Procambarus clarkii individual identification method and system based on biological characteristics

Abstract

The present disclosure provides a method and system for identifying individual procambarus clarkii of biological characteristics, relating to the technical field of deep learning, comprising: extracting local and global digital biological characteristics of each slice image and a target image, and fusing the local and global digital biological characteristics to obtain a digital biological characteristic matrix; inputting the digital biological characteristic matrix into an improved graph neural network model, generating a query and a key through an independent multi-layer perceptron, calculating edge weights through a scaling dot product of the query-key, and obtaining multiple node features after multi-head attention splicing; splicing different node features, generating an attention score for each node based on a gating mechanism aggregation module, and outputting a global representation vector of each group of point sets after global pooling operation; mapping the global representation vector to a fixed dimension through a linear network, and taking it as a final point set embedding, and finally outputting a procambarus clarkii individual identification result.

Description

Technical Field

[0001] This disclosure relates to the field of deep learning technology, specifically to a method and system for individual identification of red swamp crayfish based on biological characteristics. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] Red swamp crayfish (Procambarus clarkii) Procambarus clarkii The red swamp crayfish (Procambarus clarkii), commonly known as the crayfish or Louisiana crayfish, is a freshwater crustacean native to North America. Long-term individual tracking is a core requirement in its breeding research, but traditional physical tagging techniques have significant drawbacks: color tagging, fluorescent tagging, and RFID tagging methods are prone to tag detachment or morphological deformation due to the frequent molting behavior of the red swamp crayfish, and repeated tagging of individuals further reduces their survival rate, making it difficult to meet the needs of long-term tracking.

[0004] At the same time, the application of traditional algorithms in the field of red swamp crayfish identification also has limitations: existing technologies mostly remain at the level of species classification, and rely on morphological features (such as body size and body color), molecular biological features (such as gene sequence) and ecological behavioral features (such as habitat habits and activity patterns) to build identification models. They cannot achieve accurate differentiation of a single red swamp crayfish individual and are difficult to support long-term correlation analysis of key data such as individual growth and reproduction during the breeding process.

[0005] Existing deep learning algorithms are effective in identifying biological features, but they have limitations in recognizing small targets such as key biomarkers in arthropods, which are typically small and densely packed. Specific geometric features in the basal layer of the red swamp crayfish's carapace exhibit in-situ regeneration characteristics after molting, but current computer vision methods do not incorporate this biological characteristic. Current technologies still rely on manually labeled color points, neglecting the lifelong recognition value of naturally occurring features. Furthermore, there are currently no similar identification technologies in the aquaculture field. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes a biological feature-based individual identification method and system for Procambarus clarkii. The method involves imaging the cephalothorax (dorsal or lateral aspect) of the Procambarus clarkii and using an improved graph convolutional neural network model to achieve highly dense and small biological feature detection, enabling stable identification under different growth stages, body colors, and lighting conditions.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] Each preprocessed slice image and the complete target image are input into the biometric feature extraction model, and the local digital biometric features of each slice image and the global digital biometric features of the target image are output. The local digital biometric features of each slice image and the global digital biometric features of the target image are fused together, and after maximum suppression processing, the digital biometric feature matrix of the individual is obtained.

[0009] The individual's digital biometric matrix is ​​input into the improved graph neural network model. In multiple nested edge attention graph convolutional modules, queries and keys are generated by independent multilayer perceptrons. The edge weights are calculated by the scaled dot product of the query and key. Each biometric feature is combined with its k nearest neighbors to obtain the attention coefficient. The node features of the feature are generated by combining the location information of the biometric feature and its attention coefficient.

[0010] Different node features are concatenated, and an aggregation module based on a gating mechanism generates an attention score for each node. After global pooling, the global representation vector of the digital biometric feature matrix is ​​output.

[0011] The global representation vector is mapped to a fixed-dimensional vector through a multilayer perceptron and used as the biometric vector of the individual.

[0012] The biometric vector of the individual to be identified is compared with a preset benchmark feature library. Through similarity calculation and threshold determination, identity verification or identity retrieval is completed, and the final output is the individual identification result of the red swamp crayfish.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] A biometric-based individual identification system for the red swamp crayfish includes:

[0015] The image acquisition module is used to acquire the target image to be identified and to preprocess it to obtain sliced ​​images;

[0016] The feature extraction module is used to input each preprocessed slice image and the complete target image into the biometric feature extraction model, and output the local digital biometric features of each slice image and the global digital biometric features of the target image. The local digital biometric features of each slice image and the global digital biometric features of the target image are fused together, and after maximum suppression processing, the digital biometric feature matrix of the individual is obtained.

[0017] The node feature mining module is used to input the individual's digital biometric matrix into the improved graph neural network model. In multiple nested edge attention graph convolution modules, queries and keys are generated through independent multilayer perceptrons. The edge weights are calculated through the scaled dot product of the query and key. Each biometric feature is combined with its k nearest neighbors to obtain the attention coefficient. The node feature of the feature is generated by combining the location information of the biometric feature and its attention coefficient.

[0018] The aggregation module is used to concatenate different node features, and the aggregation module based on the gating mechanism generates an attention score for each node. After global pooling, it outputs the global representation vector of the digital biometric feature matrix.

[0019] The embedding generation module is used to map the global representation vector into a fixed-dimensional vector through a multilayer perceptron, and use it as the biometric vector of the individual.

[0020] The identification module compares the biometric vector of the individual to be identified with a preset benchmark feature library. Through similarity calculation and threshold determination, it completes identity verification or identity retrieval and finally outputs the individual identification result of the red swamp crayfish.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] A computer program product includes a computer program that, when executed by a processor, implements the biometric-based individual identification method for Procambarus clarkii.

[0023] According to some embodiments, the present disclosure adopts the following technical solutions:

[0024] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned biometric-based individual identification method for Procambarus clarkii.

[0025] According to some embodiments, the present disclosure adopts the following technical solutions:

[0026] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the biometric-based individual identification method for red swamp crayfish.

[0027] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0028] This disclosure presents a biometric-based individual identification method for *Procambarus clarkii*. The selected biometric features include molting without deformation, growth insensitivity, and antibody curvature, making it suitable for long-term tracking in breeding. The biometric extraction model designed in this disclosure provides stable identification of biometric features under different growth stages, body colors, and lighting conditions, demonstrating stability for detecting highly dense and small biometric features. It significantly improves the detection rate of small targets and is particularly suitable for tasks such as biological imaging, insect identification, and lesion detection.

[0029] This disclosed method for individual identification of *Procambarus clarkii* based on biological features combines an edge-information-driven attention mechanism (EdgeGATv2Conv) with a multi-scale fusion strategy for point set embedding to construct an improved graph neural network model, which has stronger representation capabilities and scalability. It is robust to partial loss of biological features and detection noise, with minimal impact; it is simple to operate, requires little manpower, and has extremely fast detection speed, making it well-suited for practical breeding and aquaculture production. Attached Figure Description

[0030] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0031] Figure 1 This is a flowchart of a biometric-based individual identification method for Procambarus clarkii according to an embodiment of the present disclosure;

[0032] Figure 2 This is a diagram of the improved graph neural network model architecture according to an embodiment of the present disclosure;

[0033] Figure 3 This is an architecture diagram of the edge attention graph convolution module according to an embodiment of the present disclosure; Detailed Implementation

[0034] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0035] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0036] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0037] Example 1

[0038] One embodiment of this disclosure provides a method for individual identification of *Procambarus clarkii* based on biological characteristics, the method comprising the following steps:

[0039] Step 1: Obtain the target image to be identified and preprocess it to obtain sliced ​​images;

[0040] Step 2: Input each preprocessed slice image and the complete target image into the biometric feature extraction model, and output the local digital biometric features of each slice image and the global digital biometric features of the target image. Then, fuse the local digital biometric features of each slice image with the global digital biometric features of the target image. After maximum suppression processing, the digital biometric feature matrix of the individual is obtained.

[0041] Step 3: Input the individual's digital biometric matrix into the improved graph neural network model. In multiple nested edge attention graph convolutional modules, queries and keys are generated through independent multilayer perceptrons. Edge weights are calculated through the scaled dot product of query and key. Each biometric feature is combined with its k nearest neighbors to obtain the attention coefficient. The node features of the feature are generated by combining the location information of the biometric feature and its attention coefficient.

[0042] Step 4: Concatenate the different node features, and generate an attention score for each node based on the gating mechanism aggregation module. After global pooling, output the global representation vector of the digital biometric feature matrix.

[0043] Step 5: Map the global representation vector to a fixed-dimensional vector using a multilayer perceptron, and use it as the biometric vector for that individual;

[0044] Step 6: Compare the biometric vector of the individual to be identified with the preset benchmark feature library. Through similarity calculation and threshold determination, complete the identity verification or identity retrieval, and finally output the individual identification result of the red swamp crayfish.

[0045] As one embodiment, the biometric-based individual identification method for Procambarus clarkii disclosed herein effectively captures local and global morphological distributions through biometric extraction and two-dimensional digital feature mapping, thereby achieving individual identification of Procambarus clarkii. The specific implementation process is as follows:

[0046] Step 1: Obtain the target image to be identified and preprocess it to obtain sliced ​​images;

[0047] Specifically, the camera is used to image the shrimp body, specifically the cephalothorax (dorsal or lateral aspect) of the red swamp crayfish. During the acquisition process, it is necessary to ensure a stable light source to avoid glare; ensure that the target is centered in the image; and keep the sample orientation fixed to avoid rotation.

[0048] Further image preprocessing and resolution enhancement are performed:

[0049] High-quality interpolation algorithms (such as LANCZOS4) are used to upscale the image to a specified resolution to ensure that the target is sufficiently visible in subsequent slices.

[0050] The target image is cropped according to the preset slice size (e.g., 800×1330) and overlap ratio (e.g., 20%) to obtain a slice image, which supports automatic adjustment of the slice layout according to the image resolution.

[0051] Each slice retains its starting position index in the original image for easy coordinate reconstruction. Step 2: Input each preprocessed slice image and the complete target image into the biometric extraction model, and output the local digital biometric features of each slice image and the global digital biometric features of the target image. Then, fuse the local digital biometric features of each slice image with the global digital biometric features of the target image, and after maximum suppression processing, obtain the digital biometric feature matrix of the individual.

[0052] Specifically, the black and white protrusions on the sides of the red swamp crayfish are biological features; these biological features are molting invariable and growth stable. This feature will not be deformed by the bending of the individual. When using the model for processing, the model remains robust to partial deformation and spatial distortion of the individual.

[0053] Therefore, the cephalothorax (dorsal or lateral side) of the red swamp crayfish is imaged to extract biological features;

[0054] A biometric feature extraction model is constructed, which uses CSPNEXt (Cross Stage Partial Network eXtended) as the backbone network and PAFPN (Path Aggregation Network with Cross Stage Partial blocks) as the neck structure to build a multi-scale feature pyramid. The model achieves efficient target localization and classification through a separate bounding box prediction head.

[0055] Each slice image is used as an independent input to the biometric extraction model. The model performs confidence threshold filtering on the final detection results and restores the coordinates to the original image space. It also supports full-image standard inference as a comparison to improve recall.

[0056] As one example, a complete target image is input into a biometric extraction model to obtain global digital biometric features;

[0057] After segmenting the target image according to a preset overlap ratio, the segmented image is input into a biometric feature extraction model to obtain the corresponding local digital biometric features;

[0058] By fusing global and local digital biometric features, an individual digital biometric feature matrix is ​​obtained through a standardized quality control scheme.

[0059] Compared with traditional single-image feature extraction methods, this innovation has the following advantages: it significantly improves the detection rate of small-sized targets, and is particularly suitable for tasks such as biological images, insect recognition, and lesion detection; it adapts to scenes with densely arranged targets, avoiding target occlusion and accidental deletion.

[0060] Step 3: Input the individual digital biometric feature matrix into the improved graph neural network model. In the edge attention graph convolution module, queries and keys are generated by independent multilayer perceptrons. The edge weights are calculated by the scaled dot product of query and key, and the attention coefficients are obtained by softmax. After concatenation of multi-head attention, multiple node features are obtained.

[0061] Specifically, the improved graph neural network model is an improved graph neural network model for extracting embedded representations of two-dimensional point clouds. Traditional graph attention mechanisms rely only on neighbor node information to form attention, but do not consider the relative relationship between nodes and their neighbors, i.e., the relative geometric structure.

[0062] This invention provides an improved graph neural network model for point set embedding that combines an attention mechanism based on edge-relative geometry (EdgeGATv2Conv) with a multi-scale fusion strategy, such as... Figure 2 As shown, it has stronger characterization capabilities and scalability. Specifically:

[0063] 1. Input module

[0064] The input is a set of individual feature vectors of size [B, N, D], where B represents the batch size, N represents the number of biometric points in each group, and D=2 represents the two-dimensional coordinate information.

[0065] 2. EdgeGATv2Conv (EdgeAttentionGraphConvolutional Module)

[0066] The edge attention graph convolution module (EdgeGATv2Conv) is constructed and computed in the following manner:

[0067] Node representations preserve original features (i.e., do not perform initial linear transformations, enhancing structural interpretability), such as... Figure 3 As shown, queries and keys are generated through independent multilayer perceptrons. The edge weights are calculated by scaling the dot product of query and key, and then the attention coefficients are obtained by softmax.

[0068] Node information is used to generate queries through multiple multilayer perceptrons, and the relative geometric information between a node and its neighbors is used to generate keys through multiple multilayer perceptrons. Q * K is then processed through softmax to obtain attention weights ranging from 0 to 1. The node information and the relative geometric information between a node and its neighbors are merged and then used to generate values ​​through multiple multilayer perceptrons. Information propagation is performed through an edge-aware MLP model, and edge representations are constructed by combining the relative geometric information between a node and its neighbors. The multi-head outputs are concatenated and uniformly linearly mapped to node features.

[0069] Step 4: Concatenate the different node features and generate an attention score for each node based on the gating mechanism aggregation module. After global pooling, output the global representation vector of the digital biometric matrix. Map the global representation vector to a fixed dimension through a linear network and use it as the individual's biometric vector. Finally, output the individual identification result of Procambarus clarkii.

[0070] Specifically, this disclosure uses features extracted from a bottom-up, multi-layer EdgeGATv2Conv architecture and employs a multi-scale feature fusion module to perform the following fusion process:

[0071] The node features output from different layers (such as feature1 and feature2) are concatenated and projected onto a unified high-dimensional space (1024 dimensions) through an MLP (including batch normalization and nonlinear activation).

[0072] To further enhance global semantic modeling capabilities, an aggregation module (global attention pooling module) based on a gating mechanism is constructed to generate attention scores for each node and perform a global pooling operation after softmax weighting; optional dropout can be used to improve generalization capabilities.

[0073] Specifically, the updated node feature information for each individual is processed through a multilayer perceptron and then through softmax to obtain the attention score for each node. The updated node feature information for each individual is then processed through another multilayer perceptron to obtain the value, and then weighted and summed with the attention score at the channel level to obtain a new vector. This global attention pooling module outputs the global representation vector for each set of points.

[0074] The global representation vector is input into the embedding mapping module and mapped to a fixed-dimensional vector (such as 128-dimensional) through a linear network. This vector is then used as the final point set embedding and serves as the individual's biometric vector for final individual identification.

[0075] The biometric vector of the individual to be identified is further compared with a preset benchmark feature library. Through similarity calculation and threshold determination, identity verification or identity retrieval is completed, and the identification result is output.

[0076] Specifically, a benchmark feature library is constructed using the biometric vectors and information of benchmark individuals. The benchmark individual vectors are preset stored benchmark features, which, together with preset individual information, form the database.

[0077] The number of red swamp crayfish to be identified is calculated based on the benchmark feature library. Cosine similarity is used to find the biometric vector of the individual to be identified and the six benchmark individual feature vectors with the closest cosine similarity in the benchmark feature library. The individual feature vector of the object to be identified is compared with the preset benchmark feature library. Through similarity calculation and threshold determination, identity verification or identity retrieval is completed, and the identification result is output.

[0078] As one example, L2 normalization is performed on the output to enhance distribution stability and adapt the metric learning loss function.

[0079] Example 2

[0080] One embodiment of this disclosure provides a biometric-based individual identification system for the red swamp crayfish, comprising:

[0081] The image acquisition module is used to acquire the target image to be identified and to preprocess it to obtain sliced ​​images;

[0082] The feature extraction module is used to input each preprocessed slice image and the complete target image into the biometric feature extraction model, and output the local digital biometric features of each slice image and the global digital biometric features of the target image. The local digital biometric features of each slice image and the global digital biometric features of the target image are fused together, and after maximum suppression processing, the digital biometric feature matrix of the individual is obtained.

[0083] The node feature mining module is used to input the individual's digital biometric matrix into the improved graph neural network model. In multiple nested edge attention graph convolution modules, queries and keys are generated through independent multilayer perceptrons. The edge weights are calculated through the scaled dot product of the query and key. Each biometric feature is combined with its k nearest neighbors to obtain the attention coefficient. The node feature of the feature is generated by combining the location information of the biometric feature and its attention coefficient.

[0084] The aggregation module is used to concatenate different node features, and the aggregation module based on the gating mechanism generates an attention score for each node. After global pooling, it outputs the global representation vector of the digital biometric feature matrix.

[0085] The embedding generation module is used to map the global representation vector into a fixed-dimensional vector through a multilayer perceptron, and use it as the biometric vector of the individual.

[0086] The identification module compares the biometric vector of the individual to be identified with a preset benchmark feature library. Through similarity calculation and threshold determination, it completes identity verification or identity retrieval and finally outputs the individual identification result of the red swamp crayfish.

[0087] Example 3

[0088] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the biometric-based individual identification method for Procambarus clarkii.

[0089] Example 4

[0090] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the biometric-based individual identification method for Procambarus clarkii.

[0091] Example 5

[0092] One embodiment of this disclosure provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the biometric-based individual identification method for Procambarus clarkii.

[0093] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0095] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for individual identification of Procambarus clarkii based on biological characteristics, characterized in that, include: The target image to be identified is acquired and preprocessed to obtain sliced ​​images; Each preprocessed slice image and the complete target image are input into the biometric feature extraction model, which outputs the local digital biometric features of each slice image and the global digital biometric features of the target image. The local digital biometric features of each slice image and the global digital biometric features of the target image are then fused. After maximum suppression processing, the digital biometric feature matrix of the individual is obtained. Among them, the black and white protrusions on the sides of the body of the red swamp crayfish are biometric features. The individual's digital biometric matrix is ​​input into an improved graph neural network model. Query and key are generated through independent multilayer perceptrons in multiple nested edge attention graph convolutional modules. Edge weights are calculated using the scaled dot product of query and key. Each biometric feature is combined with its k nearest neighbors to obtain an attention coefficient. The node feature of the feature is generated by combining the biometric feature's location information and its attention coefficient. Information propagation is performed through an edge-aware MLP model, and edge representations are constructed by combining the relative geometric information of nodes and neighboring nodes. The multi-head outputs are concatenated and uniformly linearly mapped to the node feature. Different node features are concatenated, and an aggregation module based on a gating mechanism generates an attention score for each node. After global pooling, the global representation vector of the digital biometric feature matrix is ​​output. The global representation vector is mapped to a fixed-dimensional vector through a multilayer perceptron and used as the biometric vector of the individual. The biometric vector of the individual to be identified is compared with a preset benchmark feature library. Through similarity calculation and threshold determination, identity verification or identity retrieval is completed, and the final output is the individual identification result of the red swamp crayfish.

2. The method for individual identification of Procambarus clarkii based on biological characteristics as described in claim 1, characterized in that, The dorsal or lateral aspect of the cephalothorax of the shrimp to be identified is imaged to obtain the target image. The resolution of the image is improved using a high-quality interpolation algorithm. Based on the preset slice size, the target image is sliced ​​in an overlapping manner so that each slice focuses on a local area.

3. The method for individual identification of Procambarus clarkii based on biological characteristics as described in claim 2, characterized in that, It supports automatically adjusting the slice layout according to the image resolution. Each slice retains its starting position index in the target image, which is convenient for coordinate reconstruction. Each slice is used as an independent input to the biometric extraction model.

4. The method for individual identification of Procambarus clarkii based on biological characteristics as described in claim 1, characterized in that, The segmented image is input into the biometrics extraction model to obtain the corresponding local digital biometrics; the target image is input into the biometrics extraction model to obtain the global digital biometrics of the target image. The local digital biometrics of each slice image are fused with the global digital biometrics of the target image. The feature extraction results are filtered by confidence threshold, and the coordinates are restored to the original image space. The detection box of each biometric target is converted into the center point coordinates.

5. The method for individual identification of Procambarus clarkii based on biological characteristics as described in claim 1, characterized in that, The individual's digital biometric matrix is ​​input into the improved graph neural network model. The individual's digital biometric matrix is ​​a point set tensor of size [B, N, D], where B represents the batch size, N represents the number of biometric points in each group, and D represents the two-dimensional coordinate information. Information propagation is carried out through the edge-aware MLP model, and edge representation is constructed by combining residual difference. After the multi-head output is concatenated, it is uniformly linearly mapped to node features. The edge attention graph convolution module is stacked in two layers to extract multiple node features.

6. The method for individual identification of Procambarus clarkii based on biological characteristics as described in claim 1, characterized in that, The multi-scale feature fusion module uses a bottom-up multi-EdgeGATv2Conv feature extraction method. In the fusion process after the second layer, the outputs of different layers are concatenated and projected to a unified high-dimensional space through MLP. The aggregation module based on the gating mechanism generates attention scores for each node. After global pooling, the global representation vector of each set of points is output.

7. A biometric-based individual identification system for Procambarus clarkii, characterized in that, include: The image acquisition module is used to acquire the target image to be identified and to preprocess it to obtain sliced ​​images; The feature extraction module is used to input each preprocessed slice image and the complete target image into the biofeature extraction model, and output the local digital biofeatures of each slice image and the global digital biofeatures of the target image. The local digital biofeatures of each slice image and the global digital biofeatures of the target image are fused and after maximum suppression processing, the digital biofeature matrix of the individual is obtained; among them, the black and white protrusions on the sides of the body of the red swamp crayfish are biofeatures. The node feature mining module is used to input the individual's digital biometric matrix into an improved graph neural network model. In multiple nested edge attention graph convolutional modules, queries and keys are generated through independent multilayer perceptrons. Edge weights are calculated by scaling the dot product of the query and key. Each biometric feature is combined with its k nearest neighbors to obtain an attention coefficient. The node feature of the feature is generated by combining the location information of the biometric feature and its attention coefficient. Information propagation is carried out through an edge-aware MLP model. The edge representation is constructed by combining the relative geometric information of the node and its neighboring nodes. The multi-head outputs are concatenated and uniformly linearly mapped to the node feature. The aggregation module is used to concatenate different node features, and the aggregation module based on the gating mechanism generates an attention score for each node. After global pooling, it outputs the global representation vector of the digital biometric feature matrix. The embedding generation module is used to map the global representation vector into a fixed-dimensional vector through a multilayer perceptron, and use it as the biometric vector of the individual. The identification module compares the biometric vector of the individual to be identified with a preset benchmark feature library. Through similarity calculation and threshold determination, it completes identity verification or identity retrieval and finally outputs the individual identification result of the red swamp crayfish.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the biometric identification method for Procambarus clarkii as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the biometric-based individual identification method for Procambarus clarkii as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the biometric-based individual identification method for Procambarus clarkii as described in any one of claims 1-6.

Images