Deep learning-based facial recognition and kinship identification system for macaques

Abstract

The invention discloses a macaque face recognition and kinship identification system based on deep learning, and relates to the technical field of face recognition. The invention comprises a dynamic capture module, a processing module, a face recognition module, a behavior analysis and kinship identification association analysis module; the dynamic capture module is used to capture macaque activity video data in an activity area; firstly, a suitable dynamic capture high-speed camera is selected to ensure that clear and continuous macaque activity video data can be captured, and cameras are arranged in the macaque activity area, the activity area is set on the macaque key area and path, and multiple cameras are arranged to cover different viewing angles and distances; the invention uses deep learning technology to quickly identify and analyze macaque facial features, thereby significantly improving the efficiency and accuracy of face recognition, and at the same time combines behavior analysis and genetic marker data to further improve the accuracy of kinship identification.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and in particular to a macaque face recognition kinship identification system based on deep learning. Background Art

[0002] With the rapid development of artificial intelligence and deep learning technologies, biometric recognition technology has been widely used in the field of wildlife protection and research. In particular, in primate research, facial recognition technology has gradually become a research hotspot due to its non-invasiveness and high efficiency. However, existing facial recognition technologies are mostly focused on humans or domestic animals. For wild animals, especially macaques with highly dynamic behaviors, their facial recognition and kinship identification systems are not yet mature.

[0003] As primates, macaques have complex social behaviors, and kinship has an important impact on group structure and social dynamics. Traditional macaque kinship identification methods mainly rely on genetic analysis, which is accurate but costly, time-consuming, and requires the capture and processing of macaques, which may cause stress and harm to the animals. In addition, genetic analysis cannot reflect the dynamic changes of macaque groups in real time, limiting its application in wildlife behavior and ecological research.

[0004] As an alternative method, facial recognition is easy to operate and has a fast response. However, the automatic detection and recognition of macaque facial features faces many challenges. First, macaque facial features are different from those of humans, and general facial recognition models are difficult to apply directly. Second, macaques have a wide range of activities and their behaviors are dynamic and changeable, making it difficult to capture dynamic facial images. Third, factors such as lighting changes, occlusion, and similarities between individuals in the natural environment increase the complexity of facial recognition.

[0005] In addition, macaque behavior analysis is another key way to understand their social structure and kinship. Current behavior analysis methods mostly rely on manual observation and recording, which is inefficient and highly subjective. Automated behavior analysis technology has not yet been popularized, especially in combining facial recognition results for individual behavior recognition and kinship inference, where technical means are limited.

[0006] In view of this, it is necessary to develop a macaque facial recognition kinship identification system based on deep learning, which can automatically capture macaque facial images, accurately identify facial features, and combine behavioral analysis results to provide efficient and objective technical means for macaque kinship research. This can not only improve research efficiency and reduce interference with animals, but also provide new perspectives and methods for wildlife protection and behavioral research.

[0007] In order to solve the above defects, a technical solution is now provided. Summary of the invention

[0008] The purpose of the present invention is to solve the problems of low accuracy, low efficiency and interference to animals in the existing macaque facial recognition and kinship identification technologies, and to propose a macaque facial recognition and kinship identification system based on deep learning.

[0009] The purpose of the present invention can be achieved through the following technical solutions:

[0010] A macaque facial recognition kinship identification system based on deep learning, including:

[0011] A dynamic capture module, used to capture video data of macaque activities in the activity area;

[0012] The processing module is used to analyze and process the collected video data. The specific steps are as follows:

[0013] The collected videos are preprocessed, including denoising and brightness adjustment. The facial detection algorithm is used to automatically detect the macaque facial area from the video data, and the macaque facial image is extracted from the video to form a facial data set. The specific process is as follows:

[0014] Extracting individual image frames from the preprocessed video data in preparation for subsequent processing, applying a face detection algorithm to identify the macaque face region in the image frame, and cutting out the detected face region from the original image frame to form a preliminary face image set;

[0015] The captured facial images are subjected to feature extraction through a deep learning model, the similarity between different facial images is calculated, and similar facial images are classified into the same category;

[0016] Then, the quality of different facial images in each category is evaluated, including collecting and quantifying the clarity, angle and occlusion area of ​​different facial images to obtain quality judgment values, and the obtained quality judgment values ​​are used as the standard for measuring the quality of facial images;

[0017] The facial recognition module is used to accurately identify the faces of macaques and identify different macaque individuals. The specific steps are as follows:

[0018] The quality judgment values ​​obtained for different facial images in each category are sorted in order of size, and one or several facial images with quality judgment values ​​from high to low are selected as representative reference facial images of the category;

[0019] Optimize the feature extraction process for the selected representative reference facial images, and use the optimized features to train the facial recognition model to adapt to the facial features of different macaque individuals;

[0020] The facial detection of representative reference facial images is verified by screening the optimized model to evaluate the accuracy of facial detection, ensure that the model can effectively distinguish the faces of different macaque individuals, and optimize the facial detection algorithm based on the facial detection results, iteratively update the facial recognition model to adapt to new facial data and improve recognition accuracy; then record and mark or number the facial features of the identified macaque individuals;

[0021] Behavioral analysis, used to analyze the behavior of macaques using macaque activity video data;

[0022] The kinship identification association analysis module is used to perform kinship identification correlation analysis based on the results of behavioral analysis.

[0023] Furthermore, the specific steps of the dynamic capture module capturing the video data of the macaque activity in the activity area are as follows:

[0024] First, we selected dynamic capture high-speed cameras and deployed them in the macaque activity areas. The activity areas were set in the macaque key areas and paths, and multiple cameras were deployed to cover different viewing angles and distances.

[0025] Enable multiple cameras for real-time monitoring and video recording, and store recorded videos in a secure location to ensure data integrity.

[0026] Furthermore, the specific operation steps of the processing module to calculate the similarity between different facial images are as follows:

[0027] Use convolutional neural networks to extract features from captured facial images. When using a custom dataset, train or adjust the deep learning model to adapt to specific facial image features.

[0028] Each facial image is processed by a deep learning model to generate a fixed-length feature vector, which is then normalized so that adjacent features have a uniform scale. The similarity between feature vectors is then calculated using a distance metric, which includes Euclidean distance, Manhattan distance, cosine similarity, and Jaccard similarity coefficient.

[0029] Then construct a similarity matrix, where each row and column represents a facial image, and the elements in the matrix represent the similarity between the corresponding images. A similarity threshold is set to distinguish between similar and dissimilar facial images.

[0030] The similarity matrix is ​​then used for cluster analysis to classify similar facial images into the same category. The similarity information is used for individual identification to determine whether different facial images belong to the same macaque individual. The accuracy of the similarity calculation is then verified using known individual information. The model parameters or similarity measurement method are adjusted based on the verification results to improve the accuracy of the similarity calculation.

[0031] Furthermore, the specific operation steps of the processing module to evaluate the quality of different facial images in each category are as follows:

[0032] First, the clarity, angle and occlusion area of ​​different facial images are collected and quantified;

[0033] The clarity quantification is performed by obtaining the following: Sharpness measurement: using edge detection algorithm to evaluate the sharpness of the image. The higher the sharpness, the clearer the image.

[0034] Laplace operator: evaluates sharpness by calculating the second-order derivative of the image;

[0035] Fourier transform: Analyze high-frequency components through Fourier transform of the image;

[0036] After normalizing the acquired sharpness metric, Laplace operator and Fourier transform high-frequency component, a triangle is established with the sharpness metric, Laplace operator and Fourier transform high-frequency component as the three sides of the triangle, and then a height is preset to establish a triangular pyramid model, and the surface area of ​​the triangular pyramid model is calculated and recorded as a clarity value, and this clarity value is used as a standard for measuring the clarity of the facial image;

[0037] Angle quantification: Key point detection: Use the facial key point detection algorithm to locate the facial feature points of the macaque, analyze the number of facial feature points in the facial image, assign an importance score to each feature point, calculate the sum of the importance scores of all feature points in the facial image, and record it as the point value;

[0038] Angle estimation: Calculate the inclination angle and rotation angle of the face according to the position of the facial feature points. After normalization, calculate the sum of the product of the inclination angle and rotation angle and the preset corresponding weights, which is recorded as the angle deviation value;

[0039] Then normalize the obtained point value and angular deviation value and enter them into the following formula: The angle evaluation value JPZ is obtained, and the obtained angle evaluation value is used as the angle standard for measuring the facial image;

[0040] Occlusion area quantification: Use skin color detection algorithm to determine the facial area, then compare it with the original image to calculate the occluded area;

[0041] Then normalize the obtained clarity value, angle evaluation value and occlusion area and enter them into the following formula: To obtain the quality judgment value ZPZ, where QXZ, JPZ and ZDM are the clarity value, corner evaluation value and occlusion area respectively, and κ1, κ2, κ3 are the preset weight coefficients of the clarity value, corner evaluation value and occlusion area respectively.

[0042] Furthermore, the specific operation steps of the behavior analysis of macaques through macaque activity video data are as follows:

[0043] Obtain the video data of macaque activities collected by the motion capture module and segment several long videos. The specific segmentation process is as follows:

[0044] First, a computer vision algorithm is used to detect multiple macaques in a long video, and a multi-target tracking algorithm is used to track each macaque in the video.

[0045] According to the research purpose, a threshold is set to determine the start and end of the video segment segmentation, and the segmentation logic is defined, including starting a new segment when two or more macaques appear in the video and ending the current segment when the number of macaques decreases to one or less;

[0046] Monitor the video stream in real time and segment it according to the results of entity detection and tracking. When the segmentation threshold is reached, mark the segmentation point and prepare to start recording a new segment. Manually check the results of automatic segmentation to ensure the accuracy of the segmentation point and correct errors in the automatic segmentation process.

[0047] Then, the segmented video clips are screened, and video clips with only one or no macaques are eliminated. The screened video clips are marked, and the number, marks and activity types of macaques in each clip are recorded;

[0048] Define the types of interactions and relationships to be analyzed, select behavior analysis algorithms, including machine learning and deep learning methods, track individual macaques in the video, ensure that each individual can be identified and tracked throughout the clip, and apply behavior recognition algorithms to automatically identify macaque interactions and behavior patterns in different video clips;

[0049] Then, by analyzing the interactions between individual macaques, identifying social structures and close relationships, and then constructing a social network diagram, visualizing the frequency and pattern of interactions between individual macaques, extracting sequence data of macaque behavior, analyzing the order and duration of behavior, and identifying recurring behavior patterns;

[0050] Define features used to describe macaque behavior, quantify features describing behavior, and provide numerical data for behavioral analysis;

[0051] The behavioral identification results, individual tracking data, and behavioral features were integrated together, and statistical methods were used to analyze the relationships and patterns between different behaviors. Visualization tools were then used to display the behaviors and interactions of the macaques, and to incorporate individual markers of the facial features of the different macaques identified.

[0052] Furthermore, the specific operation steps of the kinship identification correlation analysis module for performing kinship identification correlation analysis based on the results of the behavior analysis are as follows:

[0053] The results of the behavioral analysis were used to extract the correlation of kinship identification, including: interaction frequency, interaction duration and interaction type;

[0054] Conduct spatial relationship analysis, social network analysis, behavior sequence analysis, and prosocial behavior analysis on the extracted content;

[0055] Then, the facial features of different macaque individuals are obtained from the facial recognition module, and the corresponding genetic markers are extracted;

[0056] Integrate the genetic marker data of macaques and calculate the genetic similarity between different macaque individuals;

[0057] The behavioral data was then combined with the genetic data to improve the accuracy of macaque kinship identification. Statistical methods were used to analyze the association between behavioral traits and genetic similarities, and the individual numbers of macaques with high correlations were recorded to provide data support for subsequent DNA testing.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] (1) The present invention uses deep learning technology to quickly identify and analyze macaque facial features, significantly improving the efficiency and accuracy of facial recognition. At the same time, it combines behavioral analysis and genetic marker data to further improve the accuracy of kinship identification;

[0060] (2) The present invention uses a dynamic capture module to automatically collect video data of macaque activities in a non-invasive environment, avoiding the capture and processing of macaques in traditional methods, reducing interference with the daily lives of animals and potential stress and harm;

[0061] (3) The present invention provides an efficient method for studying kinship in macaques, which helps scientists gain a deeper understanding of the social structure and behavioral patterns of macaques and provides important technical support and decision-making basis for wildlife protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;

[0063] Figure 1This is the overall system block diagram of the present invention. DETAILED DESCRIPTION

[0064] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0066] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations.

[0067] like Figure 1 As shown, the macaque facial recognition and kinship identification system based on deep learning includes a dynamic capture module, a processing module, a facial recognition module, a behavior analysis module, and a kinship identification association analysis module;

[0068] The dynamic capture module is used to capture the video data of the macaque activities in the activity area;

[0069] First, select a suitable dynamic capture high-speed camera to ensure that clear and continuous video data of macaque activities can be captured, and deploy cameras in the macaque activity area. The activity area is set in the key area and path of the macaques. Multiple cameras are deployed to cover different viewing angles and distances; start multiple cameras for real-time monitoring and video recording, and store the recorded videos in a safe location to ensure the integrity of the data.

[0070] The processing module is used to analyze and process the collected video data;

[0071] The collected videos were preprocessed, including denoising and brightness adjustment. The facial detection algorithm was used to automatically detect the macaque facial area from the video data, and the macaque facial images were extracted from the video to form a facial data set. The specific steps are as follows: extract separate image frames from the preprocessed video data in preparation for subsequent processing, apply facial detection algorithms, including Haar cascade classifiers, HOG+SVM and MTCNN, identify the macaque facial area in the image frame, and cut out the detected facial area from the original image frame to form a preliminary facial image set; extract features from the cut facial images through a deep learning model, and calculate the similarity between different facial images. The specific process is as follows:

[0072] Use the convolutional neural network (CNN) deep learning model to extract features from the captured facial images. When a custom data set is used, the deep learning model is trained or fine-tuned to adapt to specific facial image features. Each facial image is processed by the deep learning model to generate a feature vector of fixed length. The feature vector is normalized so that adjacent features have a uniform scale. The distance measurement method is then used to calculate the similarity between feature vectors. The distance measurement methods include:

[0073] Euclidean distance: measures the straight-line distance between two points in multidimensional space; Manhattan distance: measures the "city block" distance between two points in multidimensional space; Cosine similarity: measures the degree of similarity between two vectors in direction, and is insensitive to the vector's modulus; Jaccard similarity coefficient: measures the similarity between two sets, and is applicable to situations where the feature vector contains binary or sparse data;

[0074] Then, a similarity matrix is ​​constructed, in which each row and each column represents a facial image, and the elements in the matrix represent the similarity between the corresponding images. A similarity threshold is set to distinguish similar and dissimilar facial images. Then, the similarity matrix is ​​used for cluster analysis to classify similar facial images into the same category, and the similarity information is used for individual recognition to determine whether different facial images belong to the same macaque individual. Then, the accuracy of the similarity calculation is verified by the known individual information, and the model parameters or the similarity measurement method are adjusted according to the verification results to improve the accuracy of the similarity calculation.

[0075] Then the quality of different facial images in each category is evaluated, including the quantification of the clarity, angle and occlusion area of ​​different facial images;

[0076] Among them, clarity quantification: Obtain facial images: Sharpness measurement: Use edge detection algorithms (such as Sobel operator, Canny operator) to evaluate the sharpness of the image. The higher the sharpness, the clearer the image; Laplace operator: Evaluate clarity by calculating the second-order derivative of the image. High-definition images will have more obvious Laplace responses; Fourier transform: Analyze high-frequency components through Fourier transform of the image. The more high-frequency components, the clearer the image; After normalizing the acquired sharpness measurement, Laplace operator and Fourier transform high-frequency components, respectively use the sharpness measurement, Laplace operator and Fourier transform high-frequency components as the three sides of the triangle to establish a triangle, and then preset a height to establish a triangular pyramid model, calculate the surface area of ​​the triangular pyramid model and record it as the clarity value, and use this clarity value as the standard for measuring the clarity of the facial image;

[0077] Angle quantification: Key point detection: Use facial key point detection algorithm (such as Haar feature of dlib or OpenCV) to locate the facial feature points of macaques, such as eyes, nose, mouth, etc., analyze the number of facial feature points in the facial image, and assign an importance score to each feature point. Calculate the sum of the importance scores of all feature points in the facial image and record it as the point value; Angle estimation: Calculate the inclination angle and rotation angle of the face according to the position of the facial feature points, calculate the sum of the product of the inclination angle and rotation angle with the preset corresponding weight after normalization, and record it as the angle deviation value. The larger the calculated angle deviation value, the worse the angle of the corresponding facial image; Then, the obtained point value and angle deviation value are normalized and inserted into the following formula: The angle evaluation value JPZ is obtained, and the obtained angle evaluation value is used as the angle standard for measuring the facial image;

[0078] Occlusion area quantification: Use skin color detection algorithm to determine the facial area, then compare it with the original image to calculate the occluded area;

[0079] Then normalize the obtained clarity value, corner evaluation value JPZ and occlusion area and enter them into the following formula: To obtain the quality judgment value ZPZ, where QXZ, JPZ and ZDM are the clarity value, corner evaluation value and occlusion area respectively, κ1, κ2, κ3 are the preset weight coefficients of the clarity value, corner evaluation value and occlusion area respectively, and the values ​​are set between 1.12-1.56, 0.91-0.98 and 1.11-1.63 respectively; and the obtained quality judgment value is used as the standard for measuring the quality of facial images.

[0080] The facial recognition module is used to accurately identify the faces of macaques and identify different macaque individuals;

[0081] The quality judgment values ​​obtained from different facial images in each category are sorted according to size, and one or several facial images with high to low quality judgment values ​​are selected as representative reference facial images of the category; then the representative reference facial images of each category are determined; for the selected representative reference facial images, the feature extraction process is further optimized to improve the accuracy of facial detection, and the facial recognition model is trained using the optimized features to adapt to the facial features of different macaque individuals;

[0082] By screening and optimizing the model, facial detection verification is performed on representative reference facial images to evaluate the accuracy of facial detection, ensuring that the model can effectively distinguish the faces of different macaque individuals. Based on the facial detection results, the facial detection algorithm is continuously optimized and the facial recognition model is iteratively updated to adapt to new facial data and improve recognition accuracy. The facial features of the identified macaque individuals are then recorded and marked or numbered to prepare for subsequent analysis.

[0083] Behavioral analysis is used to analyze the behavior of macaques using video data of macaque activities;

[0084] The video data of macaque activities collected by the motion capture module was obtained, and several long videos were segmented. The specific segmentation process was as follows: First, computer vision algorithms such as YOLO and SSD based on deep learning were used to detect multiple macaques in the long video, and multi-target tracking algorithms such as SORT and DeepSORT were used to track each macaque in the video; a threshold was set according to the research purpose to determine the appropriate start and end of the segmentation of the video segment, and the segmentation logic was defined, including starting a new segment when two or more macaques appeared in the video, and ending the current segment when the number of macaques decreased to one or less;

[0085] Monitor the video stream in real time and segment it according to the results of entity detection and tracking. When the segmentation threshold is reached, mark the segmentation point and prepare to start recording a new segment. Manually check the results of automatic segmentation to ensure the accuracy of the segmentation point and correct the errors in the automatic segmentation process. Then screen several segmented video segments, eliminate video segments with only one or no macaques in the segment, and annotate the screened video segments, record the number of macaque individuals, tags and activity types in each segment. Define in detail the interactive behaviors and close relationship types that need to be analyzed, select appropriate behavioral analysis algorithms, including machine learning and deep learning methods, track individual macaques in the video, ensure that each individual can be identified and tracked in the entire segment, and apply behavioral recognition algorithms to automatically identify the interactive actions and behavioral patterns of macaques in different video segments.

[0086] Then, by analyzing the interactions between individual macaques, identifying social structures and close relationships, and then constructing a social network diagram, visualizing the frequency and pattern of interactions between individual macaques, extracting sequential data of macaque behavior, analyzing the order and duration of behavior, and identifying recurring behavioral patterns, such as specific interaction rituals or close behaviors; defining characteristics used to describe macaque behavior, such as movement speed, frequency, duration, etc., quantifying the characteristics of the behavior, and providing numerical data for behavioral analysis; integrating behavior identification results, individual tracking data, and behavioral characteristics, using statistical methods to analyze the relationships and patterns between different behaviors, and then using visualization tools to display the macaques' behaviors and interactions, and incorporating individual markers of the identified facial features of different macaques to improve visualization.

[0087] The kinship identification correlation analysis module is used to perform kinship identification correlation analysis based on the results of behavioral analysis;

[0088] The results of behavioral analysis are used to extract the correlation of kinship identification, and the extracted content includes:

[0089] Interaction frequency: record the interaction frequency between different macaques. Relatives may have a higher interaction frequency. Interaction duration: measure the duration of interaction between macaques. Relatives may have longer interactions. Interaction type: mark and record the specific type of interaction, such as foraging together, mutual grooming, fighting, etc.

[0090] The extracted content was analyzed for spatial relationships, social network, behavior sequence, and prosocial behavior. The spatial relationship analysis included: Proximity: analyzing the spatial proximity of individual macaques. Relatives may be more inclined to be close to each other; Spatial sharing: recording the pattern of space sharing among individual macaques. Relatives may have closer space sharing;

[0091] Social network analysis includes: Social network construction: constructing macaque social networks based on interaction data, and relatives may form close social connections; Network centrality: analyzing the centrality of individuals in social networks, and relatives may occupy an important position in the network; Behavior sequence analysis includes: Behavior sequence: recording the behavior sequence of individual macaques, and relatives may have similar behavior patterns; Behavior synchronization: analyzing the synchronization of behavior between individual macaques, and relatives may show higher behavioral consistency; Prosocial behavior analysis includes: Prosocial behavior: recording prosocial behaviors such as mutual grooming and sharing food, which may be more common among relatives; Care behavior: observing and recording care behavior for young monkeys, which may indicate kinship;

[0092] The facial features of different macaque individuals are then obtained from the facial recognition module, and the corresponding genetic markers are extracted to integrate the genetic marker data of the macaques. This is direct evidence for kinship identification, and the genetic similarity between different macaque individuals is calculated. High similarity may indicate a kinship relationship. The behavioral data is then combined with the genetic data to improve the accuracy of macaque kinship identification. Statistical methods are used to analyze the correlation between behavioral characteristics and genetic similarity, and the numbers of macaque individuals with high correlation are recorded to provide data support for subsequent DNA testing.

[0093] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A macaque facial recognition kinship identification system based on deep learning, characterized in that: include: A dynamic capture module, used to capture video data of macaque activities in the activity area; The processing module is used to analyze and process the collected video data. The specific steps are as follows: The collected videos are preprocessed, including denoising and brightness adjustment. The facial detection algorithm is used to automatically detect the macaque facial area from the video data, and the macaque facial image is extracted from the video to form a facial data set. The specific process is as follows: Extracting individual image frames from the preprocessed video data in preparation for subsequent processing, applying a face detection algorithm to identify the macaque face region in the image frame, and cutting out the detected face region from the original image frame to form a preliminary face image set; The captured facial images are subjected to feature extraction through a deep learning model, the similarity between different facial images is calculated, and similar facial images are classified into the same category; Then, the quality of different facial images in each category is evaluated, including collecting and quantifying the clarity, angle and occlusion area of ​​different facial images to obtain quality judgment values, and the obtained quality judgment values ​​are used as the standard for measuring the quality of facial images; The facial recognition module is used to accurately identify the faces of macaques and identify different macaque individuals. The specific steps are as follows: The quality judgment values ​​obtained for different facial images in each category are sorted in order of size, and one or several facial images with quality judgment values ​​from high to low are selected as representative reference facial images of the category; Optimize the feature extraction process for the selected representative reference facial images, and use the optimized features to train the facial recognition model to adapt to the facial features of different macaque individuals; The facial detection of representative reference facial images was verified by screening the optimized model to evaluate the accuracy of facial detection and ensure that the model can effectively distinguish the faces of different macaque individuals. Based on the facial detection results, the facial detection algorithm was optimized and the facial recognition model was iteratively updated to adapt to new facial data and improve recognition accuracy. Then, the facial features of the identified macaque individuals are recorded and marked or numbered; Behavioral analysis, used to analyze the behavior of macaques using macaque activity video data; The kinship identification association analysis module is used to perform kinship identification correlation analysis based on the results of behavioral analysis.

2. The macaque face recognition and kinship identification system based on deep learning according to claim 1, characterized in that: The specific steps of the dynamic capture module capturing the video data of the macaque activity in the activity area are as follows: First, we selected dynamic capture high-speed cameras and deployed them in the macaque activity areas. The activity areas were set in the macaque key areas and paths, and multiple cameras were deployed to cover different viewing angles and distances. Enable multiple cameras for real-time monitoring and video recording, and store recorded videos in a secure location to ensure data integrity.

3. The macaque face recognition and kinship identification system based on deep learning according to claim 1, characterized in that: The specific operation steps of the processing module to calculate the similarity between different facial images are as follows: Use convolutional neural networks to extract features from captured facial images. When using a custom dataset, train or adjust the deep learning model to adapt to specific facial image features. Each facial image is processed by a deep learning model to generate a fixed-length feature vector, which is then normalized so that adjacent features have a uniform scale. The similarity between feature vectors is then calculated using a distance metric, which includes Euclidean distance, Manhattan distance, cosine similarity, and Jaccard similarity coefficient. Then construct a similarity matrix, where each row and column represents a facial image, and the elements in the matrix represent the similarity between the corresponding images. A similarity threshold is set to distinguish between similar and dissimilar facial images. The similarity matrix is ​​then used for cluster analysis to classify similar facial images into the same category. The similarity information is used for individual identification to determine whether different facial images belong to the same macaque individual. The accuracy of the similarity calculation is then verified using known individual information. The model parameters or similarity measurement method are adjusted based on the verification results to improve the accuracy of the similarity calculation.

4. The macaque face recognition and kinship identification system based on deep learning according to claim 1, characterized in that: The specific operation steps of the processing module to evaluate the quality of different facial images in each category are as follows: First, the clarity, angle and occlusion area of ​​different facial images are collected and quantified; The clarity quantification is performed by obtaining the following: Sharpness measurement: using edge detection algorithm to evaluate the sharpness of the image. The higher the sharpness, the clearer the image. Laplace operator: evaluates sharpness by calculating the second-order derivative of the image; Fourier transform: Analyze high-frequency components through Fourier transform of the image; After normalizing the acquired sharpness metric, Laplace operator and Fourier transform high-frequency component, a triangle is established with the sharpness metric, Laplace operator and Fourier transform high-frequency component as the three sides of the triangle, and then a height is preset to establish a triangular pyramid model, and the surface area of ​​the triangular pyramid model is calculated and recorded as a clarity value, and this clarity value is used as a standard for measuring the clarity of the facial image; Angle quantification: Key point detection: Use the facial key point detection algorithm to locate the facial feature points of the macaque, analyze the number of facial feature points in the facial image, assign an importance score to each feature point, calculate the sum of the importance scores of all feature points in the facial image, and record it as the point value; Angle estimation: Calculate the inclination angle and rotation angle of the face according to the position of the facial feature points. After normalization, calculate the sum of the product of the inclination angle and rotation angle and the preset corresponding weights, which is recorded as the angle deviation value; Then normalize the obtained point value and angular deviation value and enter them into the following formula: To obtain the angle evaluation value JPZ, where dw and jp are the point value and angle deviation value respectively, and the angle evaluation value obtained is used as the angle standard for measuring the facial image; Occlusion area quantification: Use skin color detection algorithm to determine the facial area, then compare it with the original image to calculate the occluded area; Then normalize the obtained clarity value, angle evaluation value and occlusion area and enter them into the following formula: To obtain the quality judgment value ZPZ, where QXZ, JPZ and ZDM are the clarity value, corner evaluation value and occlusion area respectively, and κ1, κ2, κ3 are the preset weight coefficients of the clarity value, corner evaluation value and occlusion area respectively.

5. The macaque face recognition and kinship identification system based on deep learning according to claim 1, characterized in that: The specific operation steps of the behavior analysis of macaques using macaque activity video data are as follows: Obtain the video data of macaque activities collected by the motion capture module and segment several long videos. The specific segmentation process is as follows: First, a computer vision algorithm is used to detect multiple macaques in a long video, and a multi-target tracking algorithm is used to track each macaque in the video. According to the research purpose, a threshold is set to determine the start and end of the video segment segmentation, and the segmentation logic is defined, including starting a new segment when two or more macaques appear in the video and ending the current segment when the number of macaques decreases to one or less; Monitor the video stream in real time and segment it according to the results of entity detection and tracking. When the segmentation threshold is reached, mark the segmentation point and prepare to start recording a new segment. Manually check the results of automatic segmentation to ensure the accuracy of the segmentation point and correct errors in the automatic segmentation process. Then, the segmented video clips are screened, and video clips with only one or no macaques are eliminated. The screened video clips are marked, and the number, marks and activity types of macaques in each clip are recorded; Define the types of interactions and relationships to be analyzed, select behavior analysis algorithms, including machine learning and deep learning methods, track individual macaques in the video, ensure that each individual can be identified and tracked throughout the clip, and apply behavior recognition algorithms to automatically identify macaque interactions and behavior patterns in different video clips; Then, by analyzing the interactions between individual macaques, identifying social structures and close relationships, and then constructing a social network diagram, visualizing the frequency and pattern of interactions between individual macaques, extracting sequence data of macaque behavior, analyzing the order and duration of behavior, and identifying recurring behavior patterns; Define features used to describe macaque behavior, quantify features describing behavior, and provide numerical data for behavioral analysis; The behavioral identification results, individual tracking data, and behavioral features were integrated together, and statistical methods were used to analyze the relationships and patterns between different behaviors. Visualization tools were then used to display the behaviors and interactions of the macaques, and to incorporate individual markers of the facial features of the different macaques identified.

6. The macaque face recognition and kinship identification system based on deep learning according to claim 1, characterized in that: The specific operation steps of the kinship identification correlation analysis module for performing kinship identification correlation analysis based on the results of the behavior analysis are as follows: The results of the behavioral analysis were used to extract the correlation of kinship identification, including: interaction frequency, interaction duration and interaction type; Conduct spatial relationship analysis, social network analysis, behavior sequence analysis, and prosocial behavior analysis on the extracted content; Then, the facial features of different macaque individuals are obtained from the facial recognition module, and the corresponding genetic markers are extracted; Integrate the genetic marker data of macaques and calculate the genetic similarity between different macaque individuals; The behavioral data was then combined with the genetic data to improve the accuracy of macaque kinship identification. Statistical methods were used to analyze the association between behavioral traits and genetic similarities, and the individual numbers of macaques with high correlations were recorded to provide data support for subsequent DNA testing.

Images