Face image detection method based on distributed optimization instance selection

Abstract

The invention discloses a face image detection method based on distributed optimization instance selection. The face image detection method is applied to a work and coordination node network. A work node carries out multi-objective evolutionary optimization with classification precision and compression rate as targets on a local sample; the coordination node aggregates the population by adopting direct splicing or based on a conflict instance strategy according to an iteration process and executes global evolution; variation parameters are adjusted by counting the instance frequency of the Pareto optimal solution set, and feedback information is generated to guide subsequent iteration of the working nodes; and finally, selecting an optimal SVM model for detection. Through a distributed coevolution and feedback mechanism, redundant samples can be effectively eliminated under the condition that data are not concentrated, and the recognition efficiency and classification accuracy of face detection are improved.

Description

Technical Field

[0001] This invention relates to the fields of evolutionary computation and face images, specifically to an instance selection method based on distributed optimization, which can be applied to face image detection scenarios. Background Technology

[0002] The human face is a core carrier of human identity and emotional interaction. Research on facial images mainly includes identity recognition, expression analysis, feature extraction, and dynamic behavior monitoring. With the continuous improvement of computing power and the rapid iteration of deep learning algorithms, the field of facial image processing is gradually replacing traditional, inefficient manual recognition and simple feature matching methods with automated and intelligent processing.

[0003] Faces of different individuals and with different expressions exhibit unique features in images, including facial structure, texture details, and facial contours. By utilizing facial feature extraction and matching methods, individuals can be identified from complex scenes, and facial expressions can be categorized, enabling identity verification and emotional state analysis. Obtaining accurate facial feature information is a crucial prerequisite for applications such as security monitoring, human-computer interaction, and affective computing.

[0004] Facial image technology is one of the core applications of computer vision in many fields. In real-world scenarios, facial image technology can provide real-time and efficient information such as target identity, dynamic facial expression changes, and micro-expression features. It can also continuously monitor faces in complex environments (such as changing lighting conditions and occluded scenes), forming a multi-scene, multi-dimensional collection of facial information. This efficient and accurate multi-dimensional facial information plays an irreplaceable supporting role in the development of fields such as security upgrades, smart terminal interaction, and emotion-assisted diagnosis.

[0005] Traditional SVM classification models have significant limitations when handling large amounts of facial image data and distributed data storage. When dealing with massive amounts of facial image data, traditional SVM classifiers suffer from long training and prediction times, and the influence of redundant and erroneous data on classification performance, resulting in poor classification results. Furthermore, the significant time investment in training the classification model and predicting results reduces the timeliness of the facial data. When processing distributed facial data, traditional SVM classification models fail to demonstrate good performance due to limitations in data transmission between nodes. Summary of the Invention

[0006] The present invention addresses the shortcomings of the existing technology by proposing a face image detection method based on distributed optimization instance selection. Through distributed co-evolution and bidirectional feedback mechanisms, redundant samples can be effectively eliminated when data is not centralized, thereby improving the efficiency, accuracy, and adaptability of face image detection to distributed environments.

[0007] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: The face image detection method based on distributed optimization instance selection of this invention is characterized by being applied in a network environment consisting of M worker nodes and 1 coordinating node, and includes the following steps: S1. Obtain the i-th working node. facial image sample set ,in, Represents the i-th working node The wth face image sample, For the i-th working node The set of attribute features of the w-th face image sample, and , express The d-th attribute feature, d∈ D is the total number of attribute features of a face image sample. express Category tags, ∈ C represents the total number of categories; For the i-th working node The total number of face image samples; i M is the total number of worker nodes; The complete set of face image samples of the coordination node CN is composed of the face image sample sets of M working nodes. ,in, Indicates the first Personal facial image samples, For the coordination node The set of attribute features of individual face image samples, and Include Individual attribute features; express Category tags, ; To coordinate the total number of face image samples at node CN, and ; The population that constructs the coordinating node CN includes There are 10 individuals, and each individual contains 10 individuals. There are 10 binary chromosomes, each of which corresponds to 10 binary chromosomes. If the value of the binary chromosome of a face image sample is "1", it means that the face image sample at the corresponding position is selected; if it is "0", it means that the face image sample at the corresponding position is not selected. S2. Define the global current iteration count as t, and initialize it. =1; Define the global maximum number of iterations as T; S3. Define the current algebra of the working node as... and initialize =1; Define the maximum number of iterations for a worker node as . ; Define and randomly initialize In the The nth global iteration Generation population },in, express The nth individual represents a face image sample selection scheme, and Include There are 3 binary chromosomes, each representing whether to select a face image sample at the corresponding location. The Middle If each binary chromosome is "1", it indicates selection. ,like The Middle If a binary chromosome is "0", it means no selection is made. N is the population size; S4, will The corresponding subset of face image samples is input into the SVM classifier for training, to obtain the first... The nth global iteration The nth face image classification model and calculate Classification accuracy and compression ratio Thus constitute objective function value ; S5, if Upon receiving feedback from the coordinating node, then... Each individual in the process undergoes uniform crossover and preference-based mutation operations to produce... In the The nth global iteration Offspring population Otherwise, for Each individual in the process undergoes uniform crossover and single-point mutation operations to produce... In the The nth global iteration Offspring population ; S6, will and After performing a non-dominated sort, the top N individuals are selected from the sorted population using a binary tournament strategy to form the population. In the +1 global iterations of the th +1 generation population ; S7, will Assign to ,like = Then In the The nth global iteration Generation population Send the values ​​to the coordinating node CN and execute step S8; otherwise, assign t+1 to t and return to S4 to execute sequentially. S8. Define the current algebra of the coordinating node CN as: and initialize =1; Define the maximum number of iterations for the coordinating node CN as 1. ; S9, if t < ,Will Merged into the first The nth global iteration Generation population },in, express The nth individual represents a face image sample selection scheme, and Include There are 3 binary chromosomes, each representing whether to select a face image sample at the corresponding location. The Middle If each binary chromosome is "1", it indicates selection. ,like The Middle If a binary chromosome is "0", it means no selection is made. Otherwise, the coordinating node uses an aggregation strategy based on conflict instances to generate the first... The nth global iteration Generation population ; S10, Coordinating Node CN will The nth individual The corresponding subset of face image samples is input into the SVM classifier for training, to obtain the first... The nth global iteration The nth face image classification model ,and exist On the precision and compression ratio Thus constitute objective function value Therefore, based on the objective function value, for Each individual in the process undergoes uniform crossover and single-point mutation operations to produce a coordinating node CN in the th... The nth global iteration Offspring population ; S11, will and After performing a non-dominated sort, the top N individuals are selected from the sorted population using a binary tournament strategy to form the population. In the The nth global iteration +1 generation population ; S12, will Assign to ,like After assigning t+1 to t, return to S9 to execute; like = and CN in the The nth global iteration Generation population The globally optimal individual and the mutation parameters based on global preferences are sent as feedback information to the (t+1)th global iteration. The Generation population After assigning t+1 to t, return to S4 and execute sequentially; like = and Then select the Tth generation coordinating node population. SVM model corresponding to the global optimal individual As the final face image classification model, it is used to detect the collected target face image data and obtain the target face image classification result.

[0008] The face image detection method based on distributed optimized instance selection described in this invention is also characterized in that step S9 includes: S9.1, Calculate using formula (1) The Middle individual and the j-th working node In the The nth global iteration Generation population The Middle individual The degree of conflict between , : (1) In equation (1), and They represent and The number of face image samples in the sample; express and The number of conflicting face image sample pairs in the selected set of face image samples; S9.2, Coordinating Node CN: Based on the degree of conflict, from... The top N individuals with the lowest conflict levels are selected first. .

[0009] Furthermore, the globally optimal individual and the mutation parameters based on global preferences in step 12 are obtained through the following steps: S12.1 Execute non-dominated sorting and environment selection strategies to obtain the t-th global iteration. Pareto optimal solution set of the generation ,in, Let represent the k-th optimal solution individual, and K represent the number of individuals in the optimal solution set. ; S12.2, Statistics middle The number of occurrences of the value "1" on each chromosome is obtained. The frequency of each chromosome; S12.3 Setting a redundancy judgment threshold and lower threshold The baseline mutation probability of bit flipping changes from 0 to 1. The probability of mutation from 1 to 0 and the upper threshold adjustment coefficient Lower threshold adjustment coefficient and the lower world ; If the frequency of the w-th chromosome Then, the face image sample corresponding to the w-th chromosome is marked as a key instance. Update the mutation probability bits of the w-th chromosome from 1 to 0. and the mutation probability bits of the w-th chromosome from 0 to 1 ; like Then, the face image sample corresponding to the w-th chromosome is marked as a redundant instance. Update the mutation probability of chromosome w from 1 to 0 as follows: and the mutation probability bits of the w-th chromosome from 0 to 1 Thus obtain The probability of a mutation from 0 to 1 on each chromosome And the mutation probability from 1 to 0 ; S12.4, will In Each chromosome location is mapped to the i-th working node. of Each chromosome location, thus Mapped to each working node At each chromosome position and as the globally optimal individual, and Mapped to The mutation probability at each chromosome location is used as a mutation parameter based on global preferences, thus forming feedback information.

[0010] Furthermore, the uniform crossover and preference-based mutation operations performed on each individual in step 5 include: S5.1 If feedback information is received from the coordinating node, then... A uniform crossover operation is performed on every two parent individuals to produce the first generation. The nth global iteration Intermediate offspring population ; S5.2, Traversal For each binary chromosome of each offspring individual, obtain the mutation probability (from 0 to 1 and from 1 to 0) for the corresponding position from the feedback information. Use this information to perform mutation operations on each binary chromosome and generate the mutation at the [missing information - likely the first chromosome number]. The nth global iteration Offspring population : When the The feedback information obtained from the binary chromosome is the first one. The mutation probability from 0 to 1 at each position is: The mutation probability from 1 to 0 is hour, ; If the first If a binary chromosome is "0", then the probability is... The first The first binary chromosome is flipped to "1"; if the second binary chromosome is flipped to "1"; If each binary chromosome is "1", then the probability is... The first Each binary chromosome is flipped to "0".

[0011] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the image detection method, and the processor is configured to execute the program stored in the memory.

[0012] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to perform the steps of the face image detection method.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention proposes an optimization strategy for a face image SVM classification model based on distributed optimization. By employing multi-objective evolutionary instance selection and binary chromosome encoding at each working node, local search and filtering are completed without transmitting the original face data. Combined with two-stage information aggregation (splitting aggregation and aggregation based on conflict instances) at the coordinating node, the search range of high-quality sample subsets is effectively expanded and cross-node boundary sample redundancy is reduced, thereby improving the convergence speed of the algorithm iteration and enabling the population to quickly approach the first frontier. This results in a face image SVM classification model with higher classification accuracy and higher compression ratio, improving the training and prediction efficiency of image recognition.

[0014] 2. This invention designs a distributed population aggregation strategy. Based on the unconstrained optimization in the previous stage, a parent population that better meets the low-conflict constraint is selected according to the solution selection operator based on conflict degree. More feasible solution offspring are generated through the global crossover operator and the mutation operator based on global preference, so that the obtained face image SVM classification model can have better classification performance on face image datasets with multiple distributions and multiple scenes.

[0015] 3. By simultaneously optimizing two objectives (classification accuracy ACC and compression ratio RED), this invention can achieve a better balance between face image sample selection and SVM model training. Under the constraints of multi-node and non-centralized data, the obtained face image SVM classification model shows better overall performance in terms of classification performance, training / prediction latency, and communication and privacy costs. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] In this embodiment, a face image detection method based on distributed optimization instance selection is a face image detection scheme that combines a distributed architecture with optimized instance selection. It aims to solve the technical problems of long training time, reduced accuracy due to redundant data, and performance limitations in distributed storage scenarios when traditional SVM classification models process large-scale face data. This enables efficient screening of face image data, distributed collaborative training, and improves the efficiency, accuracy, and adaptability to distributed environments of face image detection. Specifically, as... Figure 1 As shown, this method is applied to a network environment consisting of 4 worker nodes and 1 coordinating node, and includes the following steps: S1. Obtain the i-th working node. facial image sample set ,in, Represents the i-th working node The wth face image sample, For the i-th working node The set of attribute features of the w-th face image sample, and , express The d-th attribute feature, d∈ D is the total number of attribute features of a face image sample. express Category tags, ∈ C represents the total number of categories; For the i-th working node The total number of face image samples; i M is the total number of worker nodes, and 4 is assigned to M.

[0018] In this embodiment, to verify the effectiveness of the method in a distributed environment, we constructed a simulated distributed storage environment. The data used was the face_image dataset, which contains 6302 face images with labels... These represent different facial features. It means eyes. It represents the mouth. It refers to the nose. Indicates skin color, 5 represents the ear; to simulate uneven data distribution and privacy isolation in real-world scenarios, we randomly and unevenly distribute facial image data across 4 working nodes: It has 2,000 facial image data. It has 1,000 facial image data. It has 1,000 facial image data. It has 2,302 facial image data.

[0019] The complete set of face image samples for the coordination node CN is composed of the face image sample sets from the four working nodes. ,in, Indicates the first Personal facial image samples, For the coordination node The set of attribute features of individual face image samples, and Include Individual attribute features; express Category tags, ; To coordinate the total number of face image samples at node CN, and .

[0020] The population that constructs the coordinating node CN includes Each individual contains 100, which is assigned to N. There are 10 binary chromosomes, each of which corresponds to 10 binary chromosomes. If a face image sample in the dataset has a binary chromosome value of "1", it means that the face image sample at the corresponding position is selected; if it is "0", it means that the face image sample at the corresponding position is not selected.

[0021] S2. Define the global current iteration count as t, and initialize it. =1; Define the global maximum number of iterations as T, and assign 100 to T; S3. Define the current algebra of the working node as... and initialize =1; Define the maximum number of iterations for a worker node as . Assign 20 to .

[0022] Define and randomly initialize In the The nth global iteration Generation population },in, express The nth individual represents a face image sample selection scheme, and Include There are 3 binary chromosomes, each representing whether to select a face image sample at the corresponding location. The Middle If each binary chromosome is "1", it indicates selection. ,like The Middle If a binary chromosome is "0", it means no selection is made. N is the population size.

[0023] S4, will The corresponding subset of face image samples is input into the SVM classifier for training, to obtain the first... The nth global iteration The nth face image classification model and calculate Classification accuracy and compression ratio Thus constitute objective function value ; S5, if Upon receiving feedback from the coordinating node, then... Each individual in the process undergoes uniform crossover and preference-based mutation operations to produce... In the The nth global iteration Offspring population Otherwise, for Each individual in the process undergoes uniform crossover and single-point mutation operations to produce... In the The nth global iteration Offspring population .

[0024] S5.1 If feedback information is received from the coordinating node, then the population... A uniform crossover operation is performed between every two parent individuals to produce an intermediate offspring population. ; S5.2, Regarding the intermediate offspring population Traversing each individual's binary chromosome Obtain the mutation probability from 0 to 1 corresponding to that position in the feedback information. The probability of mutation from 1 to 0 If the current chromosome position is "0", then proceed with probability. Flip to "1"; if the current chromosome position is "1", then use probability... Flip to "0"; after completing the mutation operation for all individuals, the result is generated in the first... The nth global iteration Offspring population .

[0025] S6, will and After performing a non-dominated sort, the top N individuals are selected from the sorted population using a binary tournament strategy to form the population. In the +1 global iterations of the th +1 generation population ; S7, will Assign to ,like = Then In the The nth global iteration Generation population Send the values ​​to the coordinating node CN and execute step S8; otherwise, assign t+1 to t and return to S4 to execute sequentially.

[0026] In this embodiment, steps S6 and S7 employ the classic environment selection strategy of the NSGA-II algorithm. Through non-dominated sorting, individuals are stratified (Rank 1, Rank 2...), with priority given to retaining individuals in Rank 1 (i.e., those located on the Pareto front). After 20 local iterations, the worker node uploads the optimized population to the coordinator node, proceeding to the next step.

[0027] S8. Define the current algebra of the coordinating node CN as: and initialize =1; Define the maximum number of iterations for the coordinating node CN as 1. Assign 20 to ; S9, if t < ,Will Merged into the first The nth global iteration Generation population },in, express The nth individual represents a face image sample selection scheme, and Include There are 3 binary chromosomes, each representing whether to select a face image sample at the corresponding location. The Middle If each binary chromosome is "1", it indicates selection. ,like The Middle If a binary chromosome is "0", it means no selection is made. Otherwise, the coordinating node uses an aggregation strategy based on conflict instances to generate the first... The nth global iteration Generation population .

[0028] S9.1, Calculate using formula (1) The Middle individual and the j-th working node In the The nth global iteration Generation population The Middle individual The degree of conflict between , : (1) In equation (1), and They represent and The number of face image samples in the sample; express and The number of conflicting face image sample pairs in the selected face image sample set; conflicting samples refer to individuals that are close in Euclidean distance in the special space of the face image dataset.

[0029] S9.2, Coordinating Node CN: Based on the degree of conflict, from... The top N individuals with the lowest conflict levels are selected first. .

[0030] S10, Coordinating Node CN will The nth individual The corresponding subset of face image samples is input into the SVM classifier for training, to obtain the first... The nth global iteration The nth face image classification model ,and exist On the precision and compression ratio Thus constitute objective function value Therefore, based on the objective function value, for Each individual in the process undergoes uniform crossover and single-point mutation operations to produce a coordinating node CN in the th... The nth global iteration Offspring population .

[0031] S11, will and After performing a non-dominated sort, the top N individuals are selected from the sorted population using a binary tournament strategy to form the population. In the The nth global iteration +1 generation population ; S12, will Assign to ,like After assigning t+1 to t, return to S9 to execute; like = and CN in the The nth global iteration Generation population The globally optimal individual and the mutation parameters based on global preferences are sent as feedback information to the (t+1)th global iteration. The Generation population After assigning t+1 to t, return to S4 to execute sequentially.

[0032] S12.1, to Execute non-dominated sorting and environment selection strategies to obtain the t-th global iteration. Pareto optimal solution set of the generation ,in, Let represent the k-th optimal solution individual, and K represent the number of individuals in the optimal solution set. ; S12.2, Statistics middle The number of occurrences of the value "1" on each chromosome is obtained. The frequency of each chromosome; S12.3 Setting a redundancy judgment threshold and lower threshold The baseline mutation probability of bit flipping changes from 0 to 1. The probability of mutation from 1 to 0 and the upper threshold adjustment coefficient and the lower world In this embodiment, 0.8 is assigned to the upper threshold. Assign 0.2 to the lower threshold. Assign 0.2 to the upper threshold adjustment coefficient. Assign 0.15 to Assign 0.05 to the lower bound. ; If the frequency of the w-th chromosome Then, the face image sample corresponding to the w-th chromosome is marked as a key instance. Update the mutation probability bits of the w-th chromosome from 1 to 0. and the mutation probability bits of the w-th chromosome from 0 to 1 .

[0033] like Then, the face image sample corresponding to the w-th chromosome is marked as a redundant instance. Update the mutation probability of chromosome w from 1 to 0 as follows: and the mutation probability bits of the w-th chromosome from 0 to 1 Thus obtain The probability of a mutation from 0 to 1 on each chromosome And the mutation probability from 1 to 0 ; S12.4, will In Each chromosome location is mapped to the i-th working node. of Each chromosome location, thus Mapped to each working node At each chromosome position and as the globally optimal individual, and Mapped to The mutation probability at each chromosome location is used as a mutation parameter based on global preferences, thus forming feedback information.

[0034] like = and Then select the Tth generation coordinating node population. SVM model corresponding to the global optimal individual As the final face image classification model, it is used to detect the collected target face image data and obtain the target face image classification result.

[0035] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the methods described above, and the processor is configured to execute the program stored in the memory.

[0036] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

[0037] To verify the effectiveness of the present invention, we conducted a comparative experiment on the face_image dataset using the method of the present invention and the traditional SVM classification method. The results are shown in Table 1.

[0038] Table 1

[0039] Compared with traditional SVM classification methods, the method of this invention has higher accuracy in face image classification and better compression ratio for face image data. Unlike traditional SVM classification methods that use all face image data to train the SVM classification model, the method of this invention uses partial instance data to train the SVM classification model, resulting in higher data compression ratio and thus shorter prediction time. In the current era of rapid development in face recognition technology and the continuous increase in face image data, the method of this invention has significantly improved the distributed storage of face image data and the prediction time, effectively improving the effect and efficiency of face image data processing.

Claims

1. A face image detection method based on distributed optimized instance selection, characterized in that, This is applied to a network environment consisting of M worker nodes and 1 coordinating node, and includes the following steps: S1. Obtain the i-th working node. facial image sample set ,in, Represents the i-th working node The wth face image sample, For the i-th working node The set of attribute features of the w-th face image sample, and , express The d-th attribute feature, d∈ D is the total number of attribute features of a face image sample. express Category tags, ∈ C represents the total number of categories; For the i-th working node The total number of face image samples; i M is the total number of worker nodes; The complete set of face image samples of the coordination node CN is composed of the face image sample sets of M working nodes. ,in, Indicates the first Personal facial image samples, For the coordination node The set of attribute features of individual face image samples, and Include Individual attribute features; express Category tags, ; To coordinate the total number of face image samples at node CN, and ; The population that constructs the coordinating node CN includes There are 10 individuals, and each individual contains 10 individuals. There are 10 binary chromosomes, each corresponding to 10 binary chromosomes. If the value of the binary chromosome of a face image sample is "1", it means that the face image sample at the corresponding position is selected; if it is "0", it means that the face image sample at the corresponding position is not selected. S2. Define the global current iteration count as t, and initialize it. =1; Define the global maximum number of iterations as T; S3. Define the current algebra of the working node as... and initialize =1; Define the maximum number of iterations for a worker node as . ; Define and randomly initialize In the The nth global iteration Generation population },in, express The nth individual represents a face image sample selection scheme, and Include There are 3 binary chromosomes, each representing whether to select a face image sample at the corresponding location. The Middle If each binary chromosome is "1", it indicates selection. ,like The Middle If a binary chromosome is "0", it means no selection is made. N is the population size; S4, will The corresponding subset of face image samples is input into the SVM classifier for training, to obtain the first... The nth global iteration The nth face image classification model and calculate Classification accuracy and compression ratio Thus constitute objective function value ; S5, if Upon receiving feedback from the coordinating node, then... Each individual in the process undergoes uniform crossover and preference-based mutation operations to produce... In the The nth global iteration Offspring population Otherwise, for Each individual in the process undergoes uniform crossover and single-point mutation operations to produce... In the The nth global iteration Offspring population ; S6, will and After performing a non-dominated sort, the top N individuals are selected from the sorted population using a binary tournament strategy to form the population. In the +1 global iterations of the th +1 generation population ; S7, will Assign to ,like = Then In the The nth global iteration Generation population Send the values ​​to the coordinating node CN and execute step S8; otherwise, assign t+1 to t and return to S4 to execute sequentially. S8. Define the current algebra of the coordinating node CN as: and initialize =1; Define the maximum number of iterations for the coordinating node CN as 1. ; S9, if t < ,Will Merged into the first The nth global iteration Generation population },in, express The nth individual represents a face image sample selection scheme, and Include There are 3 binary chromosomes, each representing whether to select a face image sample at the corresponding location. The Middle If each binary chromosome is "1", it indicates selection. ,like The Middle If a binary chromosome is "0", it means no selection is made. Otherwise, the coordinating node uses an aggregation strategy based on conflict instances to generate the first... The nth global iteration Generation population ; S10, Coordination Node CN will The nth individual The corresponding subset of face image samples is input into the SVM classifier for training, to obtain the first... The nth global iteration The nth face image classification model ,and exist On the precision and compression ratio Thus constitute objective function value Therefore, based on the objective function value, for Each individual in the process undergoes uniform crossover and single-point mutation operations to produce a coordinating node CN in the th... The nth global iteration Offspring population ; S11, will and After performing a non-dominated sort, the top N individuals are selected from the sorted population using a binary tournament strategy to form the population. In the The nth global iteration +1 generation population ; S12, will Assign to ,like After assigning t+1 to t, return to S9 to execute; like = and CN in the The nth global iteration Generation population The globally optimal individual and the mutation parameters based on global preferences are sent as feedback information to the (t+1)th global iteration. The Generation population After assigning t+1 to t, return to S4 and execute sequentially; like = and Then select the Tth generation coordinating node population. SVM model corresponding to the global optimal individual As the final face image classification model, it is used to detect the collected target face image data and obtain the target face image classification result.

2. The face image detection method based on distributed optimized instance selection according to claim 1, characterized in that, S9 includes: S9.1, Calculate using formula (1) The Middle individual and the j-th working node In the The nth global iteration Generation population The Middle individual The degree of conflict between , : (1) In equation (1), and They represent and The number of face image samples in the sample; express and The number of conflicting face image sample pairs in the selected set of face image samples; S9.2, Coordinating Node CN: Based on the degree of conflict, from... The top N individuals with the lowest conflict levels are selected first. .

3. The face image detection method based on distributed optimized instance selection according to claim 1, characterized in that, The globally optimal individual and the mutation parameters based on global preferences in step 12 are obtained as follows: S12.1 Execute non-dominated sorting and environment selection strategies to obtain the t-th global iteration. Pareto optimal solution set of the generation ,in, Let represent the k-th optimal solution individual, and K represent the number of individuals in the optimal solution set. ; S12.2, Statistics middle The number of occurrences of the value "1" on each chromosome is obtained. The frequency of each chromosome; S12.3 Setting a redundancy judgment threshold and lower threshold The baseline mutation probability of bit flipping changes from 0 to 1. The probability of mutation from 1 to 0 and the upper threshold adjustment coefficient Lower threshold adjustment coefficient and the lower world ; If the frequency of the w-th chromosome Then, the face image sample corresponding to the w-th chromosome is marked as a key instance. Update the mutation probability bits of the w-th chromosome from 1 to 0. and the mutation probability bits of the w-th chromosome from 0 to 1 ; like Then, the face image sample corresponding to the w-th chromosome is marked as a redundant instance. Update the mutation probability of chromosome w from 1 to 0 as follows: and the mutation probability bits of the w-th chromosome from 0 to 1 Thus obtain The probability of a mutation from 0 to 1 on each chromosome And the mutation probability from 1 to 0 ; S12.4, will In Each chromosome location is mapped to the i-th working node. of Each chromosome location, thus Mapped to each working node At each chromosome position and as the globally optimal individual, and Mapped to The mutation probability at each chromosome location is used as a mutation parameter based on global preferences, thus forming feedback information.

4. The face image detection method based on distributed optimized instance selection according to claim 3, characterized in that, The uniform crossover and preference-based mutation operations performed on each individual in step 5 include: S5.1 If feedback information is received from the coordinating node, then... A uniform crossover operation is performed on every two parent individuals to produce the first generation. The nth global iteration Intermediate offspring population ; S5.2, Traversal For each binary chromosome of each offspring individual, obtain the mutation probability (from 0 to 1 and from 1 to 0) for each position from the feedback information. Use this information to perform mutation operations on each binary chromosome and generate the mutation at the [missing information - likely the first or second generation chromosome]. The nth global iteration Offspring population : When the The feedback information obtained from the binary chromosome is the first one. The mutation probability from 0 to 1 at each position is: The mutation probability from 1 to 0 is hour, ; If the first If a binary chromosome is "0", then the probability is... The first The first binary chromosome is flipped to "1"; if the second binary chromosome is flipped to "1"; If each binary chromosome is "1", then the probability is... The first Each binary chromosome is flipped to "0".

5. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing any of the image detection methods of claims 1-4, and the processor is configured to execute the program stored in the memory.

6. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when run by a processor, performs the steps of any of the face image detection methods described in claims 1-4.

Images