In this embodiment, the processor is used to execute Figure 6 The memory stores the program code and various data required to execute the above modules or submodules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program code and data required to execute all modules / submodules in the target image recognition device of this application, and the server can call the program code and data of the server to execute the functions of all submodules.

[0136] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the target image recognition method of any embodiment of the present application.

[0137] The present application also provides a computer program product, including a computer program / instruction, which implements the steps of the method described in any embodiment of the present application when executed by one or more processors.

[0138] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments of the present application can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0139] To summarize, this application performs target image recognition on object images by integrating two recognition models that are suitable for identifying target images and have high confidence in positive and negative results respectively, thereby distinguishing and summarizing various confidence situations, and obtaining more reliable target image recognition judgment results, thereby reducing manual labeling costs and improving the recognition efficiency of e-commerce platforms in processing massive product images.

[0140] It will be appreciated by those skilled in the art that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, altered, combined, or deleted. Furthermore, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted. Furthermore, the steps, measures, and schemes in the prior art that are similar to those disclosed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

[0141] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A target image recognition method, characterized in that: Used in e-commerce platform scenarios, including the following steps: Get the image of the object to be identified to see if it contains the target image; Calling the first composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image, wherein the first composite recognition model is configured such that its positive judgment result has a higher credibility than its negative judgment result; Calling the second composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image, wherein the second composite recognition model is configured such that its negative judgment result has a higher credibility than its positive judgment result; Wherein, both the first composite recognition model and the second composite recognition model include a plurality of sub-models that respectively use different neural network structures to identify whether the item image contains the target image and obtain corresponding binary classification judgment results; in the first composite recognition model, when all sub-models obtain positive judgment results, they are triggered to output positive judgment results, and when any sub-model obtains negative judgment results, they are triggered to output negative judgment results; in the second composite recognition model, when all sub-models obtain negative judgment results, they are triggered to output negative judgment results, and when any sub-model obtains positive judgment results, they are triggered to output positive judgment results; When the two-classification judgment results are the same, the corresponding judgment results are used as the recognition results of the object image.

2. The target image recognition method according to claim 1, characterized in that: The sub-models of the first composite recognition model and the second composite recognition model include a block recognition sub-model for performing image classification and recognition on a local image of the object picture to obtain a corresponding binary classification judgment result, and at least one full-image recognition sub-model for performing image classification and recognition on a full-image of the object picture to obtain a corresponding binary classification judgment result.

3. The target image recognition method according to claim 1, characterized in that: In the first composite recognition model or the second composite recognition model, each sub-model is implemented based on a different neural network model; and in the first composite recognition model and the second composite recognition model, the sub-models used by each other are at least partially different.

4. The target image recognition method according to claim 1, characterized in that: Calling the first composite recognition model to perform image classification and recognition on the object picture to obtain a binary classification judgment result indicating whether the object picture contains the target image includes the following steps: Transmitting the object image to the first preliminary screening full-image recognition sub-model of the first composite recognition model; The first preliminary screening full-image recognition sub-model recognizes the full-image of the item picture and determines a binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the operation of the first block recognition sub-model is triggered; The first block recognition sub-model recognizes the local image of the item picture and determines a binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the operation of the first fine screening full-image recognition sub-model is triggered; The first fine screening full-image recognition sub-model recognizes the full image of the object picture and determines a binary classification result. When the binary classification result is a negative judgment result, the first composite recognition model is triggered to output a negative judgment result, otherwise the first composite recognition model is triggered to output a positive judgment result.

5. The target image recognition method according to claim 1, characterized in that: Calling the second composite recognition model to perform image classification and recognition on the object picture to obtain a binary classification judgment result indicating whether the object picture contains the target image includes the following steps: Transmitting the object image to the second preliminary screening full-image recognition sub-model of the second composite recognition model; The second preliminary screening full-image recognition sub-model recognizes the full-image of the item picture and determines a binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second block recognition sub-model is triggered to run; The second block recognition sub-model recognizes the local image of the item picture and determines the binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second fine screening full-image recognition sub-model is triggered to run; The second fine screening full-image recognition sub-model recognizes the full image of the object picture and determines a binary classification result. When the binary classification result is a positive judgment result, the second composite recognition model is triggered to output a positive judgment result, otherwise the second composite recognition model is triggered to output a negative judgment result.

6. A target image recognition device, characterized in that: include: An image acquisition module is used to acquire an image of an object to be identified to determine whether it contains a target image; A first recognition module, used to call a first composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image, wherein the first composite recognition model is configured such that its positive judgment result has a higher credibility than its negative judgment result; A second recognition module, used to call a second composite recognition model to perform image classification and recognition on the object image to obtain a binary classification judgment result indicating whether the object image contains the target image, wherein the second composite recognition model is configured such that its negative judgment result has a higher credibility than its positive judgment result; The result fusion module is used to use the corresponding judgment results as the recognition results of the object image when the two-classification judgment results are the same.

7. A computer device comprising a central processing unit and a memory, characterized in that: The central processing unit is used to call and run the computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that: It stores a computer program implemented according to the method described in any one of claims 1 to 6 in the form of computer-readable instructions, and when the computer program is called and executed by a computer, the steps included in the corresponding method are executed.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

Images