Face recognition classification method based on sliding mode variable structure classification decision

Abstract

The invention provides a face recognition classification method based on a sliding mode variable structure classification decision, and belongs to the technical field of sliding mode control strategies. The method comprises the following steps: 1, extracting a feature vector of a human face, and defining an SVM classification function of which an identification target is a target person or a non-target person; 2, a weighted Gaussian function is introduced into the SVM classification function obtained in the step 1 to serve as a kernel function, and a corresponding SVM classification function is obtained; 3, in order to optimize kernel function parameters of the SVM in the step 2 by using SMC, introducing a sigmoid function to replace a sign function in an SVM classification function so as to solve and express a classification problem; secondly, establishing a relation between the SMC and the SVM through an error; and 4, obtaining a classification objective function based on the SVM classification function in the step 2 and the error in the step 3. According to the method, the classification effect of the feature vectors of the face is improved, the convergence and stability of the method are proved, and the classification accuracy of face recognition is kept and even improved while the number of the support vectors is greatly reduced.

Description

Technical Field

[0001] This invention relates to a face recognition classification method based on sliding mode variable structure classification decision, belonging to the field of sliding mode control strategy technology. Background Technology

[0002] Data classification is a core task in machine learning, especially the Support Vector Machine (SVM) algorithm, which excels in handling nonlinear datasets. Traditional SVM achieves nonlinear classification through kernel function mapping, but kernel parameters are typically optimized using open-loop methods, lacking dynamic feedback adjustment based on classification errors. Sliding mode control (SMC), as a closed-loop control strategy, possesses strong robustness in handling nonlinear and uncertain systems and is increasingly being incorporated into SVM parameter optimization, improving model adaptability and convergence speed. In the field of face recognition, because the features between different faces cannot be distinguished by simple linear classifiers, the feature vectors extracted from faces in face recognition belong to linearly inseparable datasets.

[0003] Defects and shortcomings of existing technology:

[0004] 1. Traditional SVM kernel parameter optimization methods are open-loop optimizations, which cannot dynamically adjust the model based on errors for linearly inseparable face recognition datasets, thus limiting the model's ability to match complex data distributions.

[0005] 2. Existing research combining SMC and SVM does not explicitly provide a classification objective function, resulting in insufficient model transparency and theoretical completeness.

[0006] 3. Existing optimization algorithms still have room for improvement in terms of increasing the number of support vectors. Summary of the Invention

[0007] The purpose of this invention is to solve the problems existing in the prior art and to provide a face recognition classification method based on sliding mode variable structure classification decision.

[0008] The objective of this invention is achieved through the following technical solution:

[0009] A classification method for linearly inseparable datasets based on a sliding mode control strategy includes the following steps:

[0010] Step 1: Extract the feature vector of the face and define the SVM classification function to identify whether the target is a person or not.

[0011] Step 2: Introduce a weighted Gaussian function as the kernel function into the SVM classification function obtained in Step 1 to obtain the corresponding SVM classification function;

[0012] Step 3: Optimize the kernel function parameters of the SVM in Step 2 using SMC. We introduce the sigmoid function to replace the sign function in the SVM classification function to solve and represent the classification problem; then, through error... Establish the connection between SMC and SVM;

[0013] Step 4: Based on the SVM classification function in Step 2 and the error in Step 3 Thus, the classification objective function is obtained.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] 1. In this invention, the closed-loop SMC control enhances the adaptive adjustment capability of the SVM kernel parameters, thereby improving the classification effect of facial feature vectors.

[0016] 2. This invention clearly defines a classification objective function for identifying whether the target is a target person or a non-target person, and theoretically proves the convergence and stability of the method.

[0017] 3. This invention maintains or even improves the classification accuracy of face recognition while significantly reducing the number of support vectors. Attached Figure Description

[0018] Figure 1 This is a flowchart of the face recognition classification method based on sliding mode variable structure classification decision in this invention.

[0019] Figure 2 This is a flowchart of the algorithm in Embodiment 1 of the present invention. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to the accompanying drawings: This embodiment is implemented under the premise of the technical solution of the present invention, and detailed implementation methods are given, but the protection scope of the present invention is not limited to the following embodiments.

[0021] like Figure 1 As shown in the embodiment, the face recognition classification method based on sliding mode variable structure classification decision involves the following: when the recognition target is "target person (positive class)" vs "non-target person (negative class)", for example, when the phone is unlocked, the feature vector of the face (such as facial feature distance and texture) is extracted, and SVM is used to determine whether the current face belongs to the owner of the phone.

[0022] In this invention, a classification algorithm for linearly inseparable datasets will be designed using kernel functions. First, the SVM classification function is given, defined as follows:

[0023] (1)

[0024] in, It is a hyperplane that classifies data points of different categories. The total number of feature vector samples for a face. It's a kernel function. The input is the feature vector of the face, representing the data points to be classified. The training samples are used to train the SVM model. yes The output, yes The output, It is a weight variable. It is a displacement term, designed to describe the distance between the hyperplane and the origin.

[0025] In the SVM classification results, when Correct classification was achieved when There are two possible misclassification scenarios:

[0026] (2)

[0027] in, These are the feature vectors of a human face and the data points to be classified. The actual label value.

[0028] This invention will use a weighted Gaussian function as the kernel function, that is ,in, Represents the weight parameter matrix, It is a positive semi-definite matrix. , These are parameters that need to be optimized.

[0029] Introducing the corresponding kernel function, the corresponding SVM classification function is:

[0030] (3)

[0031] To use SMC to optimize the kernel function parameters of SVM Furthermore, the sigmoid function is introduced to replace the sign function, thereby solving and representing classification problems (the necessity of introducing the sigmoid function and its expression are explained in Note 1). Then, through error... Establish the connection between SMC and SVM, and the error The specific form is: ,in:

[0032] (4)

[0033] Based on the classification function (3) and error The classification objective function is further obtained as follows:

[0034] (5)

[0035] in, It is a strictly positive number. It is the designed sliding surface. It is any non-decreasing odd function; and These are the dynamic equations for the sliding surface and the non-sliding surface, respectively. The dynamic equation for the sliding surface is... and the dynamic equations of non-sliding surfaces This ensures both the incremental stability of the algorithm and improves the accuracy of the classification algorithm.

[0036] Note 1: (1) When using SMC to optimize the kernel function parameters of SVM, the problem of finding the derivative needs to be solved during the algorithm iteration process. Since the derivative of the sign function is irreversible, the sign function is replaced with the sigmoid function to solve and describe the classification problem.

[0037] (2) Specifically, using and Two expressions (where) To handle the error separately, when Two types of misclassified data points at time, let You can get The expression.

[0038] Note 2: Sliding surface and the dynamic equations of non-sliding surfaces This ensures both the asymptotic stability of the algorithm and improves the accuracy of the classification algorithm.

[0039] Through theoretical derivation, it is proven that dynamic updating of kernel parameters can be achieved. The value is such that it satisfies the objective function (5).

[0040] Example 1

[0041] like Figure 2 As shown, an SVM algorithm based on the SMC strategy improvement is designed:

[0042] The kernel parameter algorithm based on the SMC strategy is as follows: Let the regularization parameter be... Three counters initial value and the algorithm's iterative updates ,

[0043] Input: Feature vector of the face to be classified and initial values ​​of each parameter

[0044] Output: Dataset classification results, parameters The number of support vectors and classification accuracy.

[0045] 1. Initialize parameter values ​​and set them. , initial value and ;

[0046] 2. Train an SVM based on the SMC policy to generate support vectors (SVs) and their number (NSVs);

[0047] 3. Based on the training set, test set, and validation set used for classification, calculate the training error TE and validation error VE. If TE = 0, then the perturbation... ;

[0048] 4. Use the label (MC-lbs) parameter from the misclassified training data to update the kernel parameters. ;

[0049] 5. Algorithm Check ,Check Has the maximum acceptable number of iterations been reached to improve the MNVE threshold? If so... If the maximum acceptable number of iterations has not been reached, the new value calculated is... Repeat this process until MNVE reaches its predefined threshold, completing the training.

[0050] 6. Use Classify the dataset.

[0051] The above description is merely a preferred embodiment of the present invention. These specific embodiments are different implementations based on the overall concept of the present invention, and the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A face recognition classification method based on sliding mode variable structure classification decision, characterized in that, Includes the following steps: Step 1: Extract the feature vector of the face and define the SVM classification function to identify whether the target is a person or not. Step 2: Introduce a weighted Gaussian function as the kernel function into the SVM classification function obtained in Step 1 to obtain the corresponding SVM classification function; Step 3: Optimize the kernel function parameters of the SVM in Step 2 using SMC. We introduce the sigmoid function to replace the sign function in the SVM classification function to solve and represent the classification problem; then, through error... Establish the connection between SMC and SVM; Step 4: Based on the SVM classification function in Step 2 and the error in Step 3 Thus, the classification objective function is obtained.

2. The face recognition classification method based on sliding mode variable structure classification decision according to claim 1, characterized in that, The SVM classification function described in step one is as follows: (1) Where sign is the sign function. It is a hyperplane that classifies data points of different categories. The total number of feature vector samples for a face. It's a kernel function. The input is the feature vector of the face, representing the data points to be classified. The training samples are used to train the SVM model. yes The output, yes The output, It is a weight variable. It is a displacement term.

3. The face recognition classification method based on sliding mode variable structure classification decision as described in claim 2, characterized in that, In the classification result of the SVM classification function described in step one, when Correct classification was achieved when There are two possible misclassification scenarios: (2) in, These are the feature vectors of a human face and the data points to be classified. The actual label value.

4. The face recognition classification method based on sliding mode variable structure classification decision as described in claim 2, characterized in that, The weighted Gaussian function mentioned in step two is ,in, Represents the weight parameter matrix, It is a positive semi-definite matrix. , These are parameters that need to be optimized.

5. The face recognition classification method based on sliding mode variable structure classification decision as described in claim 4, characterized in that, The corresponding SVM classification function obtained in step two is as follows: (3)。 6. The face recognition classification method based on sliding mode variable structure classification decision as described in claim 5, characterized in that, The error mentioned in step three The specific form is: ,in: (4)。 7. The face recognition classification method based on sliding mode variable structure classification decision as described in claim 6, characterized in that, The classification objective function described in step four is: (5) in, It is a strictly positive number. It is the designed sliding surface. It is any non-decreasing odd function; and These are the dynamic equations for the sliding surface and the non-sliding surface, respectively.

Images