Method and device for generating palm vein sample data

Abstract

This application provides a method and apparatus for generating palm vein sample data, which are used to improve the accuracy of obtaining palm vein sample data. The method includes: obtaining a feature matrix of a real palm vein image; wherein the feature matrix is ​​used to represent the distribution of image features of the real palm vein image; inputting the feature matrix into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image; wherein the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a method and device for generating palm vein sample data. Background Art

[0002] With the continuous development of deep learning technology, its application in various fields is becoming increasingly widespread. Deep learning technology requires a large amount of sample data. Ordinary natural images can be captured and calibrated using web crawlers. Palm vein images are used in the field of biometrics. Therefore, deep learning technology requires high accuracy of palm vein sample images. Palm vein sample images contain internal information of specific parts of the human body and are difficult to obtain online.

[0003] Currently, palm vein sample images are primarily obtained through neural network training. This typically involves performing convolution and pooling on real palm vein images to generate the corresponding sample data. However, this method can omit certain image information from real palm vein images during training, resulting in inaccurate sample data. Summary of the Invention

[0004] The present application provides a method and apparatus for generating palm vein sample data, which are used to improve the accuracy of acquiring palm vein sample images.

[0005] In a first aspect, a method for generating palm vein sample data is provided, comprising:

[0006] Obtaining a feature matrix of a real palm vein image; wherein the feature matrix is ​​used to represent the image feature distribution of the real palm vein image;

[0007] The feature matrix is ​​input into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image; wherein the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix.

[0008] In an embodiment of the present application, the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix of the real palm vein image, which avoids missing the image information of the real palm vein image during the pooling process, and is conducive to improving the accuracy of the palm vein sample data generated in the later training.

[0009] In one possible design, the feature matrix is ​​input into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image, including:

[0010] Inputting the feature matrix into the convolution layer of the generator to obtain a convolution matrix;

[0011] Inputting the convolution matrix into the pooling layer of the generator to obtain a pooling matrix;

[0012] The pooling matrix is ​​input into the fully connected layer of the generator to obtain palm vein sample data.

[0013] In an embodiment of the present application, a network structure of a generator is improved, which includes a convolutional layer, a pooling layer and a fully connected layer in sequence. The network structure of the generator is simple and easy to implement.

[0014] In one possible design, the convolution matrix is ​​input into the pooling layer of the generator to obtain a pooling matrix, including:

[0015] Performing maximum pooling aggregation processing on the convolution matrix to obtain a pooling aggregation matrix;

[0016] Inverting the pooling aggregation matrix according to a preset pooling operator to obtain a pooling aggregation inverse matrix;

[0017] Multiply the preset tensor by the inverse pooling matrix to obtain a pooling matrix.

[0018] In an embodiment of the present application, after the pooling aggregation processing, the pooling aggregation matrix is ​​inverted, and the preset tensor is multiplied by the pooling inverse matrix to obtain a pooling matrix with the same dimension as the feature matrix. No additional calculation amount is added during the pooling aggregation processing. The multiplication operation is performed after the pooling aggregation inverse matrix is ​​obtained, which relatively reduces the calculation amount.

[0019] In one possible design, a maximum pooling aggregation process is performed on the convolution matrix to obtain a pooling aggregation matrix, including:

[0020] Splitting the convolution matrix into at least one pooling region according to a preset window and a preset step size; wherein the value of the preset step size is the same as the size of the preset window;

[0021] Aggregation processing is performed on elements in each pooling area of ​​the at least one pooling area to obtain a pooling aggregation matrix.

[0022] In an embodiment of the present application, during the pooling aggregation processing, the convolution matrix is ​​split according to a preset window into at least one pooling matrix to reduce the computational complexity of the pooling aggregation processing.

[0023] In a possible design, obtaining a feature matrix of a real palm vein image includes:

[0024] Get real palm vein images;

[0025] performing binarization processing on the real palm vein image;

[0026] Acquire a region of interest (ROI) in the real palm vein image after binarization processing, and adjust the size of the ROI to a preset size;

[0027] The grayscale value corresponding to each pixel in the adjusted ROI is obtained, and the grayscale values ​​of all pixels in the ROI are arranged in a preset order, so as to obtain a feature matrix of the real palm vein image.

[0028] In an embodiment of the present application, the ROI of the real palm vein image is identified, and the size of the feature matrix of the real palm vein image is minimized without missing important features, so as to improve the efficiency of generating palm vein sample data in the later stage.

[0029] In one possible design, the adversarial network includes a discriminator, which, before inputting the feature matrix into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image, includes:

[0030] Inputting a noise matrix corresponding to a preset real palm vein image into the generator to obtain a training palm vein image corresponding to the noise matrix;

[0031] Inputting the training palm vein image into the discriminator to obtain image convolution features of the training palm vein image;

[0032] Inputting the image convolution feature into a logistic function to obtain a probability value that the training palm vein image is real;

[0033] According to the probability value, the parameters of the generator and the parameters of the discriminator are updated using the gradient descent method until the cross-entropy loss function converges, thereby obtaining the generator in the pre-trained adversarial network; wherein the convergence of the cross-entropy loss function is used to represent the loss between the generator and the discriminator.

[0034] In an embodiment of the present application, a method for training a generator and a discriminator is provided. The training method is simple and direct and can quickly train an adversarial network model.

[0035] In one possible design, based on the probability value, the parameters of the generator and the parameters of the discriminator are updated using a gradient descent method, including:

[0036] Obtaining a first parameter gradient value of the generator and a second parameter gradient value of the discriminator according to the probability value;

[0037] According to the first parameter gradient value, the current parameter value of the generator is updated, and according to the second parameter gradient value, the current parameter value of the discriminator is updated.

[0038] In an embodiment of the present application, the parameters of the generator and the discriminator are updated using the gradient descent method to increase the speed of training the generator and the discriminator.

[0039] In one possible design, the frequency of updating the parameters of the generator is greater than or equal to the frequency of updating the parameters of the discriminator.

[0040] In the embodiment of the present application, the frequency of updating the parameters of the generator and the discriminator is flexibly set to improve the flexibility of training the generator and the discriminator.

[0041] In a second aspect, a device for generating palm vein sample data is provided, comprising:

[0042] An acquisition module, configured to acquire a feature matrix of a real palm vein image; wherein the feature matrix is ​​used to represent the distribution of image features of the real palm vein image;

[0043] A processing module is used to input the feature matrix into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image; wherein the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix.

[0044] In a third aspect, a device for generating palm vein sample data is provided, comprising:

[0045] at least one processor, and

[0046] a memory communicatively coupled to the at least one processor;

[0047] The memory stores instructions that can be executed by the at least one processor, and the at least one processor implements the method as described in the first aspect by executing the instructions stored in the memory.

[0048] According to a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions. When the computer instructions are executed on a computer, the computer executes the method as described in any one of the first aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of a scenario of a method for generating palm vein sample data provided in an embodiment of the present application;

[0050] Figure 2A flowchart of a method for generating palm vein sample data provided in an embodiment of the present application;

[0051] Figure 3 A schematic diagram of the process of processing a real palm vein sample image provided in an embodiment of the present application;

[0052] Figure 4 A structural diagram of a generator provided in an embodiment of the present application;

[0053] Figure 5 A schematic diagram of the pooling process provided in an embodiment of the present application;

[0054] Figure 6 A schematic diagram of palm vein sample data generated according to an embodiment of the present application;

[0055] Figure 7 A schematic diagram of the structure of a device for generating palm vein sample data provided in an embodiment of the present application;

[0056] Figure 8 A schematic diagram of the structure of a device for generating palm vein sample data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to better understand the technical solutions provided by the embodiments of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0058] In order to better understand the technical solutions provided by the embodiments of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0059] In order to improve the accuracy of obtaining palm vein sample data, the present application embodiment provides a method for generating palm vein sample data. Figure 1 The application scenarios of this method are described.

[0060] The present embodiment uses an adversarial network to generate palm vein sample data. The adversarial network can run on a server, which can be a single server or a server cluster. The adversarial network includes a generator and a discriminator. The generator generates training palm vein sample images based on noise samples, and the discriminator determines whether the training palm vein sample images are genuine. During the continuous training process, the generator and discriminator are optimized. After training is completed, the generator can generate corresponding palm vein sample data.

[0061] based on Figure 1 The following describes the method for generating palm vein sample data involved in this application. Figure 2 , the specific process of this method is as follows:

[0062] Step 201: Obtain a feature matrix of a real palm vein image; wherein the feature matrix is ​​used to represent the distribution of image features of the real palm vein image;

[0063] In step 202, the feature matrix is ​​input into the generator in the pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image; wherein the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix.

[0064] The method is executed by a device for generating palm vein sample data, and the device for generating palm vein sample data can be implemented by a computer.

[0065] The overall idea of ​​the embodiment of this application is described below:

[0066] Before generating the palm vein sample data, the device for generating the palm vein sample data needs to first obtain the feature matrix of the real palm vein image. The feature matrix is ​​used to represent the image feature distribution of the real palm vein image. The device for generating the palm vein sample data obtains the palm vein sample data corresponding to the real palm vein image based on the feature matrix of the real palm vein image and a pre-trained generator.

[0067] The specific execution method of step 201 is described below with examples.

[0068] One way to perform step 201 is:

[0069] Obtain the feature matrix of the real palm vein image from other devices.

[0070] Specifically, other devices can communicate with the device that generates palm vein sample data. After processing the real palm vein image, the other devices send the feature matrix of the real palm vein image to the device that generates palm vein sample data, thereby the device that generates palm vein sample data obtains the feature matrix.

[0071] One way to perform step 201 is:

[0072] Get real palm vein images;

[0073] Perform binarization processing on the real palm vein image;

[0074] Obtaining a region of interest (ROI) in the real palm vein image after binarization processing, and adjusting the size of the ROI to a preset size;

[0075] The eigenvalue corresponding to each window in the adjusted ROI is obtained, and the eigenvalues ​​of all windows in the ROI are arranged in a preset order to obtain the feature matrix of the real palm vein image.

[0076] Specifically, the device for generating palm vein sample data obtains a real palm vein image, binarizes the real palm vein image, obtains a ROI in the real palm vein image, adjusts the size of the real palm vein image to a uniform size, and thereby obtains the eigenvalue of each pixel in the ROI as a feature matrix of the real palm vein image.

[0077] The feature value is used to represent the image feature information of the ROI, such as grayscale value or brightness value.

[0078] For example, see Figure 3 , Figure 3 From a to e, it represents the process of processing the real palm vein image. Figure 3 a in the image represents the real palm vein image after binarization. Calibrate a to obtain Figure 3 The ROI shown in b is shown in FIG. 3 , c is a schematic diagram after the size of the ROI in b is adjusted to a preset size, d is a diagram after the ROI in c is filtered, and e is a schematic diagram after the filtering process of d. The device for generating palm vein sample data can obtain the grayscale value of each pixel in the image from FIG. e, and arrange the grayscale values ​​in sequence to obtain the feature matrix corresponding to the real palm vein image.

[0079] After executing step 201, step 202 is executed, that is, the feature matrix is ​​input into the generator in the pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image; wherein the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix.

[0080] Before executing step 202, the device generating the palm vein sample data needs to first obtain the generator and discriminator in the adversarial network. The following describes the method for training the generator and discriminator of the adversarial network. Based on the generator and discriminator, a cross entropy loss function is constructed. The cross entropy loss function is specifically expressed as follows:

[0081]

[0082] Where D represents the discriminator, G represents the generator, x represents the real palm vein image, z represents the noise sample input to the generator, G(z) represents the training palm vein image generated by the generator based on the noise sample, D(x) represents the probability that the discriminator judges that the real palm vein image is real, and D(G(z)) represents the probability that the discriminator judges that the training palm vein image generated by the generator is real. It refers to the expected probability value of the real palm vein image being judged as real. It represents the expected probability that the training palm vein image generated based on the noise sample is judged to be real.

[0083] The noise samples are input into the generator, which outputs the corresponding training palm vein images.

[0084] For example, see Figure 4 ,In the process of training noise samples, the generator includes convolutional layer, pooling layer and fully connected layer in sequence.,After the noise samples pass through the convolutional layer, pooling layer and fully connected layer in sequence, the training palm vein image is output.

[0085] The discriminator performs convolution processing on the training palm vein image to obtain the convolution feature corresponding to the training palm vein image. The convolution feature is input into the logistic function to output the probability value of the training palm vein image being real. The specific expression of the logistic function is as follows:

[0086]

[0087] According to the probability value, the parameters of the generator and the discriminator are updated using the gradient descent method. The specific update method is as follows:

[0088] According to the probability value, the gradient of the parameters of the generator and the gradient of the parameters of the discriminator are obtained. The calculation formula of the gradient of the parameters of the generator is as follows:

[0089]

[0090] The calculation formula of the gradient of the discriminator parameters is as follows:

[0091]

[0092] Among them, D(x i ) represents the probability that the discriminator determines that the real palm vein image is a real image, D(G(Z (i) )) represents the probability value, and m represents the total number of samples used in this calculation.

[0093] The parameter values ​​of the generator are updated according to the gradient of the parameters of the generator, and the parameter values ​​of the discriminator are updated according to the gradient of the parameters of the discriminator until the cross entropy loss function converges, thus completing the training process of the generator and discriminator.

[0094] In one possible implementation, when the generator updates its parameter values, the gradient of the generator's parameters is subtracted. When the discriminator updates its parameter values, the gradient of the discriminator's parameters is added.

[0095] In one possible implementation, the frequency of parameter updates of the generator is greater than or equal to the frequency of parameter updates of the discriminator.

[0096] Specifically, each time the parameters of the generator are updated, the parameters of the discriminator may be updated, or the parameters of the discriminator may be updated after the parameters of the generator have been updated a preset number of times.

[0097] The following describes how the apparatus for generating palm vein sample data executes step 202 .

[0098] Input the feature matrix into the convolution layer of the generator to obtain the convolution matrix;

[0099] Input the convolution matrix into the pooling layer of the generator to obtain the pooling matrix;

[0100] The pooling matrix is ​​input into the fully connected layer of the generator to obtain palm vein sample data.

[0101] Specifically, after the generator is trained, the various parameters in the generator have been determined, and after the feature matrix passes through the convolution layer, pooling layer and fully connected layer in sequence, the generator can generate the corresponding palm vein sample data.

[0102] In a possible embodiment, the convolution matrix is ​​input into the pooling layer of the generator to obtain a pooling matrix, including:

[0103] Perform maximum pooling aggregation processing on the convolution matrix to obtain a pooling aggregation matrix;

[0104] Invert the pooling aggregation matrix according to the preset pooling operator to obtain the pooling aggregation inverse matrix;

[0105] Multiply the preset tensor by the inverse pooling matrix to obtain the pooling matrix.

[0106] In a possible embodiment, the device for generating palm vein sample data performs maximum pooling aggregation processing on the convolution matrix to obtain the pooling aggregation matrix as follows:

[0107] Splitting the convolution matrix into at least one pooling region according to a preset window and a preset step size; wherein the value of the preset step size is the same as the size of the preset window;

[0108] Aggregate the elements in each pooling area of ​​the at least one pooling area to obtain a pooling aggregation matrix.

[0109] Specifically, the apparatus for generating palm vein sample data splits the convolution matrix into multiple pooling regions according to a preset window and a preset step size. To reduce the dimensionality of the convolution matrix, the apparatus for generating palm vein sample data may set the size of the preset window to be the same as the preset step size. After obtaining the multiple pooling regions, the apparatus for generating palm vein sample data takes the value of each pooling region as the maximum value in that region, thereby obtaining a pooling aggregation matrix.

[0110] For example, see Figure 5 , the preset window is 2*2, the preset step size is 2, and the device generating palm vein sample data divides the convolution matrix into 4 pooling areas, such as Figure 5 As shown in a in , the value in the first pooling area is taken as the maximum value 120, so as to obtain the value of each pooling area in turn, and finally obtain the pooling aggregation matrix, as shown in Figure 5 As shown in b in , we can see that the dimension of the convolution matrix is ​​4*4, and the dimension of the pooling aggregation matrix is ​​2*2.

[0111] In a possible embodiment, after obtaining multiple pooling regions, the apparatus for generating palm vein sample data sets the non-maximum value in each pooling region to 0 and retains the maximum value in each pooling region, thereby obtaining a pooling aggregation matrix.

[0112] In this embodiment, the non-maximum values ​​in each pooling region are set to 0, suppressing all gradient values ​​except the maximum value, which is beneficial for refining the edges of the central pixel. Furthermore, retaining the non-maximum values ​​can reduce the computational complexity of the pooling aggregation process while maintaining the dimensionality of the pooling aggregation matrix.

[0113] After obtaining the pooling aggregation matrix, the device generating the palm vein sample data performs dimensionality increase processing on the pooling aggregation matrix in the following manner:

[0114] The pooling matrix is ​​inverted using the preset pooling operator to obtain the inverse pooling matrix. The preset tensor is then multiplied by the inverse pooling matrix to obtain the pooling matrix. Finally, the dimensions of the pooling matrix are made equal to those of the feature matrix, preserving more detailed information from the real palm vein sample image and ensuring that the resulting palm vein sample data is more realistic.

[0115] The device for generating palm vein sample data inputs the pooling matrix into the fully connected layer of the generator to obtain the palm vein sample data.

[0116] For example, Figure 3 The feature matrix corresponding to the image shown in e is input into Figure 4 In the generator trained in , we can obtain Figure 6 Multiple palm vein sample data shown in .

[0117] Based on the method for generating palm vein sample data discussed above, the present embodiment provides a device for generating palm vein sample data, please refer to Figure 7 The device for generating palm vein sample data includes an acquisition module 701 and a processing module 702, wherein:

[0118] An acquisition module 701 is configured to acquire a feature matrix of a real palm vein image; wherein the feature matrix is ​​used to represent the distribution of image features of the real palm vein image;

[0119] Processing module 702 is used to input the feature matrix into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image; wherein the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix.

[0120] Based on the method for generating palm vein sample data discussed above, the present embodiment provides a device for generating palm vein sample data, please refer to Figure 8 The device for generating palm vein sample data includes a processor 801 and a memory 802, wherein:

[0121] The processor 801 is communicatively connected to the memory 802;

[0122] The memory 802 stores instructions that can be executed by the processor 801. The processor 801 implements the following by executing the instructions stored in the memory 802: Figure 2 The method described.

[0123] Based on the method for generating palm vein sample data discussed above, the present application embodiment provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions. When the computer instructions are executed on a computer, the computer executes the following Figure 2 The method described.

[0124] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0125] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0128] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0129] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A method for generating palm vein sample data, characterized in that: include: Obtaining a feature matrix of a real palm vein image; wherein the feature matrix is ​​used to represent the image feature distribution of the real palm vein image; The feature matrix is ​​input into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image; wherein the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix, wherein: Inputting the feature matrix into the generator in the pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image includes: inputting the feature matrix into the convolution layer of the generator to obtain a convolution matrix; inputting the convolution matrix into the pooling layer of the generator to obtain a pooling matrix; inputting the pooling matrix into the fully connected layer of the generator to obtain palm vein sample data, wherein inputting the convolution matrix into the pooling layer of the generator to obtain a pooling matrix includes: performing maximum pooling aggregation processing on the convolution matrix to obtain to a pooling aggregation matrix; inverting the pooling aggregation matrix according to a preset pooling operator to obtain a pooling aggregation inverse matrix; multiplying a preset tensor by the pooling inverse matrix to obtain a pooling matrix, wherein the convolution matrix is ​​subjected to maximum pooling aggregation processing to obtain a pooling aggregation matrix, including: splitting the convolution matrix into at least one pooling area according to a preset window and a preset step size, and aggregating the elements in each pooling area in the at least one pooling area to obtain a pooling aggregation matrix, wherein the value of the preset step size is the same as the size of the preset window.

2. The method according to claim 1, wherein Get the feature matrix of the real palm vein image, including: Get real palm vein images; performing binarization processing on the real palm vein image; Acquire a region of interest (ROI) in the real palm vein image after binarization processing, and adjust the size of the ROI to a preset size; The grayscale value corresponding to each pixel in the adjusted ROI is obtained, and the grayscale values ​​of all pixels in the ROI are arranged in a preset order, so as to obtain a feature matrix of the real palm vein image.

3. The method according to claim 1, wherein The adversarial network includes a discriminator, which, before inputting the feature matrix into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image, includes: Inputting a noise matrix corresponding to a preset real palm vein image into the generator to obtain a training palm vein image corresponding to the noise matrix; Inputting the training palm vein image into the discriminator to obtain image convolution features of the training palm vein image; Inputting the image convolution feature into a logistic function to obtain a probability value that the training palm vein image is real; According to the probability value, the parameters of the generator and the parameters of the discriminator are updated using the gradient descent method until the cross-entropy loss function converges, thereby obtaining the generator in the pre-trained adversarial network; wherein the convergence of the cross-entropy loss function is used to represent the loss between the generator and the discriminator.

4. The method according to claim 3, wherein According to the probability value, the parameters of the generator and the parameters of the discriminator are updated using a gradient descent method, including: Obtaining a first parameter gradient value of the generator and a second parameter gradient value of the discriminator according to the probability value; According to the first parameter gradient value, the current parameter value of the generator is updated, and according to the second parameter gradient value, the current parameter value of the discriminator is updated.

5. The method according to claim 4, wherein The frequency of updating the parameters of the generator is greater than or equal to the frequency of updating the parameters of the discriminator.

6. A device for generating palm vein sample data, characterized in that: include: An acquisition module, configured to acquire a feature matrix of a real palm vein image; wherein the feature matrix is ​​used to represent the distribution of image features of the real palm vein image; A processing module is configured to input the feature matrix into a generator in a pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image; wherein the dimension of the pooling matrix output by the pooling layer in the generator is the same as the dimension of the feature matrix, wherein: Inputting the feature matrix into the generator in the pre-trained adversarial network to obtain palm vein sample data corresponding to the real palm vein image includes: inputting the feature matrix into the convolution layer of the generator to obtain a convolution matrix; inputting the convolution matrix into the pooling layer of the generator to obtain a pooling matrix; inputting the pooling matrix into the fully connected layer of the generator to obtain palm vein sample data, wherein inputting the convolution matrix into the pooling layer of the generator to obtain a pooling matrix includes: performing maximum pooling aggregation processing on the convolution matrix to obtain to a pooling aggregation matrix; inverting the pooling aggregation matrix according to a preset pooling operator to obtain a pooling aggregation inverse matrix; multiplying a preset tensor by the pooling inverse matrix to obtain a pooling matrix, wherein the convolution matrix is ​​subjected to maximum pooling aggregation processing to obtain a pooling aggregation matrix, including: splitting the convolution matrix into at least one pooling area according to a preset window and a preset step size, and aggregating the elements in each pooling area in the at least one pooling area to obtain a pooling aggregation matrix, wherein the value of the preset step size is the same as the size of the preset window.

7. A device for generating palm vein sample data, characterized in that: include: at least one processor, and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the at least one processor implements the method according to any one of claims 1 to 5 by executing the instructions stored in the memory.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and when the computer instructions are executed on a computer, the computer is caused to perform the method according to any one of claims 1 to 5.

Images