A method and device for identifying target features, and computer storage medium

Abstract

The present invention discloses a method and device for identifying target features, and a computer storage medium, which relate to the technical field of computer vision and pattern recognition, and are used to identify target features across domains based on target attributes, thereby improving the recognition accuracy of target features. The recognition method comprises: acquiring a target domain image; determining target information according to the target domain image, the target information comprising attribute information and feature information of multiple targets; establishing a target feature relationship graph according to the attribute information and feature information of multiple targets; and performing target feature recognition on the target feature relationship graph using a graph neural network. The device executes the recognition method proposed in the above technical solution. The target feature recognition method provided by the present invention is used in image recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and pattern recognition, and in particular to a method and device for recognizing target features, and a computer storage medium. Background Art

[0002] Person re-identification refers to the technology of accurately identifying specific pedestrians captured by non-overlapping cameras across time and location. Pedestrian re-identification technology can be simply divided into single-domain pedestrian re-identification technology and cross-domain pedestrian re-identification technology. Single-domain pedestrian re-identification refers to the model being trained and tested on the same dataset; while cross-domain pedestrian re-identification uses different datasets for training and testing. Its purpose is to solve the problem of pedestrian re-identification technology being applied from the laboratory to actual unfamiliar environments.

[0003] With the widespread application of deep learning, single-domain person re-identification research has achieved high re-identification accuracy. However, current person re-identification technology still faces considerable difficulties when facing cross-domain problems. Simply applying the model trained in the source domain to the target domain for testing will lead to a sharp drop in pedestrian re-identification performance, limiting the application of cross-domain pedestrian re-identification technology. Summary of the invention

[0004] The object of the present invention is to provide a method and device for identifying target features, and a computer storage medium, so as to identify target features across domains based on target attributes and improve the recognition accuracy of target features.

[0005] In order to achieve the above object, the present invention provides a method for identifying target features, comprising:

[0006] Obtain target domain image;

[0007] Determining target information according to the target domain image, wherein the target information includes attribute information and feature information of multiple targets;

[0008] Establishing a target feature relationship graph based on the attribute information and feature information of the plurality of targets;

[0009] A graph neural network is used to perform target feature recognition on the target feature relationship graph.

[0010] Optionally, determining target information according to the target domain image includes:

[0011] Processing the target domain image using the first target recognition model to obtain feature information of multiple targets;

[0012] The target image information is processed using the second target recognition model to obtain attribute information of multiple targets.

[0013] Optionally, the first target recognition model is a source domain target recognition model; the source domain of the source domain target recognition model and the target domain of the target domain image are the same area or different areas.

[0014] Optionally, the training method of the first target recognition model is a supervised training method or a semi-supervised training method; and / or,

[0015] The loss function of the first target recognition model includes a joint cross entropy loss function and / or a triple loss function; and / or,

[0016] The second target recognition model is a YOLOv3 attribute recognition algorithm or a Knn attribute recognition algorithm.

[0017] Optionally, establishing a target feature relationship graph according to the attribute information and feature information of the plurality of targets includes:

[0018] Determining attribute similarities of the two objects according to the attribute information of the two objects;

[0019] When the attribute similarity between the two targets meets the feature association condition, the feature association information of the two targets is added to the feature information of the multiple targets to obtain a target feature relationship graph.

[0020] Optionally, the attribute information of each target is an attribute vector of the target; wherein,

[0021] The attribute similarity is cosine similarity cos(θ), and the feature association condition is cos(θ)≥threshold, -1<threshold<1; or,

[0022] The attribute similarity is the number n of matching elements in the attribute vectors of the two targets, and the feature association condition is that the number n of matching elements in the attribute vectors of the two targets is equal to k, N / 2<k≤N, and N is the total number N of elements in the attribute vectors of the targets.

[0023] Optionally, the training set of the graph neural network is a target feature relationship graph based on the target domain.

[0024] Optionally, the type of loss function of the training set of the graph neural network is Euclidean distance or cosine distance.

[0025] Compared with the prior art, in the target feature recognition method provided by the present invention, no matter what model is used to determine the attribute information and feature information of multiple targets, a target feature relationship graph can be established based on the attribute information and feature information of multiple targets, so that the target feature relationship graph is associated with the attribute information of each target. Based on this, when the target feature relationship graph is used to identify the target features using the graph neural network, the feature correlation degree contained in the feature information of each target obtained is relatively high, thereby ensuring the accuracy of target feature recognition. It can be seen that when the target feature recognition method provided by the present invention is applied to cross-domain recognition, the source domain recognition model can be used to identify the target features of the target domain, and the detection object (target feature relationship graph) can be reconstructed in combination with the target feature similarity to ensure the accuracy of identifying the target features in the graph neural network using the graph neural network, thereby improving the application scope of cross-domain pedestrian re-identification technology.

[0026] The present invention also provides a target feature recognition device, characterized in that it includes a processor and a communication interface coupled to the processor; the processor is used to run a computer program or instruction to implement the target feature recognition method described in the above technical solution.

[0027] Compared with the prior art, the beneficial effects of the target feature recognition device provided by the present invention are the same as the beneficial effects of the above-mentioned target feature recognition method, which will not be elaborated here.

[0028] The present invention also provides a computer storage medium, characterized in that instructions are stored in the computer storage medium, and when the instructions are executed, the target feature recognition method described in the above technical solution is implemented.

[0029] Compared with the prior art, the beneficial effects of the computer storage medium provided by the present invention are the same as the beneficial effects of the above-mentioned target feature recognition method, which will not be described in detail here. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0031] Figure 1 Application scenarios of the target feature recognition method provided by the embodiment of the present invention;

[0032] Figure 2 A flow chart of a method for identifying target features provided by an embodiment of the present invention;

[0033] Figure 3 A schematic diagram is established for a target feature relationship diagram in an embodiment of the present invention;

[0034] Figure 4A structural block diagram of a target feature recognition device provided by an embodiment of the present invention;

[0035] Figure 5 A schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention;

[0036] Figure 6 A schematic diagram of the structure of a chip provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, words such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and their order is not limited. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit them to be different.

[0038] It should be noted that, in the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0039] In the present invention, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b, c can be single or multiple.

[0040] The embodiment of the present invention provides a method for identifying target features. The target features that can be identified can be features of active targets and pedestrian features of stationary targets. Active targets can be pedestrians, vehicles, animals, etc. Stationary targets can be power facilities, buildings, etc. The following example describes the method for identifying target features provided by the embodiment of the present invention. All networks based on the recognition method can be built using the deep learning PyTorch framework.

[0041] Figure 1 The following examples illustrate the application scenarios of the target feature recognition method provided by the embodiment of the present invention. Figure 1 As shown, the application scenario takes a campus as the target domain, and multiple cameras 100 are arranged in the campus, and these cameras communicate with the monitoring platform 200 remotely. The monitoring platform 100 may include not only a monitor with a display screen, but also a host 220 and a gateway 230 of the monitor. Multiple cameras 100 can transmit the captured images to the host 220 through the gateway 230. The host 220 can recognize the image and display the recognition result on the display screen for users to understand the actual situation of the campus. The communication method here can choose LAN wireless communication such as wifi zigbee, or 3G, 4G or 5G communication based on mobile communication technology. Of course, optical fiber or power line carrier technology can also be selected for wired communication.

[0042] The target feature recognition method provided in the embodiment of the present invention can be executed by an electronic device or a chip in the electronic device. The following embodiment is illustrated by taking the monitoring platform as the execution subject.

[0043] Figure 2 The following is a flow chart illustrating a method for identifying target features provided by an embodiment of the present invention. Figure 2 As shown, the target feature recognition method provided by the embodiment of the present invention includes:

[0044] Step 101: The monitoring platform acquires a target domain image. The target domain image can be acquired by a camera, and can be a grayscale image or a color image. The target domain image acquired by the camera can be received by the host through the gateway, and the host in the monitoring platform performs target feature recognition on the target domain image.

[0045] Step 102: The monitoring platform determines target information based on the target domain image. The target information includes attribute information and feature information of multiple targets. The attribute information and feature information can be output by the same convolutional neural network or by different convolutional neural networks.

[0046] In practical applications, the monitoring platform can use the first target recognition model to process the target domain image to obtain the feature information of multiple targets; and use the second target recognition model to process the target image information to obtain the attribute information of multiple targets.

[0047] The first target recognition model is a source domain target recognition model. Here, the source domain is defined as the training set acquisition area of ​​the first target recognition model. The target domain refers to the acquisition area of ​​the target domain image. The source domain and the target domain can be the same area or different areas.

[0048] For example, when the source domain and the target domain are in the same area, using the source domain target recognition model to recognize the target domain image can greatly improve the accuracy of pedestrian re-identification, thereby improving the recognition accuracy of subsequent pedestrian features.

[0049] When the source domain and the target domain are different regions, the training set collection region of the first object recognition model is different from the collection region of the target domain image. Taking Xi'an as an example, Weiyang District can be the source domain and Yanta District can be the target domain.

[0050] In one example, the training method of the first target recognition model may be a supervised training method or a semi-supervised training method. The second target recognition model may determine the target attribute by using a YOLOv3 attribute recognition algorithm or a Knn attribute recognition algorithm.

[0051] For example, when the training method of the first target recognition model can be a supervised training method, a convolutional neural network (CNN) is used to train the source domain pedestrian data by combining the cross entropy loss function and the triplet loss function. Here, the convolutional neural network can be ResNet50_IBN_a, which has better generalization performance than ResNet50. Based on this, after processing the target domain image using the trained ResNet50_IBN_a model, the obtained pedestrian feature accuracy is relatively high.

[0052] At the same time, during the training of source domain pedestrian data, you can also use the Warmup learning rate preset method, label smoothing regularization method, set the Last Stride (the step size of the last layer of convolution kernel of each convolution block) to 1, add BNNeck (batch normalization) layers and other methods to make the cross-domain generalization of the convolutional neural network model better.

[0053] Step 103: The monitoring platform establishes a target feature relationship graph based on the attribute information and feature information of multiple targets. Specifically, the monitoring platform determines the attribute similarity of two targets based on the attribute information of two targets; when the attribute similarity of two targets meets the feature association condition, the feature association information of the two targets is added to the feature information of multiple targets to obtain a target feature relationship graph. At this time, the attribute information of each target is used as a node, and the similarity of the two nodes is calculated. When the similarity meets the feature association condition, an edge is established between the two nodes so that there is a feature association between the two nodes, and a target feature relationship graph is obtained.

[0054] Inside the computer, the attribute information and feature information of each target can be expressed in the form of a matrix or vector. Assuming that these targets are pedestrians, the attribute information of a pedestrian can include the pedestrian's gender, clothing color, shoe type, etc. At this time, when the attribute information of the pedestrian is expressed in the form of a vector, each element in the vector represents an attribute. The feature information of the pedestrian can include the pedestrian's fatness, height, etc. At this time, when the attribute information of the pedestrian is expressed in the form of a vector, each element in the vector represents a feature.

[0055] Specifically, the attribute information of each target is the attribute vector of the target, and the attribute similarity can be expressed by cosine similarity cos(θ). The feature association condition is cos(θ)≥threshold, -1<threshold<1. For cosine similarity cos(θ), its expression can be A is the attribute vector of one target, and B is the attribute vector of another target. When the threshold is 0.60-0.85, the similarity detection result is better.

[0056] Figure 3 A schematic diagram is established for the target feature relationship diagram in the embodiment of the present invention. Figure 3 As shown in the figure, if each pedestrian has n attribute annotations, the attribute corresponding to each pedestrian is an n-dimensional attribute vector. For example, there are 23 pedestrian attribute annotations in ten categories in the standard dataset DukeMTMC-attribute, so n=23. The attribute annotations of this standard dataset are binary attribute annotations. The binary value here is +1 or -1. At this time, the attribute vector of each pedestrian is a 23-dimensional attribute vector, in which the vector elements are +1 or -1.

[0057] like Figure 3 As shown, when cos(θ)≥threshold, it indicates that there is an edge connection between the feature node of pedestrian a and the feature node of pedestrian b, and an edge connection is established between the node of pedestrian a and the node of pedestrian b, and the edge connection can be represented in the form of a data matrix.

[0058] Of course, the attribute similarity can also be the number of matching elements n in the attribute vectors of the two targets. In this case, the feature association condition is that the number of matching elements n in the attribute vectors of the two targets is equal to k, N / 2<k≤N, and N is the total number of elements N in the attribute vectors of the targets.

[0059] Taking cross-domain pedestrian recognition as an example, if the attribute vector of the pedestrian is a 3D attribute vector, for example, the attribute vector of pedestrian a is: The attribute vector of pedestrian b is There are 2 elements matching in the attribute vector of pedestrian a and the attribute vector of pedestrian b.

[0060] In one case, the feature association condition is that the number of matching elements in the attribute vectors of the two pedestrians is greater than or equal to the total number of elements in the attribute vector of the target, which is 3. Since the number of matching elements in the attribute vectors of the two pedestrians is less than the total number of elements in the attribute vector of the target, which is 3, it means that the similarity between the attribute vector of pedestrian a and the attribute vector of pedestrian b does not meet the feature association condition.

[0061] In another case, the feature association condition is that the number of matching elements in the attribute vectors of the two pedestrians is greater than or equal to the total number of elements in the attribute vector of the target, which is 2. Since the number of matching elements in the attribute vectors of the two pedestrians is equal to 2, it means that the similarity between the attribute vector of pedestrian a and the attribute vector of pedestrian b meets the feature association condition, and therefore, an edge can be established between the feature information of pedestrian a and the feature information of pedestrian b.

[0062] As can be seen from the above, the above target feature relationship diagram essentially adds the relationship information between the attribute information of the targets to the attribute information of the determined multiple targets.

[0063] Step 104: The monitoring platform uses a graph neural network to perform target feature recognition on the target feature relationship graph.

[0064] For targets with complex attributes such as pedestrians, the target has many attribute elements. When determining the attribute similarity between two targets, the attribute differences between the two targets will be relatively large. Therefore, the target feature relationship graph established by the monitoring platform based on the attribute information and feature information of multiple targets can fully reflect the feature graph based on the similarity between target attributes. On this basis, when the monitoring platform uses the graph neural network to perform target feature recognition on the target feature relationship graph, the target features with relatively high attribute complexity can be improved. The dimension of the attribute vector of the target with relatively high attribute complexity defined here can be determined according to actual needs. For example, the dimension can be at least 10 dimensions.

[0065] For targets with fewer attributes, such as cars, the target has fewer attribute elements. When determining the attribute similarity between two targets, the attribute differences between the two targets are not large, resulting in the target feature relationship graph being unable to fully reflect the differences between the targets. On this basis, when the monitoring platform uses the graph neural network to identify target features on the target feature relationship graph, the recognition accuracy of target features with relatively low attribute complexity is not high.

[0066] The training set of the above-mentioned graph neural network is a target feature relationship graph based on the target domain. This target feature relationship graph of the target domain can be established in the manner of step 103. During the establishment process, the feature information of the target used can be output by the source domain target recognition model or the source domain convolutional neural network. The training process of the graph neural network is described in detail below.

[0067] Use the ResNet50_IBN_a convolutional neural network, and combine the cross entropy loss function and the triplet loss function to train the source domain pedestrian data to obtain the source domain convolutional neural network. Input the query data (query) and gallery data (gallery) as the target domain data into the source domain convolutional neural network to extract the feature information of multiple pedestrians in the target domain. Each pedestrian feature information exists in the form of a feature vector. Take each pedestrian feature information as a node, and determine the similarity between the two through the attribute information of each pair of pedestrians. The similarity determination method can refer to step 103, which will not be described in detail here.

[0068] When the similarity of the attribute information of two pedestrians meets the feature association condition, an edge is established between the feature information of the two pedestrians to associate the two and obtain a target feature relationship graph based on the pedestrian attribute information. The graph neural network selects GraphSAGE graph neural network, which has 3 layers. The target feature relationship graph is input into the GraphSAGE graph neural network, and trained in GraphSAGE based on the graph unsupervised loss algorithm. After 20 iterations, multiple pedestrian features are obtained.

[0069] In order to test the training results, the distance between each pedestrian feature output by the GraphSAGE neural network and the features of each pedestrian in the query data can be calculated. Of course, the k-loop distance re-ranking method is added here to return the retrieval results according to the distance sorting, thereby improving the re-identification accuracy.

[0070] At the same time, since the distance between each pedestrian feature and the features of each pedestrian in the query data (query) is calculated to evaluate its recognition accuracy, the loss function type of the training set of the graph neural network is Euclidean distance or cosine distance or other distance metrics.

[0071] As can be seen from the above, in the target feature recognition method provided by the embodiment of the present invention, no matter what model is used to determine the attribute information and feature information of multiple targets, a target feature relationship graph can be established based on the attribute information and feature information of multiple targets, so that the target feature relationship graph is associated with the attribute information of each target. Based on this, when the target feature relationship graph is used for target feature recognition using a graph neural network, the feature correlation degree contained in the feature information obtained for each target is relatively high, thereby ensuring the accuracy of target feature recognition. It can be seen that when the target feature recognition method provided by the embodiment of the present invention is applied to cross-domain recognition, the source domain recognition model can be used to identify the target features of the target domain, and the detection object (target feature relationship graph) can be reconstructed in combination with the target feature similarity to ensure the accuracy of identifying the target features in the graph neural network using the graph neural network, thereby improving the application scope of cross-domain pedestrian re-identification technology.

[0072] The above mainly introduces the solution provided by the embodiment of the present invention from the perspective of the monitoring platform. It can be understood that in order to realize the above functions, the monitoring platform includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present invention can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0073] The embodiment of the present invention can divide the functional modules of the monitoring platform according to the above method example. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present invention is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0074] In the case of using the corresponding integrated unit, Figure 4 FIG. 2 is a block diagram showing a target feature recognition device according to an embodiment of the present invention. Figure 4 As shown, the target feature recognition device can be Figure 1 The monitoring platform shown may also be a chip applied to the monitoring platform. The target feature recognition device 300 includes: a communication unit and a processing unit.

[0075] The communication unit 301 is used to support the monitoring platform to execute step 101 of the target feature recognition method in the above embodiment.

[0076] The processing unit 302 is used to support the monitoring platform to execute steps 102 to 104 of the target feature recognition method in the above embodiment.

[0077] In some possible implementations, the processing unit 302 is used to process the target domain image using the first target recognition model to obtain feature information of multiple targets; and to process the target image information using the second target recognition model to obtain attribute information of multiple targets.

[0078] In some possible implementations, the processing unit 302 is used to determine the attribute similarity of two targets according to the attribute information of two targets; when the attribute similarity of two targets meets the feature association condition, the feature association information of the two targets is added to the feature information of multiple targets to obtain a target feature relationship graph.

[0079] In some possible implementations, the above-mentioned target feature recognition device may further include a storage unit 303 for storing program codes and data of the base station.

[0080] Among them, the processing unit 302 can be a processor or a controller, for example, a central processing unit (CPU), a general processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. The processor can also be a combination that implements a computing function, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The communication unit can be a transceiver, a transceiver circuit or a communication interface, and the like. The storage unit can be a memory.

[0081] When the processing unit 302 is a processor, the communication unit 301 is a transceiver, and the storage unit 303 is a memory, the electronic device involved in the embodiment of the present invention can be Figure 5 Electronic equipment shown.

[0082] Figure 5 FIG. 1 is a schematic diagram showing the hardware structure of an electronic device provided by an embodiment of the present invention. Figure 5 As shown, the electronic device 400 includes a processor 410 and a communication interface 420 .

[0083] like Figure 5 As shown, the processor 410 may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the present invention. The communication interface may be one or more. The communication interface 420 may use any transceiver-like device for communicating with other devices or communication networks.

[0084] like Figure 5 As shown, the electronic device may further include a communication line 440. The communication line 440 may include a path for transmitting information between the components.

[0085] Optional, such as Figure 5 As shown, the electronic device may further include a memory 430. The memory 430 is used to store computer-executable instructions for executing the solution of the present invention, and is controlled by the processor to execute. The processor 410 is used to execute the computer-executable instructions stored in the memory, thereby implementing the method provided by the embodiment of the present invention.

[0086] like Figure 5 As shown, the memory 430 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 430 may exist independently and be connected to the processor 410 via a communication line 440. The memory 430 may also be integrated with the processor 410.

[0087] Optionally, the computer-executable instructions in the embodiment of the present invention may also be referred to as application program codes, which is not specifically limited in the embodiment of the present invention.

[0088] In a specific implementation, as an example, Figure 5 As shown, the processor 410 may include one or more CPUs, such as Figure 5 CPU0 and CPU1 in.

[0089] In a specific implementation, as an example, Figure 5 As shown, the electronic device 400 may include multiple processors, such as Figure 5 The processor 410 and the processor 450 in the embodiment of the present invention are shown in FIG. Each of these processors may be a single-core processor or a multi-core processor.

[0090] Figure 6 Schematic diagram of the structure of the chip provided by the embodiment of the present invention. Figure 6 As shown, the chip 500 includes one or more (including two) processors 510 and a communication interface 520 .

[0091] Optional, such as Figure 6 As shown, the chip 500 also includes a memory 530, which may include a read-only memory and a random access memory, and provides operation instructions and data to the processor 510. A portion of the memory 530 may also include a non-volatile random access memory (NVRAM).

[0092] In some embodiments, Figure 6 As shown, the memory 530 stores the following elements, execution modules or data structures, or their subsets, or their extended sets.

[0093] In the embodiment of the present invention, Figure 6 As shown, the corresponding operation is performed by calling the operation instruction stored in the memory 530 (the operation instruction may be stored in the operating system).

[0094] like Figure 6 As shown, the processor 510 controls the processing operations of any one of the terminal devices, and the processor may also be referred to as a central processing unit (CPU).

[0095] like Figure 6 As shown, the memory 530 may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory 530 may also include an NVRAM. For example, in an application, the memory, the communication interface, and the memory are coupled together through a bus system, wherein the bus system may include a power bus, a control bus, and a status signal bus in addition to a data bus. However, for the sake of clarity, in Figure 6 Various buses are labeled as bus system 540 .

[0096] The method disclosed in the above embodiment of the present invention can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general-purpose processor, a digital signal processor (digital signal processing, DSP), an ASIC, a field-programmable gate array (field-programmable gate array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0097] In one possible implementation, Figure 6 As shown, the communication interface 520 is used to execute Figure 2 The receiving step of step 101 in the embodiment shown. The processor 510 is used to execute Figure 2 The processing steps of step 102 to step 104 in the embodiment shown.

[0098] On the one hand, a computer-readable storage medium is provided, in which instructions are stored. When the instructions are executed, the functions performed by the monitoring platform in the above embodiment are implemented.

[0099] On the one hand, a chip is provided, which is applied to a terminal device. The chip includes at least one processor and a communication interface. The communication interface is coupled to at least one processor, and the processor is used to run instructions to implement the functions performed by the monitoring platform in the above embodiment.

[0100] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instruction is loaded and executed on a computer, the process or function described in the embodiment of the present invention is executed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal, a user device or other programmable device. The computer program or instruction can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer program or instruction can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, a hard disk, or a tape; it can also be an optical medium, such as a digital video disc (DVD); it can also be a semiconductor medium, such as a solid state drive (SSD).

[0101] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "one" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0102] Although the present invention has been described in conjunction with specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present invention. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present invention as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present invention. Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, the present invention is intended to include such modifications and variations if they fall within the scope of the claims of the present invention and their equivalents.

Claims

1. A method for identifying target features, characterized in that: include: Obtain target domain image; Determining target information according to the target domain image, wherein the target information includes attribute information and feature information of multiple targets; Establishing a target feature relationship graph based on the attribute information and feature information of the plurality of targets; Using a graph neural network to perform target feature recognition on the target feature relationship graph; The determining of target information according to the target domain image includes: using a first target recognition model to process the target domain image to obtain feature information of multiple targets; using a second target recognition model to process target image information to obtain attribute information of multiple targets; the first target recognition model is a source domain target recognition model; the source domain of the source domain target recognition model and the target domain of the target domain image are the same area or different areas; The target feature relationship diagram is established based on the attribute information and feature information of the multiple targets, including: determining the attribute similarity of the two targets based on the attribute information of the two targets; when the attribute similarity of the two targets meets the feature association condition, adding the feature association information of the two targets to the feature information of the multiple targets to obtain the target feature relationship diagram; the attribute information of each target is the attribute vector of the target; wherein the attribute similarity is the number n of element matches in the attribute vectors of the two targets, and the feature association condition is that the number n of element matches in the attribute vectors of the two targets is equal to k, N / 2<k≤N, and N is the total number N of elements in the attribute vectors of the targets; or, the attribute similarity is cosine similarity , the feature association condition is , .

2. The target feature recognition method according to claim 1, characterized in that: The training method of the first target recognition model is a supervised training method or a semi-supervised training method; and / or, The loss function of the first target recognition model includes a cross entropy loss function and / or a triple loss function; and / or, The second target recognition model is a YOLOv3 attribute recognition algorithm or a Knn attribute recognition algorithm.

3. The target feature recognition method according to claim 1 or 2, characterized in that: The training set of the graph neural network is a target feature relationship graph based on the target domain.

4. The target feature recognition method according to claim 1 or 2, characterized in that: The type of loss function of the training set of the graph neural network is Euclidean distance or cosine distance.

5. A target feature recognition device, characterized in that: It comprises a processor and a communication interface coupled to the processor; the processor is used to run a computer program or instruction to implement the target feature recognition method as claimed in any one of claims 1 to 4.

6. A computer storage medium, characterized in that: The computer storage medium stores instructions, and when the instructions are executed, the target feature recognition method according to any one of claims 1 to 4 is implemented.

Images