A face recognition method for missing tag data at the user side

Abstract

The present invention provides a face recognition method for missing label data of the user side, including selecting an initial model and determining a sample selection strategy. Supplementary samples are screened out from the unlabeled sample set based on the sample selection strategy to obtain a supplementary sample set. The supplementary sample set is added to the labeled sample set to obtain a first training set. The first training set includes labeled samples and unlabeled samples, and the initial model is trained hybridly using the first training set to obtain a face recognition model. In semi-supervised training, first, the labeled samples are used to learn the mapping relationship between the input features and the output labels, and then the sample types of the unlabeled samples are predicted. The unlabeled samples contain a large amount of unstructured or unannotated data. The present invention combines the labeled samples and the unlabeled samples, which can solve the problem of falling into a local optimal solution when using active learning alone, accelerate the convergence process of the initial model in the case of lack of label data, and thus improve the efficiency and accuracy of training the initial model.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and particularly to a face recognition method for missing label data at the user end. Background Art

[0002] Existing face recognition methods have certain limitations when label data is missing. Traditional face recognition methods include algorithms based on feature extraction and machine learning models. Algorithms based on feature extraction include the LBPH algorithm and the HOG algorithm, and machine learning models include the CNN model. Algorithms based on feature extraction and machine learning models usually rely on a complete label data set for training. In the case of missing label data, the accuracy of face recognition will be greatly reduced.

[0003] Although deep learning models have achieved good results in face recognition, deep learning models also rely on a large amount of labeled data. When the label data is insufficient, the performance of deep learning models will be significantly affected. In addition, some data completion methods, such as generative adversarial networks and autoencoders, although they can generate missing label data, these generated labels are often not accurate enough and may introduce noise, affecting the accuracy of recognition results. Unsupervised learning and semi-supervised learning methods try to learn through less labeled or unlabeled data, but in the absence of labels, the accuracy of face recognition is not high, and there are generally problems with poor robustness. In addition, deep learning and semi-supervised learning usually require a large amount of computing resources and time, which limits the real-time application of deep learning and semi-supervised learning. Generally speaking, the existing technology not only depends on the effect of data completion when label data is missing, but also faces problems such as high computational complexity, low data completion accuracy, and lack of a unified theoretical framework. These problems limit the popularization and application of existing face recognition algorithms in practical applications.

[0004] Therefore, a face recognition method is needed that can reduce the amount of data that needs to be labeled, select the most informative samples to gradually improve the face recognition ability of the face recognition model, and has high computational efficiency. Summary of the Invention

[0005] To overcome the problems existing in the related art, the purpose of the present invention is to provide a face recognition method for missing label data at the user end, which can reduce the amount of data that needs to be labeled, select the most informative samples to gradually improve the face recognition ability of the face recognition model, and has high computational efficiency.

[0006] A face recognition method for missing label data at the user end includes:

[0007] Select an initial model and determine a sample selection strategy;

[0008] Select supplementary samples from the unlabeled sample set based on the sample selection strategy to obtain a supplementary sample set;

[0009] Add the supplementary sample set to the labeled sample set to obtain a first training set; the first training set includes labeled samples and unlabeled samples;

[0010] Use the first training set to perform hybrid training on the initial model to obtain a face recognition model; wherein, the hybrid training includes alternating iterative active learning and semi-supervised training;

[0011] Use the face recognition model to perform face recognition on the user's face.

[0012] In a preferred technical solution of the present invention, the using the first training set to perform hybrid training on the initial model to obtain a face recognition model includes:

[0013] Use the first training set to perform semi-supervised training on the initial model to generate a first set of pseudo-labeled samples;

[0014] Annotate some of the pseudo-labeled samples in the first set of pseudo-labeled samples to obtain a first updated set of pseudo-labeled samples;

[0015] Add the first updated set of pseudo-labeled samples to the first training set to obtain a second training set;

[0016] Use the second training set to perform semi-supervised training on the initial model to generate a second set of pseudo-labeled samples.

[0017] In a preferred technical solution of the present invention, the selecting the initial model includes:

[0018] Select one model from a neural network model, a decision tree model, a support vector machine model, and a logistic regression model as the initial model.

[0019] In a preferred technical solution of the present invention, the determining the sample selection strategy includes:

[0020] Select one sampling method from uncertainty sampling, diversity sampling, and information entropy sampling as the sample selection strategy.

[0021] In a preferred technical solution of the present invention, the selecting supplementary samples from the unlabeled sample set based on the sample selection strategy to obtain a supplementary sample set includes:

[0022] If the sample selection strategy is uncertainty sampling, select an uncertainty index;

[0023] Select supplementary samples from the unlabeled sample set according to the uncertainty index;

[0024] Take all the supplementary samples as a supplementary sample set.

[0025] In a preferred technical solution of the present invention, the screening of supplementary samples from the unlabeled sample set according to the uncertainty index includes:

[0026] If the uncertainty index is entropy, sort the unlabeled samples in descending order according to the magnitude of the entropy to obtain an entropy sequence;

[0027] Take the unlabeled samples corresponding to the first entropy to the K1-th entropy in the entropy sequence as supplementary samples.

[0028] In a preferred technical solution of the present invention, the screening of supplementary samples from the unlabeled sample set according to the uncertainty index includes:

[0029] If the uncertainty index is confidence, sort the unlabeled samples in ascending order according to the magnitude of the confidence to obtain a confidence sequence;

[0030] Take the unlabeled samples corresponding to the first confidence to the K2-th confidence in the confidence sequence as supplementary samples.

[0031] In a preferred technical solution of the present invention, the screening of supplementary samples from the unlabeled sample set based on the sample selection strategy to obtain a supplementary sample set includes:

[0032] If the sample selection strategy is diversity sampling, use clustering or maximizing the margin method to screen supplementary samples from the unlabeled sample set;

[0033] Take all the supplementary samples as a supplementary sample set.

[0034] In a preferred technical solution of the present invention, after obtaining the first updated pseudo-labeled sample set, it further includes:

[0035] If the number of labeled pseudo-labeled samples is greater than or equal to the labeled sample threshold, or during the training of the initial model, the face recognition accuracy has converged, stop training the initial model to obtain a face recognition model.

[0036] In a preferred technical solution of the present invention, the semi-supervised training of the initial model using the first training set to generate a first pseudo-labeled sample set includes:

[0037] Input the labeled samples in the first training set into the initial model so that the initial model learns the mapping relationship between the input features and the output labels;

[0038] Input the unlabeled samples in the first training set into the initial model, predict the unlabeled samples to obtain a predicted output;

[0039] Compare the N predicted outputs with the corresponding actual outputs respectively to obtain the label matching degree;

[0040] Adjust the model parameters of the initial model according to the label matching degree.

[0041] The beneficial effects of the present invention are as follows:

[0042] The face recognition method for missing label data at the user end provided by the present invention includes selecting an initial model and determining a sample selection strategy. Based on the sample selection strategy, supplementary samples are screened out from the unlabeled sample set to obtain a supplementary sample set. The present invention is based on active learning and uses the sample selection strategy to select the most informative unlabeled samples with a high uncertainty of categories. Add the supplementary sample set to the labeled sample set to supplement the labeled sample set and obtain a first training set. The first training set contains both labeled samples and unlabeled samples. Use the first training set to perform hybrid training on the initial model to obtain a face recognition model; wherein, the hybrid training includes alternating iterative active learning and semi-supervised training. In semi-supervised training, first use the labeled samples to learn the mapping relationship between the input features and the output labels, and then the initial model predicts the unlabeled samples based on this mapping relationship to obtain the sample types of the unlabeled samples. Compare the predicted output label categories with the actual output label categories, and continuously update the model parameters of the initial model according to the comparison results. When the model parameters converge or the number of training times reaches the training times threshold, a face recognition model is obtained. Unlabeled samples contain a large amount of unstructured or unannotated data. By using unlabeled data, semi-supervised learning can improve the accuracy of face recognition of the initial model and save the time consumed for annotating samples at the same time. The present invention combines labeled samples and unlabeled samples, which can solve the problems of getting stuck in local optimal solutions when using active learning alone, resulting in low efficiency and accuracy of the learning process, and the problem that active learning requires a large number of correctly labeled samples. Active learning can focus on those samples that are the most difficult to classify or the most representative, thereby accelerating the convergence process of the initial model and improving the accuracy of the initial model in the case of scarce labeled data. In terms of semi-supervised learning, the present invention further optimizes the utilization of unlabeled data by combining the effective labels obtained by active learning, and improves the training effect of the model. Description of the Drawings

[0043] Figure 1 is a flowchart of the face recognition method for missing label data at the user end of the present invention;

[0044] Figure 2 is a flowchart of hybrid training of the initial model by combining active learning and semi-supervised training according to the present invention. Detailed Embodiments

[0045] The preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0046] Example 1

[0047] As Figure 1 shown, this embodiment provides a face recognition method for missing tag data at the user end, including:

[0048] S1: Select an initial model and determine a sample selection strategy.

[0049] S2: Screen out supplementary samples from the unlabeled sample set based on the sample selection strategy to obtain a supplementary sample set.

[0050] S3: Add the supplementary sample set to the labeled sample set to obtain a first training set; the first training set includes labeled samples and unlabeled samples.

[0051] S4: Use the first training set to perform hybrid training on the initial model to obtain a face recognition model; wherein, the hybrid training includes alternating iterative active learning and semi-supervised training.

[0052] S5: Use the face recognition model to perform face recognition on the user's face.

[0053] The user end of the present invention can be a face recognition attendance machine or a face recognition security monitoring system, which is not limited herein.

[0054] The selection of the initial model includes:

[0055] Select one model from a neural network model, a decision tree model, a support vector machine model, and a logistic regression model as the initial model.

[0056] The determination of the sample selection strategy includes:

[0057] Select one sampling method from uncertainty sampling, diversity sampling, and information entropy sampling as the sample selection strategy.

[0058] The initial model can also be a support vector machine model, a k-nearest neighbor model, a naive Bayes model, a random forest model, a linear regression model, or a logistic regression model. In this embodiment, the initial model is a neural network model as an example.

[0059] Screening supplementary samples from the unlabeled sample set based on the sample selection strategy to obtain a supplementary sample set, including:

[0060] S21: If the sample selection strategy is uncertainty sampling, select the uncertainty index.

[0061] S22: Screen supplementary samples from the unlabeled sample set according to the uncertainty index.

[0062] S23: Use all the supplementary samples as the supplementary sample set.

[0063] Uncertainty sampling is to select those samples with uncertain or incorrect classification results under the current model, and use some indicators to measure the uncertainty of samples, such as entropy and confidence, to sort the samples in ascending or descending order, and then select a part of the samples ranked in the front as the training set. Common methods of uncertainty sampling include minimizing the margin, information entropy, and entropy reduction, etc.

[0064] The purpose of diversity sampling is to select samples with different characteristics from the data set as much as possible. If the sample selection strategy is diversity sampling, cluster or maximize the margin to screen supplementary samples from the unlabeled sample set, and use all the supplementary samples as the supplementary sample set.

[0065] Information entropy sampling: Information entropy sampling is a method based on uncertainty sampling, which selects the most valuable samples by calculating the information entropy of each sample. Calculate the classification entropy of each sample under the current model. The larger the entropy value, the more uncertain the classification result of the sample, and the more necessary it is to add the sample to the labeled sample set.

[0066] The screening of supplementary samples from the unlabeled sample set according to the uncertainty index includes:

[0067] S221: If the uncertainty index is entropy, sort the unlabeled samples in descending order according to the size of the entropy to obtain an entropy sequence.

[0068] S222: Use the unlabeled samples corresponding to the first entropy to the K1-th entropy in the entropy sequence as supplementary samples.

[0069] Entropy is a function describing the state of a system, and entropy is used to measure the degree of chaos of the system. The larger the entropy value, the more uncertain the classification result of the sample, and the more necessary it is to add it to the training set. Therefore, sort the unlabeled samples in descending order according to the size of the entropy to obtain an entropy sequence. The earlier the entropy in the entropy sequence, the more additional information the unlabeled sample corresponding to the entropy can bring. Use the unlabeled samples corresponding to the first entropy to the K1-th entropy in the entropy sequence as supplementary samples.

[0070] Preferably, the value of K1 is set according to the total number of entropies in the entropy sequence. Multiply the total number of entropies in the entropy sequence by 20%. If the product is a decimal, round it to the nearest integer using the rounding method. For example, if there are 100 entropies in the entropy sequence, that is, the number of unlabeled samples is 100, the first 20 entropies in the entropy sequence correspond to unlabeled samples as supplementary samples.

[0071] Screening out supplementary samples from the unlabeled sample set according to the uncertainty index includes:

[0072] If the uncertainty index is confidence, sort the unlabeled samples in ascending order according to the magnitude of the confidence to obtain a confidence sequence;

[0073] Take the unlabeled samples corresponding to the first confidence to the K2 - th confidence in the confidence sequence as supplementary samples.

[0074] The lower the confidence, the greater the additional information that the corresponding unlabeled sample can bring, and the more the unlabeled sample should be added to the labeled sample set. Therefore, sort the unlabeled samples in ascending order according to the confidence. The confidences of the first K2 confidences in the confidence sequence are relatively low. Take the K2 unlabeled samples with lower confidences as supplementary samples.

[0075] Preferably, before screening out supplementary samples from the unlabeled sample set based on the sample selection strategy, it also includes constructing an unlabeled sample set, which contains real unlabeled samples and simulated unlabeled samples. The simulated unlabeled samples are obtained by performing a non - linear transformation on the real unlabeled samples. Calculate the mean value of P real unlabeled samples to obtain the average face, where P≥2. Use the following formula to perform a non - linear transformation on the average face:

[0076] ;

[0077] where, K i is the i - th simulated unlabeled sample, AVG i is the i - th average face, MAX is the set maximum pixel value, and the maximum pixel value can be 255 or other values, which is not limited here.

[0078] Performing a non - linear transformation on the average face can highlight the facial region with moderate pixel values, such as the facial region with an average value of 128, while suppressing the facial regions with too low and too high pixel values, so that the simulated unlabeled samples have facial features not included in the real unlabeled samples, thereby supplementing the real unlabeled samples. Add the simulated unlabeled samples to the real unlabeled samples to obtain the unlabeled sample set.

[0079] In security monitoring, face recognition technology is usually used for automated identity authentication, abnormal behavior detection, intrusion warning, etc. Due to the large volume of monitoring video data and the high proportion of unlabeled data, traditional supervised learning methods may face the problem of scarce labeled data. At this time, active learning and semi-supervised training can effectively alleviate this problem.

[0080] In this embodiment, a small amount of labeled data, such as face images of people with known identities, is used to train the initial model, and then active learning is used to select the most uncertain samples for annotation. When the monitoring system encounters an unseen face, active learning will select the most difficult-to-classify samples, such as face images under certain specific angles and specific lighting conditions, and then manually annotate these highly uncertain samples. As the amount of monitoring data increases, semi-supervised learning can use a large number of unlabeled face images to train the initial model without manual annotation. By extracting face features from unlabeled samples, the recognition effect can be improved and the dependence on a large number of labeled samples can be reduced.

[0081] The face recognition method for missing user - end label data provided in this embodiment includes selecting an initial model and determining a sample selection strategy. Based on the sample selection strategy, supplementary samples are screened out from the unlabeled sample set to obtain a supplementary sample set. The present invention is based on active learning and uses the sample selection strategy to select the unlabeled samples with the most information content and high uncertainty in categories. The supplementary sample set is added to the labeled sample set to supplement the labeled sample set, obtaining a first training set. The first training set contains both labeled samples and unlabeled samples. The initial model is subjected to hybrid training using the first training set to obtain a face recognition model; among them, the hybrid training includes alternating iterative active learning and semi - supervised training. In semi - supervised training, first, the labeled samples are used to learn the mapping relationship between the input features and the output labels, and then the initial model predicts the unlabeled samples based on this mapping relationship to obtain the sample types of the unlabeled samples. The predicted label categories are compared with the actual output label categories, and the model parameters of the initial model are continuously updated according to the comparison results. When the model parameters converge or the number of training times reaches the training - time threshold, a face recognition model is obtained. Unlabeled samples contain a large amount of unstructured or unannotated data. By utilizing unlabeled data, semi - supervised learning can improve the accuracy of face recognition of the initial model while saving the time consumed for annotating samples. The present invention combines labeled samples and unlabeled samples, which can solve the problem that using only active learning may fall into a local optimal solution, resulting in low efficiency and accuracy of the learning process, and the problem that active learning requires a large number of correctly labeled samples. Active learning can focus on the most difficult - to - classify or most representative samples, thereby accelerating the convergence process of the initial model and improving the accuracy of the initial model in the case of scarce labeled data. In terms of semi - supervised learning, the present invention further optimizes the utilization of unlabeled data by combining the effective labels obtained from active learning, improving the training effect of the model.

[0082] Embodiment 2

[0083] This embodiment provides a face recognition method for missing user - end label data. This embodiment only describes the differences from Embodiment 1, as Figure 2 shown, using the first training set to perform hybrid training on the initial model to obtain a face recognition model includes:

[0084] S41: Perform semi - supervised training on the initial model using the first training set to generate a first pseudo - label sample set.

[0085] S42: Label some of the pseudo - label samples in the first pseudo - label sample set to obtain a first updated pseudo - label sample set.

[0086] S43: Add the first updated pseudo - label sample set to the first training set to obtain a second training set.

[0087] S44: Use the second training set to perform semi-supervised training on the initial model to generate a second set of pseudo-label samples.

[0088] In traditional supervised learning, the algorithm only uses labeled data with known outputs or labels. However, in many real-world scenarios, the labeled data may be scarce or costly to obtain, and only a large amount of unlabeled data can be used. Semi-supervised learning is a type of machine learning that uses both labeled and unlabeled data to train a neural network model.

[0089] In the active learning stage, a small number of supplementary samples in the first training set are labeled. In the semi-supervised learning stage, first, the labeled samples are used to learn the mapping relationship between the input features and the output labels, and then, based on the mapping relationship, the unlabeled samples are predicted to generate pseudo-label samples. The labels of the pseudo-label samples are compared with the actual labels of the pseudo-label samples, and the model parameters of the initial model are adjusted according to the comparison results to optimize the performance of the initial model.

[0090] Use semi-supervised learning to generate a first set of pseudo-label samples, which contains multiple pseudo-label samples. The pseudo-label samples in the first set of pseudo-label samples are labeled through active learning to obtain a first updated set of pseudo-label samples.

[0091] After obtaining the first updated set of pseudo-label samples, it further includes:

[0092] If the number of labeled pseudo-label samples is greater than or equal to the annotation sample threshold, or during the training process of the initial model, the face recognition accuracy has converged, then stop active learning.

[0093] If the number of labeled pseudo-label samples is less than the annotation sample threshold and the face recognition accuracy of the initial model has not converged, then continue active learning. Execute steps S43 - S44. After step S44, some of the pseudo-label samples in the second set of pseudo-label samples are labeled to obtain a second updated set of pseudo-label samples. The second updated set of pseudo-label samples is added to the second training set to obtain a third training set. Use the third training set to perform semi-supervised training on the initial model to generate a third set of pseudo-label samples. Repeat the above process, continuously generating sets of pseudo-label samples and training sets until the number of labeled pseudo-label samples is greater than or equal to the annotation sample threshold, or the face recognition accuracy of the initial model converges, and stop training the initial model to obtain a face recognition model.

[0094] The step of using the first training set to perform semi-supervised training on the initial model to generate a first set of pseudo-label samples includes:

[0095] S411: Input the labeled samples in the first training set into the initial model, so that the initial model learns the mapping relationship between the input features and the output labels.

[0096] S412: Input the unlabeled samples in the first training set into the initial model, make predictions on the unlabeled samples, and obtain the predicted outputs.

[0097] S413: Compare the N predicted outputs with the corresponding actual outputs respectively to obtain the label matching degree.

[0098] S414: Adjust the model parameters of the initial model according to the label matching degree.

[0099] During the semi-supervised learning process, if the pseudo-labels predicted in the initial stage are inaccurate, it may affect the subsequent training and lead to a decline in model performance. In addition, the generation of pseudo-labels often lacks an effective quality control mechanism, which easily causes mislabeled data to be passed to the next round of training, ultimately affecting the recognition accuracy. Therefore, the iterative active learning and semi-supervised training fusion strategy adopted by the present invention can gradually improve the generalization ability and robustness of the model while continuously supplementing pseudo-labels and dynamically adjusting the sample selection strategy.

[0100] The lower the label matching degree, the more times the model parameters are adjusted, and the greater the amplitude of each adjustment of the model parameters. The higher the label matching degree, the fewer times the model parameters are adjusted, and the smaller the amplitude of each adjustment of the model parameters. During the training process of the initial model, if the face recognition accuracy of the initial model has converged, stop training the initial model to obtain the face recognition model.

[0101] This embodiment uses the first training set to perform semi-supervised training on the initial model to generate the first pseudo-label sample set, including inputting the labeled samples in the first training set into the initial model, so that the initial model learns the mapping relationship between the input features and the output labels. Input the unlabeled samples in the first training set into the initial model, make predictions on the unlabeled samples, and obtain the predicted outputs. Compare the N predicted outputs with the corresponding actual outputs respectively to obtain the label matching degree, and adjust the model parameters of the initial model according to the label matching degree. The iterative active learning and semi-supervised training fusion strategy adopted by the present invention can gradually improve the generalization performance and robustness of the model while continuously supplementing pseudo-labels and dynamically adjusting the sample selection strategy. Compared with traditional single methods, the present invention can more effectively handle the problem of insufficient labeled samples, especially in scenarios with a large amount of data or high annotation costs, showing good application potential. In addition, the framework proposed by the present invention can adapt the initial model to different types of data by continuously updating the initial model.

[0102] Embodiment 3

[0103] In an embodiment of the present application, a computer device is further provided. The computer device may be a server. Among them, the computer device includes a processor, a memory, a network interface, and a database connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection.

[0104] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the face recognition method for missing user-end tag data described in any one of Embodiments 1-2. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0105] It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, apparatus, article, or method including that element.

[0106] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be included in the patent protection scope of the present application by the same token.

Claims

1. A face recognition method for missing label data at the user end, characterized in that, Including: Select an initial model and determine a sample selection strategy; Based on the sample selection strategy, screen supplementary samples from the unlabeled sample set to obtain a supplementary sample set, including: If the sample selection strategy is uncertainty sampling, select an uncertainty metric; Screen supplementary samples from the unlabeled sample set according to the uncertainty metric, including: If the uncertainty metric is entropy, sort the unlabeled samples in descending order according to the magnitude of the entropy to obtain an entropy sequence; Use the first K1 unlabeled samples corresponding to the first entropy to the K1-th entropy in the entropy sequence as supplementary samples; set the value of K1 according to the total number of entropies in the entropy sequence, and multiply the total number of entropies in the entropy sequence by 20% to obtain K1; Use all the supplementary samples as the supplementary sample set; Add the supplementary sample set to the labeled sample set to obtain a first training set; the first training set includes labeled samples and unlabeled samples; Use the first training set to perform hybrid training on the initial model to obtain a face recognition model, including: Use the first training set to perform semi-supervised training on the initial model to generate a first set of pseudo-labeled samples, including: Input the labeled samples in the first training set into the initial model so that the initial model learns the mapping relationship between the input features and the output labels; Input the unlabeled samples in the first training set into the initial model, predict the unlabeled samples to obtain a predicted output; Compare N of the predicted outputs with the corresponding actual outputs respectively to obtain a label matching degree; Adjust the model parameters of the initial model according to the label matching degree; Annotate some of the pseudo-labeled samples in the first set of pseudo-labeled samples to obtain a first updated set of pseudo-labeled samples; Add the first updated set of pseudo-labeled samples to the first training set to obtain a second training set; Use the second training set to perform semi-supervised training on the initial model to generate a second set of pseudo-labeled samples; wherein, the hybrid training includes alternating iterative active learning and semi-supervised training; Use the face recognition model to perform face recognition on the user's face; Before screening supplementary samples from the unlabeled sample set based on the sample selection strategy, it also includes constructing an unlabeled sample set, which contains real unlabeled samples and simulated unlabeled samples; the simulated unlabeled samples are obtained by performing a non-linear transformation on the real unlabeled samples, perform a mean operation on P real unlabeled samples to obtain an average face, where P≥2; use the following formula to perform a non-linear transformation on the average face: ； Among them, K i is the i-th simulated unlabeled sample, AVG i is the i-th average face, and MAX is the set maximum pixel value.

2. The face recognition method for missing user - end tag data according to claim 1, wherein, The selection of the initial model includes: Select one model from a neural network model, a decision tree model, a support vector machine model, and a logistic regression model as the initial model.

3. The face recognition method for missing user - end tag data according to claim 1, characterized in that, The determination of the sample selection strategy includes: Select one sampling method from uncertainty sampling, diversity sampling, and information entropy sampling as the sample selection strategy.

4. The face recognition method for missing user - end tag data according to claim 1, characterized in that, The screening of supplementary samples from the unlabeled sample set according to the uncertainty metric includes: If the uncertainty metric is confidence, sort the unlabeled samples in ascending order according to the magnitude of the confidence to obtain a confidence sequence; Use the unlabeled samples corresponding to the first confidence level to the K2-th confidence level in the confidence level sequence as supplementary samples; the confidence levels of the unlabeled samples corresponding to the first K2 confidence levels in the confidence level sequence are low, and use the K2 unlabeled samples with low confidence levels as supplementary samples.

5. The face recognition method for missing user - end label data according to claim 1, wherein Screening supplementary samples from the unlabeled sample set based on the sample selection strategy to obtain a supplementary sample set, including: If the sample selection strategy is diversity sampling, use clustering or maximizing the margin method to screen supplementary samples from the unlabeled sample set; Use all the supplementary samples as the supplementary sample set.

6. The face recognition method for missing user-side tag data according to claim 1, characterized in that After obtaining the first updated pseudo-labeled sample set, it further includes: If the number of labeled pseudo-labeled samples is greater than or equal to the labeled sample threshold, or during the training of the initial model, the face recognition accuracy has converged, stop training the initial model to obtain a face recognition model.

Images