Target object quantity prediction method and device, storage medium and electronic equipment

Abstract

Embodiments of the present application disclose a target object quantity prediction method and device, a storage medium and an electronic device. The method can be applied to the field of artificial intelligence. The method renders a visual information graph based on position distribution comparison information between a target region and each associated region, area distribution comparison information, and quantity comparison information of target objects between the target region and each associated region. Information in the visual information graph is mined from a graph theory perspective to obtain graph feature information. A region topology graph corresponding to the target region and spatiotemporal features in the target object quantities of the target region and each associated region are mined. The spatiotemporal features and the graph feature information are fused to predict a target quantity of target objects in the target region at a target time. This prediction method does not require accurate position information of target objects and quantity distribution data of administrative regions, thereby protecting the privacy and security of target objects and administrative units and improving the quantity prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to a method, apparatus, storage medium, and electronic device for predicting the number of target objects. Background Technology

[0002] The accuracy of predicting infectious diseases using relevant technologies is still somewhat insufficient. Taking the prediction of the number of infectious disease infections as an example, relevant technologies mainly predict the number of future infections by using historical infection numbers, weather, and manually extracted geographical features. Common modeling methods include differential equation models, time series models, and deep learning models. However, these technologies usually require accurate location information or infection data from administrative units. This information is difficult to obtain, and the prediction accuracy will decrease when this information is unavailable, thus failing to meet the requirements for predicting infectious diseases. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, embodiments of this application provide a method, apparatus, storage medium, and electronic device for predicting the number of target objects, thereby resolving the issue of insufficient accuracy in predicting infectious phenomena in related technologies.

[0004] On one hand, embodiments of this application provide a method for predicting the number of target objects, the method comprising:

[0005] Obtain a regional topology map corresponding to the target region. The regional topology map represents the orientational topology relationship between the target region and each associated region. The associated regions are regions adjacent to the target region.

[0006] The number of target objects in the target area and each of the associated areas within a preset time interval is obtained, and the target objects have the ability to spread and move.

[0007] A visualization information map is generated based on the number of target objects. The visualization information map represents the spatial comparison information between the target area and each of the associated areas, as well as the comparison information on the number of target objects between the target area and each of the associated areas. The spatial comparison information includes area comparison information and distribution location comparison information.

[0008] Spatiotemporal features are extracted from the number of target objects and the region topology to obtain spatiotemporal information;

[0009] Graph feature extraction is performed on the visualized information graph to obtain graph feature information;

[0010] The spatiotemporal information and the graph feature information are fused to obtain fused feature information;

[0011] The quantity prediction is performed on the fused feature information to obtain the target quantity of the target object in the target region at the target time, where the target time is the time after the preset time interval.

[0012] On the other hand, embodiments of this application provide a target object quantity prediction device, the device comprising:

[0013] A multimodal information acquisition module is used to acquire a regional topology map corresponding to a target area, wherein the regional topology map represents the directional topological relationship between the target area and each associated area, and the associated area is the area adjacent to the target area; acquire the number of target objects in the target area and each of the associated areas within a preset time interval, wherein the target objects have the ability to spread and move; and generate a visualization information map based on the number of target objects, wherein the visualization information map represents spatial comparison information between the target area and each of the associated areas, and comparison information on the number of target objects between the target area and each of the associated areas, wherein the spatial comparison information includes area comparison information and distribution location comparison information;

[0014] The multimodal information processing module is used to extract spatiotemporal features from the number of target objects and the regional topology map to obtain spatiotemporal information; extract graph features from the visualized information map to obtain graph feature information; fuse the spatiotemporal information and the graph feature information to obtain fused feature information; and predict the number of target objects in the target region at a target time, wherein the target time is a time after the preset time interval.

[0015] On the other hand, embodiments of this application provide a computer-readable storage medium storing at least one instruction or at least one program, wherein the at least one instruction or at least one program is loaded and executed by a processor to implement the above-described method for predicting the number of target objects.

[0016] On the other hand, embodiments of this application provide an electronic device, including at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the at least one processor implements the above-described target object quantity prediction method by executing the instructions stored in the memory.

[0017] On the other hand, embodiments of this application provide a computer program product, including a computer program or instructions, which, when executed by a processor, implement the above-described method for predicting the number of target objects.

[0018] This application provides a method, apparatus, storage medium, and electronic device for predicting the number of target objects. The method renders a visual information map based on the location distribution comparison information and area distribution comparison information between the target area and its associated areas, as well as the number comparison information of target objects between the target area and its associated areas. Information is mined from this visual information map from a graphics perspective to obtain graph feature information. Furthermore, the method mines the regional topology map corresponding to the target area, as well as the spatiotemporal features of the number of target objects in the target area and its associated areas. The spatiotemporal features and graph feature information are fused to predict the number of target objects in the target area at a target time. This prediction method does not require precise location information of the target objects, nor does it require the distribution data of the number of target objects in administrative areas, thus effectively protecting the privacy and security of the target objects and administrative units. Moreover, without needing this information, it improves the accuracy of the number prediction by relying on graph feature mining, ensuring that the accuracy of the target object number prediction still meets the prediction requirements. The data required by this application is readily available, and the prediction results are highly accurate, making the technical solution provided by this application easy to implement and possessing broad application prospects. Attached Figure Description

[0019] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a regional distribution diagram of infectious disease infection status provided in an embodiment of this application;

[0021] Figure 2 This is a schematic diagram showing the location distribution of infectious disease infection situations provided in an embodiment of this application;

[0022] Figure 3 This is a community distribution diagram of infectious disease infection status provided in an embodiment of this application;

[0023] Figure 4 This is a schematic diagram illustrating the implementation framework of the target object quantity prediction method provided in the embodiments of this application;

[0024] Figure 5 This is a flowchart illustrating a method for predicting the number of target objects provided in an embodiment of this application;

[0025] Figure 6 This is a schematic diagram of the orientation relationship of regional elements provided in the embodiments of this application;

[0026] Figure 7 This is a schematic diagram of the scatter plot generation method provided in the embodiments of this application;

[0027] Figure 8 This is a schematic diagram of a scatter plot provided in an embodiment of this application;

[0028] Figure 9 This is a schematic diagram of the overall framework provided in the embodiments of this application;

[0029] Figure 10 This is a block diagram of a target object quantity prediction device provided in an embodiment of this application;

[0030] Figure 11 This is a schematic diagram of the hardware structure of a device for implementing the method provided in the embodiments of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the embodiments of this application.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of the embodiments of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the present application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0033] To make the objectives, technical solutions, and advantages disclosed in the embodiments of this application clearer, the embodiments of this application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of this application and are not intended to limit the embodiments of this application.

[0034] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "multiple" means two or more. To facilitate understanding of the above-described technical solutions and their resulting technical effects in the embodiments of this application, the embodiments of this application first explain the relevant technical terms:

[0035] Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing. It encompasses network technologies, information technologies, integration technologies, management platform technologies, and application technologies based on cloud computing business models. These technologies can form resource pools, allowing for flexible and convenient on-demand use. Cloud computing technology will become a crucial support. Backend services of technical network systems require substantial computing and storage resources, such as video websites, image websites, and many portal websites. With the rapid development and application of the internet industry, each resource may have its own identification mark in the future, requiring transmission to backend systems for logical processing. Data at different levels will be processed separately, and various industry data will require robust system support, which can only be achieved through cloud computing.

[0036] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0037] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / switching systems, and mechatronics. AI software technologies encompass several major areas, including computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0038] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0039] Deep learning: The concept of deep learning originates from the research of artificial neural networks and belongs to the field of machine learning. A multilayer perceptron with multiple hidden layers is a type of deep learning architecture. Deep learning discovers distributed feature representations of data by combining low-level features to form more abstract high-level representations of attribute categories or features.

[0040] Transformer: A model that uses a self-attention structure to extract pairwise interactions between elements in a sequence, widely used in natural language processing, image processing, and temporal prediction. Transformer is essentially an Encoder-Decoder model based on a multi-head attention mechanism. The input to the Transformer Encoder is the word embedding representation of a sentence and its corresponding positional encoding information. The core layer of the model is a multi-head attention mechanism. This mechanism uses multiple attention mechanisms to perform independent computations to obtain more layers of semantic information, and then concatenates and combines the results to obtain the final result. The Add&Norm layer sums and normalizes the input and output of the Multi-Head Attention layer before passing it to the Feed Forward layer. Finally, an Add&Norm process is performed again to output the final word vector matrix. Transformer is a combination of fully connected (or one-dimensional convolution) and attention. The algorithm has good parallelism and is suitable for current hardware environments.

[0041] Infectious disease area risk prediction: Based on the historical number of infectious disease infections, environmental factors, climate and other factors within a designated area, predict the future risk of infectious disease infection in a target area related to the designated area. The risk value can be calculated from the future number of infections.

[0042] ViT is an image classification model composed of a multi-layer Transformer structure. This model first divides the image into multiple local blocks, and then extracts the interaction features between different blocks through the SelfAttention mechanism, which can effectively model the feature relationships between different locations in the image.

[0043] Graph Neural Networks: Graph neural networks are a method for processing graph information in deep learning. In order to process edge relationships of various categories in a graph simultaneously, relational graph neural networks first process edge relationships of various categories independently, and then aggregate the results of all categories.

[0044] Traditional infectious disease regional risk prediction schemes mainly predict future infection numbers based on historical infection numbers, weather, and manually extracted geographical features. Common modeling methods include differential equation models, time series models, and deep learning models.

[0045] Early infectious disease risk prediction models primarily relied on differential equation theory. These models first extracted parameters S (susceptible individuals), E (exposed individuals), I (infected individuals), and R (recovered individuals) from infection data, and then calculated the proportions of each population group in each period using differential equations. This method is highly dependent on data completeness; it is difficult to construct a complete model without data on deaths and recoveries. Furthermore, these models cannot fit changes in the number of new infections, and their flexibility has room for improvement.

[0046] Time series forecasting models are often used for predicting the number of people infected with infectious diseases. These models consist of two parts: an autoregressive model (AR) and a moving average model (MA). In the data fitting process, the time series data is first stabilized using the differencing method, and then AR and MA are combined to predict the number of new infections in the future.

[0047] Deep learning models primarily utilize historical infection numbers, weather changes, and population movement data to build their models. These works rely heavily on pure data features, which can easily overlook valuable features when modeling geographic information. Furthermore, these methods only employ historical infection numbers, neglecting the impact of inter-regional population movement changes on the spread of the epidemic.

[0048] In summary, the methods used in these related technologies primarily construct infectious disease prediction models through purely numerical features, with limited exploration of visual image features representing infection status. This makes it difficult to capture complete geographic information in infectious disease regional risk prediction tasks. Please refer to [the relevant resources / references]. Figure 1 It shows a regional distribution diagram of infectious disease infection. Related technical solutions tend to overlook the spatial relationship between the surrounding areas and the central area. Figure 1The central region, filled with solid material, is surrounded by three closely connected smaller regions (filled with diagonal stripes) exhibiting high infection rates. For the central region, a large number of infected individuals clustered in the same location pose a greater risk than those dispersed in various directions; however, this information is difficult for relevant technologies to capture. Capturing information on the locational distribution of the surrounding regions related to the central region and the infection status within each of these regions is crucial for predicting the risk of infectious diseases in the central region. However, relevant technologies struggle to accurately model this type of information, thus hindering its ability to improve the accuracy of infectious disease risk prediction in the central region.

[0049] Related technologies typically require a high degree of precision in locating the infection. Please refer to [reference needed]. Figure 2 It shows a schematic diagram of the location distribution of infectious disease infections. Related technologies, when used for regional risk prediction of infectious diseases, need to obtain... Figure 2 The precise location distribution information shown is incredibly difficult to obtain. Alternatively, please refer to... Figure 3 It shows a community distribution diagram of infectious disease infection, with communities represented by grid areas. Related technologies can also predict the future number of infections in a target street area, for example, by obtaining the location of infection points at the street level, assuming the system can obtain the number of infections in the next level of communities. However, obtaining this information remains quite difficult.

[0050] Regardless of Figure 2 Information or Figure 3 Information is difficult to obtain, and using Figure 2 The technical solutions related to the information in the data have high accuracy requirements for the address information of infected individuals (e.g., when building a street-level infectious disease population prediction model, it is necessary to obtain the number of infected individuals at the community level). In practical application scenarios, they will be subject to certain limitations. Since the visualization image features are not used as auxiliary information for the task of predicting the risk of infectious disease areas, the related technologies are difficult to achieve the expected accuracy of infectious disease area prediction when it is difficult to obtain the address information of infected individuals and community infection information.

[0051] In view of this, this application provides a target object quantity prediction scheme. This scheme renders a visual information map based on the location distribution comparison information and area distribution comparison information between the target area and its associated areas, as well as the target object quantity comparison information between the target area and its associated areas. Information is mined from this visual information map from a graphics perspective to obtain graph feature information. Furthermore, the scheme mines the regional topology map corresponding to the target area, as well as the spatiotemporal features of the target object quantity in the target area and its associated areas. The spatiotemporal features and graph feature information are fused to predict the target object quantity in the target area at a target time. This prediction method does not require precise location information of the target objects, nor does it require the quantity distribution data of target objects in administrative regions, thus effectively protecting the privacy and security of the target objects and administrative units. Moreover, without needing this information, it improves the accuracy of quantity prediction by relying on graph feature mining, ensuring that the accuracy of target object quantity prediction still meets the prediction requirements. Therefore, using high-accuracy quantity prediction, the target area can be accurately classified into infection risk levels. The data required by this application embodiment is readily available, and the prediction results are highly accurate, making the technical solution provided by this application embodiment easy to implement and possessing broad application prospects.

[0052] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating the implementation framework of the target object quantity prediction method provided in the embodiments of this specification, such as... Figure 4 As shown, the implementation framework may include at least a client 10 and a server 20. The client 10 and server 20 communicate via a network 30. This implementation framework can also be considered a target object quantity prediction system, which is used to train a target object quantity prediction model and provide target object quantity prediction services based on the model. The server 20 may be located in a cloud environment, and the server 20 is a node in the target object quantity prediction system in the cloud environment. This node can be any node in the cloud environment.

[0053] The server 20 can first train a target object quantity prediction model. Once the target object quantity prediction model is obtained, the server 20 can provide target object quantity prediction services to external parties. Upon receiving a target object quantity prediction request from client 10, a regional topology map corresponding to the target area can be obtained. This regional topology map represents the directional topological relationship between the target area and each associated area, where the associated areas are adjacent to the target area. The number of target objects in the target area and each associated area within a preset time interval is obtained. Target objects possess infectivity and mobility. A visualization information map is generated based on the number of target objects. This visualization information map represents spatial comparison information between the target area and each associated area, as well as comparison information on the number of target objects between the target area and each associated area. The spatial comparison information includes area comparison information and distribution location comparison information. Spatiotemporal features are extracted from the number of target objects and the regional topology map to obtain spatiotemporal information. Graph features are extracted from the visualization information map to obtain graph feature information. The spatiotemporal information and the graph feature information are fused to obtain fused feature information. Quantity prediction is performed on the fused feature information to obtain the target quantity of target objects in the target area at a target time, where the target time is a time after the preset time interval.

[0054] The framework described in this application embodiment can provide the ability to predict the number of target objects required for applications in various scenarios, including but not limited to cloud technology, cloud gaming, cloud rendering, artificial intelligence, smart transportation, assisted driving, video media, smart communities, and instant messaging. Each component in this framework can be a terminal device or a server. Terminal devices include, but are not limited to, mobile phones, computers, intelligent voice exchange devices, smart home appliances, and in-vehicle terminals.

[0055] The following describes a method for predicting the number of target objects according to an embodiment of this application. Figure 5 This document illustrates a flowchart of a target object quantity prediction method provided in an embodiment of this application. This target object quantity prediction method can be executed based on the server 20 described above. While this application provides the method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive methods. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual systems, terminal devices, or server products, the method can be executed sequentially according to the embodiments or accompanying drawings, or in parallel (e.g., in a parallel processor or multi-threaded processing environment). The above method may include:

[0056] S501. Obtain the regional topology map corresponding to the target area. The regional topology map represents the orientational topology relationship between the target area and each associated area. The associated areas are the areas adjacent to the target area.

[0057] This application does not limit the method for obtaining the regional topology map. It can be obtained directly using geographic information data with topological information, or it can be generated by itself using geographic information data without topological information. In some embodiments, the latitude and longitude boundary coordinates of the target area and related areas can be extracted from the geographic information data. The degree of adjacency of the areas is analyzed by the latitude and longitude overlap relationship of adjacent regional elements. In this application embodiment, the target area or related area is a type of regional element.

[0058] Please refer to Figure 6 It shows a schematic diagram of the orientation relationship of the regional elements. Figure 6 In the left image, the solid-filled area represents the target area, and the striped-filled area represents the associated area. These two types of area elements can share multiple common latitude and longitude coordinates. Therefore, edge relationships can be constructed between two adjacent area elements, with the number of shared latitude and longitude coordinates used as the edge weight, thus obtaining... Figure 6 The region topology graph in the right figure is G = (V, E), where V represents each region element and E represents the edges and weights between each region element determined by the number of latitude and longitude boundary points.

[0059] S502. Obtain the number of target objects in the aforementioned target area and each of the aforementioned associated areas within a preset time interval, wherein the target objects possess the ability to spread and move.

[0060] The embodiments of this application do not limit the target object, which only needs to have the ability to spread and move. For example, the target object can be a patient with an infectious disease, a healthy person carrying an infectious virus, or a mobile organism with the ability to spread.

[0061] This application embodiment does not limit the preset time interval, and can have at least one time window. The time windows may overlap or not overlap, and the spans of different time windows may be different or the same. This application embodiment does not limit these aspects. Each time window may include at least two time points. At each time point, the number (sub-quantity) of target objects in the target area and related areas can be collected. These sub-quantities all contribute to the aforementioned number of target objects. For example, if the preset time interval is January 1st to January 10th, three time windows can be set: time window 1 corresponds to January 1st to January 8th, time window 2 corresponds to January 2nd to January 9th, and time window 3 corresponds to January 3rd to January 10th. Each time window includes 8 time points, one time point per day. Each area element at each time point corresponds to a sub-quantity, which represents the number of target objects in that area element at that time point.

[0062] S503. Generate a visualization information map based on the number of target objects. The visualization information map represents the spatial comparison information between the target area and each of the associated areas, as well as the comparison information on the number of target objects between the target area and each of the associated areas. The spatial comparison information includes area comparison information and distribution location comparison information.

[0063] This application does not limit the method for generating the visualization infographic, as long as the generated visualization infographic includes the information in S503. It should be noted that this information should have visual distinguishability, because this application uses graph mining to analyze the information in the visualization infographic to improve the accuracy of quantity prediction. If the visual distinguishability of different information, i.e., the visual effect is not obvious, it is difficult to meet the quantity prediction accuracy requirements of this application. Therefore, common grayscale images or gridded images in related technologies are difficult to meet the requirements of the visualization infographic of this application. This application provides a method for generating a scatter plot visualization infographic, which can be used as the visualization infographic in S503. Please refer to... Figure 7 The diagram illustrates a scatter plot generation method, which includes:

[0064] S701. Based on the geographical location information of the target area and each of the related areas, determine the centroid and area of ​​the target area and each of the related areas.

[0065] This application does not limit the methods for calculating the centroid and area. Taking the calculation of the centroid of a target region as an example, in one embodiment, the centroid point of the region can be calculated through the latitude and longitude boundaries of the target region. Specifically, the irregular target region is first divided into multiple small triangles, and the area A of each triangle is calculated. iWith center of mass The centroid of the target region is obtained by the weighted average of the centroids of all the smaller triangles, such as... As shown in the embodiment of this application, i represents the triangle identifier, x represents the horizontal coordinate, y represents the vertical coordinate, and C represents the centroid.

[0066] S702. Determine the scaling values ​​of the target region and each of the associated regions based on their respective areas and the preset mapping relationship, wherein the mapping relationship represents the correspondence between the area and the scaling value.

[0067] The embodiments of this application do not limit the method of constructing the mapping relationship. The mapping relationship only needs to be constructed according to the following principle: map the views of the regional elements belonging to different area scales to appropriate scaling values ​​to ensure that most of the images of the regional elements and other surrounding regional elements can be displayed by the scatter visualization view.

[0068] S703. Render the target region and the associated region according to the above scaling values ​​to obtain the region distribution map.

[0069] Specifically, after obtaining the spatial location information of the target area and each of the associated areas, the display position of each point in the target area and each of the associated areas on the electronic device can be determined based on the spatial location information and the scaling value, thereby rendering the target area and each of the associated areas to obtain a regional distribution map.

[0070] S704. Render the corresponding pattern at each centroid in the above-mentioned regional distribution map to obtain the above-mentioned visualization information map. The area of ​​the pattern corresponding to the centroid is proportional to the number of target objects in the region element corresponding to the centroid. The region element is the above-mentioned target region or the above-mentioned associated region.

[0071] In this embodiment, the rendering methods used in steps S703 and S704 are not limited. That is, the fill and line patterns of the target area and each associated area are not limited, nor are the fill and line patterns of the patterns at the centroids. The pattern type is also not limited; for example, it can be a star pattern, a circular pattern, etc. However, the patterns corresponding to the target area and each associated area should be the same pattern. To enhance the expressive power of the visualization infographic, different rendering methods can be used in S703 and S704. Specifically, in this embodiment, the target area and each associated area can be filled using a first fill method to obtain the area distribution map. The patterns corresponding to each centroid can be filled using a second fill method to obtain the visualization infographic. The second fill method is different from the first fill method. This increases the distinguishability of different information in the resulting visualization infographic—the scatter plot—and improves the graph information mining effect. This embodiment does not limit the fill methods; color filling, pattern filling, etc., can be used. For example, in this embodiment, the Leaflet.js tool can be used to fill the target area and related areas with red and yellow blocks with an opacity of 0.2 respectively. Circular blocks with an opacity of 0.5 are then filled with the centroid of the area elements as the center and the number of infected individuals as the radius, resulting in a scatter plot. Leaflet.js is a mapping library. It is a flexible, lightweight, open-source library for creating interactive maps. Leaflet is a framework for presenting map data. The data and base map layers are provided by the developers. The map consists of tile layers, browser support, default interactivity, panning, and zooming capabilities. More custom layers and plugins can also be added. This map library converts data into map layers and provides excellent support; it works well on major desktop and mobile platforms, making it a supporting library for mobile and large-screen maps.

[0072] Please refer to Figure 8 It shows a schematic diagram of a scatter plot. Figure 8 Each circle forms a scatter plot, representing the number of infected people within the region element where the circle's center is located. In this embodiment, a corresponding scatter plot can be generated for each region element at each time point under each window, and one or more scatter plots can be selected as the execution result of S503.

[0073] S504. Perform spatiotemporal feature extraction on the number of the above target objects and the above regional topology map to obtain spatiotemporal information.

[0074] This application does not limit the spatiotemporal feature extraction method. Some related neural networks, such as Transformer or graph neural networks, can be used alone or in combination to perform spatiotemporal feature extraction operations. To achieve better spatiotemporal feature extraction results, this application provides a spatiotemporal feature extraction method, which includes the following steps:

[0075] (1) Obtain the adjacency matrix based on the above region topology map. The adjacency matrix represents the adjacency relationship and degree of adjacency of each of the above region elements.

[0076] Specifically, this degree of adjacency can be represented by the number of shared latitude and longitude coordinates mentioned above, or by the normalized result of the number of shared latitude and longitude coordinates.

[0077] (2) Based on the number of target objects mentioned above, obtain the time sequence information of the quantity of each of the above-mentioned regional elements in each of the above-mentioned window intervals.

[0078] (3) For each window, multimodal feature extraction is performed on the above adjacency matrix and the corresponding quantity time series information to obtain the first information; multimodal feature extraction is performed based on the above first information and the above adjacency matrix to obtain the second information; the above quantity time series information, the above first information and the above second information are fused to obtain the spatiotemporal information under the above window.

[0079] This application does not limit the multimodal feature extraction method. For example, based on a message-propagation neural network, activation operations, message-propagation-based information extraction operations, and normalization operations are sequentially performed on the adjacency matrix and quantity time-series information to obtain the first information. The second information is obtained in the same way as the first information. Based on the same principle, more information, such as the third, fourth, and fifth information, can also be obtained. The obtained information and the quantity time-series information corresponding to the above window are fused to obtain the spatiotemporal information under the above window. This application does not specify the fusion method. For example, a combination of fully connected operations, normalization operations, and message-propagation-based information extraction operations can be used for fusion.

[0080] This application does not limit the preset activation layer. For example, the activation layer can use the Sigmoid activation function. The Sigmoid function is a common S-shaped function in biology, also known as an S-shaped growth curve. In information science, due to its monotonically increasing and inversely monotonically increasing properties, the Sigmoid function is often used as the activation function of neural networks, mapping variables to the range [0,1]. Alternatively, pRelU can also be used. The pRelU activation function is a RelU function with a function. The RelU function, also known as the Linear Rectification Function or Modified Linear Unit, is a commonly used activation function in artificial neural networks, usually referring to nonlinear functions represented by the ramp function and its variants.

[0081] (4) Encode the spatiotemporal information under each window using long short-term memory to obtain the above spatiotemporal information.

[0082] Based on the spatiotemporal information obtained under the window, the spatiotemporal information under each window can be encoded using long short-term memory encoding to obtain the aforementioned spatiotemporal information.

[0083] S504. Extract graph features from the above-mentioned visualization information graph to obtain graph feature information.

[0084] This application does not limit the specific method for graph feature extraction; related technologies can be referenced. To improve the graph feature extraction effect, this application provides a graph feature extraction method, which includes the following steps: performing image segmentation on the above-mentioned visualization information graph to obtain multiple image sub-blocks; performing linear radial encoding and positional encoding on each of the above-mentioned image sub-blocks to obtain multiple encoding results; and performing information extraction based on a self-attention mechanism on each of the above-mentioned encoding results to obtain the above-mentioned graph feature information. This application does not limit the specific methods of linear radial encoding, positional encoding, and information extraction based on a self-attention mechanism; it can be implemented using VIT combined with Transformer.

[0085] S505. The above-mentioned spatiotemporal information and the above-mentioned graph feature information are fused to obtain fused feature information; the above-mentioned fused feature information is used to predict the quantity of the target object in the target area at the target time, and the above-mentioned target time is the time after the above-mentioned preset time interval.

[0086] This application embodiment does not limit the target time, which is a time located after the above-mentioned preset time interval. For example, if the preset time interval is January 1 to January 10, then the target time can be January 11, February 1, etc. The relationship between the target time and the preset time interval only needs to be consistent with the relationship between the sample time point and the sample time interval used when training the target quantity prediction model. The target quantity prediction model is the model that implements the above steps S503 to S505.

[0087] The aforementioned target quantity prediction model includes a first model, a second model, and a third model. This model is trained using the following method: First, a sample region and the number of target objects corresponding to that region at a given time point are obtained. Second, a sample time interval is determined based on the given time point. Third, a sample region topology map is obtained. Fourth, the number of target objects in the sample region and its associated regions within the given time interval is obtained. Fifth, a sample visualization information map is generated based on the number of target objects. Sixth, spatiotemporal features are extracted from the target object number and the sample region topology map using the first model to obtain sample spatiotemporal information. Seventh, graph features are extracted from the visualization information map using the second model to obtain sample graph feature information. Eighth, the spatiotemporal information and the graph feature information are fused using the third model to obtain fused feature information. Finally, a quantity prediction is performed on the fused feature information to obtain the predicted quantity of the sample region at the given time point. Finally, the parameters of the first, second, and third models are adjusted based on the difference between the predicted quantity and the number of target objects at the given time point to obtain the final target quantity prediction model.

[0088] This application embodiment does not elaborate on the execution process of the target quantity prediction model; please refer to the preceding text. In this application embodiment, the above parameters can be adjusted based on gradient descent. Gradient descent is a commonly used method in machine learning and deep learning for first-order optimization of network parameters. In this application embodiment, gradient descent can guide the above parameters to adjust in a direction that reduces the aforementioned differences. When the number of adjustments reaches a preset threshold, or when the aforementioned differences are less than a preset difference threshold, parameter tuning is stopped, and the target quantity prediction model is obtained. Of course, this application embodiment does not limit the method for measuring the aforementioned differences; for example, cross-entropy loss can be used.

[0089] This application does not limit the specific structures of the first, second, and third models. In one embodiment, the first model includes a sequentially connected multi-layer message-passing neural network, with the last layer being a long short-term memory network. The above-mentioned extraction of spatiotemporal features from the number of target objects and the topology of the sample region based on the first model to obtain sample spatiotemporal information includes: performing spatial information-based feature extraction on the number of target objects and the topology of the sample region based on the multi-layer message-passing neural network to obtain spatial feature extraction results; and performing time-based feature extraction on the spatial feature extraction results based on the long short-term memory network to obtain the sample spatiotemporal information. The third model includes a sequentially connected plurality of feature connection layers, a fully connected layer, and an information output layer. The above-mentioned fusion of the sample spatiotemporal information and the sample graph feature information based on the third model to obtain sample fusion feature information includes: inputting the sample spatiotemporal information and the sample graph feature information into a sequentially connected plurality of feature connection layers to obtain the sample fusion feature information. The above-mentioned quantity prediction of the sample fusion feature information to obtain the predicted quantity of the sample region at the sample time point includes: inputting the sample fusion feature information into a fully connected layer and an information output layer in sequence to obtain the predicted quantity of the sample region at the sample time point.

[0090] In this embodiment, after predicting the target quantity corresponding to the target area at the target time, the infection risk level of the target area can be determined based on the predicted target quantity. This embodiment does not limit the determination method; for example, a correspondence between the floating range of the target quantity and the infection risk level can be established, and the infection risk level of the target area can be determined based on this correspondence. In a specific implementation, risk level reference information can be obtained, including at least two infection risk levels and a density interval corresponding to each infection risk level. Based on the predicted target quantity and the area of ​​the target area, the predicted density corresponding to the target area is determined. The density interval containing the predicted density is determined from the risk level reference information. The infection risk level pointed to by the density interval containing the predicted density is determined as the infection risk level of the target area. For example, if density interval A1 is {a1, b1}, its corresponding infection risk level is low risk; density interval A2 is {a2, b2}, its corresponding infection risk level is medium risk; and density interval A3 is {a3, b3}, its corresponding infection risk level is high risk, where b1 is less than or equal to a2, and b2 is less than or equal to a3. If the predicted density is greater than a1 and less than b1, the target area is a low-risk area.

[0091] This application's embodiments introduce scattered point visualization image features into quantity prediction and infectious risk level prediction for the first time. Taking regional risk prediction in infectious disease scenarios as an example, this application's embodiments can introduce visualization features and improve the model's prediction performance without requiring precise location of infected individuals. Table 1 shows the experimental data on the execution effect of this application's embodiments. The experiment compares the prediction effects of this application's embodiments with related technologies. In Table 1, AVG, LAST_DAY, AVG_WINDOW, LSTM, and TARGET represent the average value method, the previous day's infection number prediction method, the sliding window prediction method, the long short-term memory network method, and the method in this application's embodiments, respectively. The prediction execution effects of each method are evident in Table 1. As can be seen from Table 1, the mean absolute error (MAE) of this application's embodiments is lower than that of baseline models such as AVG, the previous day's infection number prediction (LAST_DAY), and the sliding window prediction (AVG_WINDOW). Furthermore, the method using the embodiments of this application outperforms the models in the related art on three different settings (predicting the next 1-3 days (Uptonext3Days), predicting the next 1-7 days (Uptonext7Days), and predicting the next 1-14 days (Upto next14Days)).

[0092] Table 1

[0093]

[0094] You can refer to this. Figure 9 This document illustrates the overall framework of the solution in this application embodiment. In the infectious disease prediction scenario, the technical solution in this application embodiment can extract the adjacency relationships between regional elements based on the geographic information data of each regional element, obtaining a regional topology map of all adjacent regional elements of the target region. In the visualization stage, this application embodiment can obtain the number of infections in the target region and related regions at each time point in each window based on historical infection data, and display the infection status of each regional element through the area of ​​scatter plot circles. In the multimodal feature extraction and fusion stage, this application embodiment first extracts spatiotemporal features, and then extracts image features from the scatter plot visualization view, predicting the future number of infections in the target region through feature fusion. This application embodiment uses scatter plots to visualize the number of infections in the target region or related regions, without relying on precise address information of infected individuals, which can better promote the method of using map features to improve prediction results to practical applications. In terms of product form, this application embodiment can predict the future number of infections based on the historical number of infections in each region, visually display the risk of infectious disease outbreaks in each region, and assist in infectious disease early warning and prevention.

[0095] This application provides a method, apparatus, storage medium, and electronic device for predicting the number of target objects. The method renders a visual information map based on the location distribution comparison information and area distribution comparison information between the target area and its associated areas, as well as the number comparison information of target objects between the target area and its associated areas. Information is mined from this visual information map from a graphics perspective to obtain graph feature information. Furthermore, the method mines the regional topology map corresponding to the target area, as well as the spatiotemporal features of the number of target objects in the target area and its associated areas. The spatiotemporal features and graph feature information are fused to predict the number of target objects in the target area at a target time. This prediction method does not require precise location information of the target objects, nor does it require the distribution data of the number of target objects in administrative areas, thus effectively protecting the privacy and security of the target objects and administrative units. Moreover, without needing this information, it improves the accuracy of the number prediction by relying on graph feature mining, ensuring that the accuracy of the target object number prediction still meets the prediction requirements. The data required by this application is readily available, and the prediction results are highly accurate, making the technical solution provided by this application easy to implement and possessing broad application prospects.

[0096] Please refer to Figure 10 The diagram illustrates a block diagram of a target object quantity prediction device in this embodiment, the device comprising:

[0097] The multimodal information acquisition module 1001 is used to acquire a regional topology map corresponding to a target area, wherein the regional topology map represents the directional topological relationship between the target area and each associated area, and the associated areas are areas adjacent to the target area; acquire the number of target objects in the target area and each of the associated areas within a preset time interval, wherein the target objects have the ability to spread and move; and generate a visualization information map based on the number of target objects, wherein the visualization information map represents spatial comparison information between the target area and each of the associated areas, and comparison information on the number of target objects between the target area and each of the associated areas, wherein the spatial comparison information includes area comparison information and distribution location comparison information;

[0098] The multimodal information processing module 1002 is used to extract spatiotemporal features from the number of target objects and the topological map of the region to obtain spatiotemporal information; extract graph features from the visualization information map to obtain graph feature information; fuse the spatiotemporal information and the graph feature information to obtain fused feature information; and predict the number of target objects in the target region at a target time, wherein the target time is a time after the preset time interval.

[0099] In one embodiment, the multimodal information acquisition module 1001 described above is used to perform the following operations:

[0100] Based on the geographical location information of the target area and each of the aforementioned related areas, determine the centroid and area of ​​each of the target area and each of the aforementioned related areas.

[0101] Based on the respective areas and the preset mapping relationship, the scaling values ​​of the target area and each of the associated areas are determined, and the mapping relationship represents the correspondence between the area and the scaling value.

[0102] Render the target region and the associated region based on the above scaling values ​​to obtain a region distribution map;

[0103] The corresponding pattern is rendered at each centroid in the above regional distribution map to obtain the above visualization information map. The area of ​​the pattern corresponding to the centroid is proportional to the number of target objects in the region element corresponding to the centroid. The region element is the target region or the associated region.

[0104] In one embodiment, the multimodal information acquisition module 1001 described above is used to perform the following operations:

[0105] The target area and each of the related areas are filled using the first fill method to obtain the area distribution map.

[0106] The pattern corresponding to each centroid is filled using a second filling method to obtain the above-mentioned visualization information map. The second filling method is different from the first filling method.

[0107] In one embodiment, the preset time interval includes at least one window interval, each of the window intervals includes at least two sub-intervals, the number of target objects includes the number of target objects in each region element of each of the sub-intervals, and the region element is the target region or the associated region. The multimodal information processing module 1002 is used to perform the following operations:

[0108] Based on the above region topology map, an adjacency matrix is ​​obtained. The adjacency matrix represents the adjacency relationship and degree of adjacency of each element in the above region.

[0109] Based on the number of target objects mentioned above, the time-series information of the quantity of each of the above-mentioned regional elements in each of the above-mentioned window intervals is obtained;

[0110] For each window, multimodal feature extraction is performed on the adjacency matrix and the corresponding quantity time-series information to obtain first information; multimodal feature extraction is performed on the first information and the adjacency matrix to obtain second information; the quantity time-series information corresponding to the window, the first information and the second information are fused to obtain the spatiotemporal information under the window.

[0111] In one embodiment, the multimodal information processing module 1002 is used to perform the following operation: encode the spatiotemporal information under each window using long short-term memory to obtain the spatiotemporal information.

[0112] In one embodiment, the multimodal information processing module 1002 is configured to perform the following operation: perform image segmentation on the above-mentioned visualization information map to obtain multiple image sub-blocks;

[0113] Linear radial and positional coding are performed on each of the above image sub-blocks to obtain multiple coding results;

[0114] Information is extracted from the above encoding results based on a self-attention mechanism to obtain the graph feature information.

[0115] In one embodiment, the above method is implemented based on a target quantity prediction model, which includes a first model, a second model, and a third model. The apparatus further includes a training module 903 for performing the following operations:

[0116] Obtain the sample region and the number of target objects corresponding to the sample region at the sample time point;

[0117] Determine the sample time interval based on the above sample time points;

[0118] Obtain the sample region topology map corresponding to the above sample regions;

[0119] Obtain the number of target objects in the sample time interval for each of the above sample regions and their associated regions.

[0120] Generate a sample visualization information map based on the number of target objects mentioned above;

[0121] Based on the first model described above, spatiotemporal features are extracted from the number of target objects in the sample and the topological map of the sample region to obtain the spatiotemporal information of the sample.

[0122] Based on the second model described above, graph features are extracted from the above-mentioned sample visualization information graph to obtain sample graph feature information;

[0123] Based on the third model described above, the spatiotemporal information of the samples and the feature information of the sample maps are fused to obtain sample fusion feature information; and the quantity of the sample fusion feature information is predicted to obtain the predicted quantity of the sample region at the sample time point.

[0124] Based on the difference between the predicted quantity and the number of target objects corresponding to the above sample time points, the parameters of the first model, the second model, and the third model are adjusted to obtain the target quantity prediction model.

[0125] In one embodiment, the first model includes a sequentially connected multilayer message-passing neural network and a long short-term memory network; the third model includes a sequentially connected multiple feature connection layers, a fully connected layer, and an information output layer; and the training module 903 is used to perform the following operations:

[0126] Based on a multi-layer message passing neural network, spatial feature extraction is performed on the number of target objects in the above samples and the topology of the sample region to obtain spatial feature extraction results.

[0127] Based on the Long Short-Term Memory network, the spatial feature extraction results above are subjected to time-based feature extraction to obtain the spatiotemporal information of the above samples.

[0128] The above-mentioned sample spatiotemporal information and sample map feature information are input into multiple feature connection layers that are sequentially connected to obtain the above-mentioned sample fusion feature information.

[0129] By sequentially connecting the above sample fusion feature information into a fully connected layer and an information output layer, the predicted quantity of the above sample region at the sample time point is obtained.

[0130] In one embodiment, the multimodal information processing module 1002 is configured to perform the following operations:

[0131] Obtain risk level reference information, which includes at least two infection risk levels and the density range corresponding to each infection risk level.

[0132] Based on the predicted number of targets and the area of ​​the target region, determine the predicted density corresponding to the target region.

[0133] Determine the density range in which the predicted density is located from the aforementioned risk level reference information;

[0134] The infection risk level pointed to by the density interval where the predicted density is located is determined as the infection risk level of the target area.

[0135] The apparatus portion of the embodiments in this application is based on the same inventive concept as the method embodiment, and will not be described in detail here.

[0136] Furthermore, Figure 11 A schematic diagram of a hardware structure for implementing the method provided in the embodiments of this application is shown. This device can participate in or include the apparatus or system provided in the embodiments of this application. Figure 11As shown, device 10 may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 11 The structure shown is for illustrative purposes only and does not limit the structure of the electronic device described above. For example, device 10 may also include a... Figure 11 The more or fewer components shown, or having the same Figure 11 The different configurations shown.

[0137] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuitry are generally referred to herein as "data processing circuitry". This data processing circuitry may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be wholly or partially integrated into any other element within device 10 (or mobile device). As involved in the embodiments of this application, this data processing circuitry serves as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0138] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method described in the embodiments of this application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the above-described method for predicting the number of target objects. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the device 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0139] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of device 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a radio frequency (RF) module used for wireless communication with the Internet.

[0140] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of device 10 (or mobile device).

[0141] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0142] The embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and server embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0143] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0144] The instructions in the aforementioned storage medium can execute a method for predicting the number of target objects, the method comprising:

[0145] Obtain the regional topology map corresponding to the target area. The regional topology map represents the orientational topology relationship between the target area and each associated area. The associated areas are the areas adjacent to the target area.

[0146] Obtain the number of target objects in the aforementioned target area and each of the aforementioned associated areas within a preset time interval. The target objects possess both infectivity and mobility.

[0147] A visualization information map is generated based on the number of target objects mentioned above. The visualization information map represents the spatial comparison information between the target area and each of the associated areas, as well as the comparison information on the number of target objects between the target area and each of the associated areas. The spatial comparison information includes area comparison information and distribution location comparison information.

[0148] Spatiotemporal features are extracted from the number of the target objects and the topological map of the region to obtain spatiotemporal information.

[0149] Graph feature extraction is performed on the above-mentioned visualized information graph to obtain graph feature information;

[0150] The spatiotemporal information and the graph feature information mentioned above are fused to obtain fused feature information;

[0151] The quantity prediction is performed on the above-mentioned fused feature information to obtain the target quantity of the target object in the target area at the target time, where the target time is the time after the above-mentioned preset time interval.

[0152] In one embodiment, obtaining the visualization information map corresponding to the target area includes:

[0153] Based on the geographical location information of the target area and each of the aforementioned related areas, determine the centroid and area of ​​each of the target area and each of the aforementioned related areas.

[0154] Based on the respective areas and the preset mapping relationship, the scaling values ​​of the target area and each of the associated areas are determined, and the mapping relationship represents the correspondence between the area and the scaling value.

[0155] Render the target region and the associated region based on the above scaling values ​​to obtain a region distribution map;

[0156] The corresponding pattern is rendered at each centroid in the above regional distribution map to obtain the above visualization information map. The area of ​​the pattern corresponding to the centroid is proportional to the number of target objects in the region element corresponding to the centroid. The region element is the target region or the associated region.

[0157] In one embodiment, the above-mentioned rendering of the target region and the associated region according to each of the above-mentioned scaling values ​​to obtain the region distribution map includes: filling the target region and the associated region using a first filling method to obtain the region distribution map.

[0158] The above-mentioned rendering of the corresponding pattern at each centroid in the above-mentioned regional distribution map to obtain the above-mentioned visualization information map includes: filling the pattern corresponding to each centroid using a second filling method to obtain the above-mentioned visualization information map, wherein the above-mentioned second filling method is different from the above-mentioned first filling method.

[0159] In one embodiment, the preset time interval includes at least one window interval, each window interval includes at least two sub-intervals, the number of target objects includes the number of target objects in each region element of each sub-interval, the region element is the target region or the associated region, and the spatiotemporal feature extraction of the number of target objects and the region topology map to obtain spatiotemporal information includes:

[0160] Based on the above region topology map, an adjacency matrix is ​​obtained. The adjacency matrix represents the adjacency relationship and degree of adjacency of each element in the above region.

[0161] Based on the number of target objects mentioned above, the time-series information of the quantity of each of the above-mentioned regional elements in each of the above-mentioned window intervals is obtained;

[0162] For each window, multimodal feature extraction is performed on the adjacency matrix and the corresponding quantity time-series information to obtain first information; multimodal feature extraction is performed on the first information and the adjacency matrix to obtain second information; the quantity time-series information corresponding to the window, the first information and the second information are fused to obtain the spatiotemporal information under the window.

[0163] In one embodiment, the above-mentioned extraction of spatiotemporal features from the number of target objects and the region topology map to obtain spatiotemporal information further includes:

[0164] The spatiotemporal information under each window is encoded using long short-term memory to obtain the aforementioned spatiotemporal information.

[0165] In one embodiment, the above-mentioned graph feature extraction of the above-mentioned visualized information graph to obtain graph feature information includes:

[0166] The above-mentioned visualization infographic is segmented to obtain multiple image sub-blocks;

[0167] Linear radial and positional coding are performed on each of the above image sub-blocks to obtain multiple coding results;

[0168] Information is extracted from the above encoding results based on a self-attention mechanism to obtain the graph feature information.

[0169] In one embodiment, the above method is implemented based on a target quantity prediction model, which includes a first model, a second model, and a third model, and is trained using the following method:

[0170] Obtain the sample region and the number of target objects corresponding to the sample region at the sample time point;

[0171] Determine the sample time interval based on the above sample time points;

[0172] Obtain the sample region topology map corresponding to the above sample regions;

[0173] Obtain the number of target objects in the sample time interval for each of the above sample regions and their associated regions.

[0174] Generate a sample visualization information map based on the number of target objects mentioned above;

[0175] Based on the first model described above, spatiotemporal features are extracted from the number of target objects in the sample and the topological map of the sample region to obtain the spatiotemporal information of the sample.

[0176] Based on the second model described above, graph features are extracted from the above-mentioned sample visualization information graph to obtain sample graph feature information;

[0177] Based on the third model described above, the spatiotemporal information of the samples and the feature information of the sample maps are fused to obtain sample fusion feature information; and the quantity of the sample fusion feature information is predicted to obtain the predicted quantity of the sample region at the sample time point.

[0178] Based on the difference between the predicted quantity and the number of target objects corresponding to the above sample time points, the parameters of the first model, the second model, and the third model are adjusted to obtain the target quantity prediction model.

[0179] In one embodiment, the first model includes a sequentially connected multilayer message-passing neural network and a long short-term memory network, and the third model includes a sequentially connected multiple feature connection layers, a fully connected layer, and an information output layer. Based on the first model, spatiotemporal feature extraction is performed on the number of target objects in the sample and the topological map of the sample region to obtain the spatiotemporal information of the sample, including:

[0180] Based on a multi-layer message passing neural network, spatial feature extraction is performed on the number of target objects in the above samples and the topology of the sample region to obtain spatial feature extraction results.

[0181] Based on the Long Short-Term Memory network, the spatial feature extraction results above are subjected to time-based feature extraction to obtain the spatiotemporal information of the above samples.

[0182] The above-mentioned sample spatiotemporal information and sample graph feature information are fused based on the third model to obtain sample fusion feature information, including: inputting the above-mentioned sample spatiotemporal information and sample graph feature information into multiple feature connection layers that are sequentially connected to obtain the above-mentioned sample fusion feature information;

[0183] The above-mentioned quantity prediction of the sample fusion feature information to obtain the predicted quantity of the sample region at the sample time point includes: inputting the sample fusion feature information into a fully connected layer and an information output layer in sequence to obtain the predicted quantity of the sample region at the sample time point.

[0184] In one embodiment, the above method further includes:

[0185] Obtain risk level reference information, which includes at least two infection risk levels and the density range corresponding to each infection risk level.

[0186] Based on the predicted number of targets and the area of ​​the target region, determine the predicted density corresponding to the target region.

[0187] Determine the density range in which the predicted density is located from the aforementioned risk level reference information;

[0188] The infection risk level pointed to by the density interval where the predicted density is located is determined as the infection risk level of the target area.

[0189] The above are merely preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present application should be included within the protection scope of the present application.

Claims

1. A method for predicting the number of target objects, characterized in that, The method includes: Obtain a regional topology map corresponding to the target region. The regional topology map represents the orientational topology relationship between the target region and each associated region. The associated regions are regions adjacent to the target region. The number of target objects in the target area and each of the associated areas within a preset time interval is obtained, and the target objects have the ability to spread and move. Based on the geographical location information of the target area and each of the associated areas, the centroid and area of ​​each of the target area and each of the associated areas are determined; based on the area and a preset mapping relationship, the scaling value of each of the target area and each of the associated areas is determined, where the mapping relationship represents the correspondence between area and scaling value; the target area and each of the associated areas are rendered according to the scaling value to obtain a regional distribution map; corresponding patterns are rendered at each centroid in the regional distribution map to obtain a visualization information map, where the area of ​​the pattern corresponding to the centroid is proportional to the number of target objects in the region element corresponding to the centroid, where the region element is the target area or the associated area; the visualization information map represents the spatial comparison information between the target area and each of the associated areas, as well as the comparison information on the number of target objects between the target area and each of the associated areas, where the spatial comparison information includes area comparison information and distribution location comparison information; For the number of target objects and the region topology map, a specific architecture based on message passing neural network multi-layer feature extraction combined with long short-term memory network encoding is used to extract spatiotemporal features and obtain spatiotemporal information. The visualized information graph is segmented to obtain multiple image sub-blocks; each image sub-block is subjected to linear radialization and positional encoding to obtain multiple encoding results; information extraction based on a self-attention mechanism is performed on each encoding result to obtain graph feature information; linear radialization, positional encoding, and information extraction based on a self-attention mechanism are all implemented using VIT combined with Transformer; The spatiotemporal information and the graph feature information are fused to obtain fused feature information; The quantity prediction is performed on the fused feature information to obtain the target quantity of the target object in the target region at the target time, where the target time is the time after the preset time interval.

2. The method according to claim 1, characterized in that, The step of rendering the target region and the associated regions according to the scaling values ​​to obtain a region distribution map includes: filling the target region and the associated regions using a first filling method to obtain the region distribution map; The step of rendering the corresponding pattern at each centroid in the regional distribution map to obtain the visualization information map includes: filling the pattern corresponding to each centroid with a second filling method to obtain the visualization information map, wherein the second filling method is different from the first filling method.

3. The method according to claim 1 or 2, characterized in that, The preset time interval includes at least one window interval, each window interval includes at least two sub-intervals, the number of target objects includes the number of target objects in each region element of each sub-interval, and the region element is the target region or the associated region. The step of extracting spatiotemporal features from the number of target objects and the region topology map to obtain spatiotemporal information includes: An adjacency matrix is ​​obtained based on the region topology graph. The adjacency matrix represents the adjacency relationship and degree of adjacency of each region element. Based on the number of target objects, obtain the time-series information of the quantity of each region element in each window interval; For each window, multimodal feature extraction is performed on the adjacency matrix and the corresponding quantity time-series information to obtain first information; multimodal feature extraction is performed based on the first information and the adjacency matrix to obtain second information; the quantity time-series information corresponding to the window, the first information, and the second information are fused to obtain the spatiotemporal information under the window.

4. The method according to claim 3, characterized in that, The step of extracting spatiotemporal features from the number of target objects and the region topology map to obtain spatiotemporal information further includes: The spatiotemporal information under each window is encoded using long short-term memory to obtain the spatiotemporal information.

5. The method according to claim 1, characterized in that, The method is implemented based on a target quantity prediction model, which includes a first model, a second model, and a third model. The target quantity prediction model is trained using the following method: Obtain the sample region and the number of target objects corresponding to the sample region at the sample time point; Determine the sample time interval based on the sample time points; Obtain the sample region topology map corresponding to the sample region; Obtain the number of target objects in the sample area and each associated sample area within the sample time interval; Generate a sample visualization information map based on the number of target objects in the sample; Based on the first model, spatiotemporal features are extracted from the number of target objects in the sample and the topology of the sample region to obtain the spatiotemporal information of the sample. Based on the second model, graph features are extracted from the sample visualization information graph to obtain sample graph feature information; Based on the third model, the spatiotemporal information of the sample and the feature information of the sample map are fused to obtain the sample fusion feature information; And perform quantity prediction on the sample fusion feature information to obtain the predicted quantity of the sample region at the sample time point; Based on the difference between the predicted quantity and the number of target objects corresponding to the sample time point, the parameters of the first model, the second model, and the third model are adjusted to obtain the target quantity prediction model.

6. The method according to claim 5, characterized in that, The first model includes a sequentially connected multi-layer message-passing neural network and a long short-term memory network. The third model includes a sequentially connected multiple feature connection layers, a fully connected layer, and an information output layer. The step of extracting spatiotemporal features from the number of target objects in the sample and the topological map of the sample region based on the first model to obtain the spatiotemporal information of the sample includes: Based on the multi-layer message passing neural network, spatial feature extraction is performed on the number of target objects in the sample and the topology of the sample region to obtain spatial feature extraction results. Based on the Long Short-Term Memory Network, the spatial feature extraction results are subjected to time-based feature extraction to obtain the sample spatiotemporal information; The step of fusing the sample spatiotemporal information and the sample graph feature information based on the third model to obtain sample fusion feature information includes: inputting the sample spatiotemporal information and the sample graph feature information into the multiple feature connection layers that are sequentially connected to obtain the sample fusion feature information; The step of predicting the number of samples based on the sample fusion feature information to obtain the predicted number of the sample region at the sample time point includes: inputting the sample fusion feature information into the fully connected layer and the information output layer that are sequentially connected to obtain the predicted number of the sample region at the sample time point.

7. The method according to claim 1, characterized in that, The method further includes: Obtain risk level reference information, which includes at least two infection risk levels and a density range corresponding to each infection risk level; Based on the predicted number of targets and the area of ​​the target region, the predicted density corresponding to the target region is determined; The density range in which the predicted density is located is determined from the risk level reference information; The infection risk level pointed to by the density interval where the predicted density is located is determined as the infection risk level of the target area.

8. A target object quantity prediction device, characterized in that, The device includes: A multimodal information acquisition module is used to acquire a regional topology map corresponding to a target area, wherein the regional topology map represents the directional topological relationship between the target area and each associated area, and the associated areas are areas adjacent to the target area; acquire the number of target objects in the target area and each of the associated areas within a preset time interval, wherein the target objects have infectivity and mobility; and determine the centroid and area of ​​the target area and each of the associated areas based on their respective geographical location information; and determine the scaling value of the target area and each of the associated areas based on their respective areas and a preset mapping relationship. The mapping relationship represents the correspondence between area and scaling value; the target area and the associated areas are rendered according to the scaling value to obtain a region distribution map; a corresponding pattern is rendered at each centroid in the region distribution map to obtain a visualization information map, wherein the area of ​​the pattern corresponding to the centroid is proportional to the number of target objects in the region element corresponding to the centroid, and the region element is the target area or the associated area. The visualization information map represents the spatial comparison information between the target area and the associated areas, as well as the comparison information of the number of target objects between the target area and the associated areas. The spatial comparison information includes area comparison information and distribution location comparison information. The multimodal information processing module is used to extract spatiotemporal features from the target object quantity and the region topology map using a specific architecture combining multi-layer feature extraction based on message passing neural networks and long short-term memory network encoding to obtain spatiotemporal information; to perform image segmentation on the visualized information map to obtain multiple image sub-blocks; to perform linear radialization and positional encoding on each image sub-block to obtain multiple encoding results; to perform information extraction based on a self-attention mechanism on each encoding result to obtain graph feature information; linear radialization, positional encoding, and information extraction based on a self-attention mechanism are all implemented using VIT combined with Transformer; to fuse the spatiotemporal information and the graph feature information to obtain fused feature information; and to perform quantity prediction on the fused feature information to obtain the target quantity of the target object in the target region at a target time, where the target time is a time after the preset time interval.

9. The apparatus according to claim 8, characterized in that, The multimodal information acquisition module is used to perform the following operations: The target region and each of the associated regions are filled using a first filling method to obtain the region distribution map; The pattern corresponding to each centroid is filled using a second filling method to obtain the visualization information map. The second filling method is different from the first filling method.

10. The apparatus according to claim 8 or 9, characterized in that, The preset time interval includes at least one window interval, each window interval includes at least two sub-intervals, the number of target objects includes the number of target objects in each region element of each sub-interval, and the region element is the target region or the associated region. The multimodal information processing module is used to perform the following operations: An adjacency matrix is ​​obtained based on the region topology graph. The adjacency matrix represents the adjacency relationship and degree of adjacency of each region element. Based on the number of target objects, obtain the time-series information of the quantity of each region element in each window interval; For each window, multimodal feature extraction is performed on the adjacency matrix and the corresponding quantity time-series information to obtain first information; multimodal feature extraction is performed based on the first information and the adjacency matrix to obtain second information; By integrating the quantity time-series information corresponding to the window, the first information, and the second information, the spatiotemporal information under the window is obtained.

11. The apparatus according to claim 10, characterized in that, The multimodal information processing module is used to perform the following operation: encode the spatiotemporal information under each window using long short-term memory to obtain the spatiotemporal information.

12. The apparatus according to claim 8, characterized in that, The device is implemented based on a target quantity prediction model, which includes a first model, a second model, and a third model. The device also includes a training module for performing the following operations: Obtain the sample region and the number of target objects corresponding to the sample region at the sample time point; Determine the sample time interval based on the sample time points; Obtain the sample region topology map corresponding to the sample region; Obtain the number of target objects in the sample area and each associated sample area within the sample time interval; Generate a sample visualization information map based on the number of target objects in the sample; Based on the first model, spatiotemporal features are extracted from the number of target objects in the sample and the topology of the sample region to obtain the spatiotemporal information of the sample. Based on the second model, graph features are extracted from the sample visualization information graph to obtain sample graph feature information; Based on the third model, the spatiotemporal information of the sample and the feature information of the sample map are fused to obtain the sample fusion feature information; And perform quantity prediction on the sample fusion feature information to obtain the predicted quantity of the sample region at the sample time point; Based on the difference between the predicted quantity and the number of target objects corresponding to the sample time point, the parameters of the first model, the second model, and the third model are adjusted to obtain the target quantity prediction model.

13. The apparatus according to claim 12, characterized in that, The first model includes a sequentially connected multi-layer message-passing neural network and a long short-term memory network; the third model includes a sequentially connected multiple feature connection layers, a fully connected layer, and an information output layer; the training module is used to perform the following operations: Based on a multi-layer message passing neural network, spatial feature extraction is performed on the number of target objects in the sample and the topology of the sample region to obtain spatial feature extraction results. Based on the long short-term memory network, the spatial feature extraction results are subjected to time-based feature extraction to obtain the spatiotemporal information of the sample. The sample spatiotemporal information and the sample map feature information are input into multiple feature connection layers that are sequentially connected to obtain the sample fusion feature information; The sample fusion feature information is input into a fully connected layer and an information output layer that are sequentially connected to obtain the predicted number of samples at the sample time point.

14. The apparatus according to claim 8, characterized in that, The multimodal information processing module is used to perform the following operations: Obtain risk level reference information, which includes at least two infection risk levels and a density range corresponding to each infection risk level; Based on the predicted number of targets and the area of ​​the target region, the predicted density corresponding to the target region is determined; The density range in which the predicted density is located is determined from the risk level reference information; The infection risk level pointed to by the density interval where the predicted density is located is determined as the infection risk level of the target area.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction or at least one program, which is loaded and executed by a processor to implement a target object quantity prediction method as described in any one of claims 1 to 7.

16. An electronic device, characterized in that, The method includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the at least one processor implements a target object quantity prediction method as described in any one of claims 1 to 7 by executing the instructions stored in the memory.

17. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, implement a target object quantity prediction method as described in any one of claims 1 to 7.

Images