Open set finger vein recognition method and system based on contrastive learning and generative learning

Abstract

The present invention discloses an open set finger vein recognition method and system based on contrastive learning and generative learning. The finger vein training image is input into the contrastive learning model for feature extraction, and the contrastive learning model is optimized based on the obtained contrastive learning features; the training image is input into the generative model to calculate the reconstructed image, and the generative model is optimized based on the reconstruction loss and KL divergence; the alignment loss is calculated based on the contrastive learning features and the generative learning features, and the generative model is further optimized based on the alignment loss and the reconstruction loss; the finger vein test image is input into the optimized generative model, and it is judged whether the reconstruction loss is greater than the preset loss threshold. If so, the test image is judged as an unknown category, otherwise, the test image is judged as a known category; the test image judged as a known category is input into the optimized contrastive learning model for category prediction. The distinction between known category images and unknown category images is achieved through the reconstruction loss threshold, and the erroneous recognition of unknown users is eliminated.

Description

Technical Field

[0001] The present invention belongs to the technical field of biometric feature recognition, and in particular relates to an open-set finger vein recognition method and system based on contrastive learning and generative learning. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Finger vein recognition is a biometric identification technology that uses the veins inside the finger for identity authentication. Since the veins are hidden under the skin and have the characteristics of liveness authentication, finger vein recognition technology is considered to be a safe and reliable means of identity authentication. In traditional identity authentication, passwords may be cracked or forgotten, and ID cards may be lost or stolen. In biometric identification, facial images are easily affected by environmental changes and aging, and fingerprints are easily worn or stolen. In contrast, finger vein recognition technology has higher security and stability.

[0004] Traditional finger vein recognition research is mostly aimed at closed set recognition problems, that is, both the registration image and the test image are from the registered users known to the recognition system. In practical applications, the test image input to the recognition system may not belong to any registered user, which involves the open set finger vein recognition problem. Open set finger vein recognition is that only some categories of images have been seen during model training, and all possible categories need to be identified during testing, including categories seen during training (known categories) and categories not seen during training (unknown categories). For example, for a finger vein recognition access control system installed in the lobby of a company, it is inevitable that users of unknown categories try to use the access control to enter the company. At this time, the test image input to the recognition system is an image of unknown category and does not belong to any registered user. The traditional recognition method matches the test image with the registered images in the database, and the test image is identified as the category to which the registered image with the largest matching score belongs. In this process, there is a risk of identifying an image of unknown category as a registered image in the system, causing the recognition system to be illegally invaded, resulting in serious consequences. In this case, studying open set finger vein recognition technology is of great significance to improving the security and reliability of the recognition system. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides an open-set finger vein recognition method and system based on contrastive learning and generative learning. For open-set finger vein recognition, the distinction between known category images and unknown category images is achieved by reconstructing the loss threshold, thereby eliminating the incorrect identification of unknown users, and using the contrastive learning model to further predict the category of known category images, so as to correctly identify registered category images and improve the security of identity authentication.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, the present invention provides an open set finger vein recognition method based on contrastive learning and generative learning, comprising:

[0008] Acquire a finger vein training image and a finger vein test image;

[0009] Inputting the finger vein training image into a contrastive learning model based on channel attention enhancement for feature extraction to obtain contrastive learning features, calculating contrastive loss based on the contrastive learning features, and optimizing parameters of the contrastive learning model using contrastive loss;

[0010] The finger vein training image is input into a generative model based on a variational autoencoder for feature extraction to obtain generative learning features, and then a reconstructed image is calculated; a reconstruction loss is calculated based on the finger vein training image and the reconstructed image, a KL divergence is calculated based on the output of the generative model encoder, and the parameters of the generative model are optimized using the reconstruction loss and the KL divergence;

[0011] Calculating an alignment loss based on the contrastive learning features and the generative learning features, and further optimizing the parameters of the generative model based on the alignment loss and the reconstruction loss;

[0012] The finger vein image to be tested is input into the optimized generation model to determine whether the reconstruction loss is greater than a preset loss threshold. If so, the finger vein image to be tested is determined as an unknown category; otherwise, the finger vein image to be tested is determined as a known category; the image to be tested determined to be a known category is input into the optimized contrast learning model to pre-extract features and perform category prediction.

[0013] A further technical solution is that the contrastive learning model based on channel attention enhancement includes an improved ResNet50, a global average pooling layer and a projection head in sequence. In the improved ResNet50, a compression-excitation module is added after each residual block to enhance feature representation.

[0014] A further technical solution is that the compression-excitation module initializes the channel weights through global average pooling, then corrects the channel weights through two fully connected layers and an activation function, and weights the features in the channel dimension based on the corrected weights to strengthen important features and suppress unimportant features.

[0015] According to a further technical solution, the generative model based on the variational autoencoder includes an encoder and a decoder, wherein the encoder encodes the input finger vein training image into a latent space representation, and the decoder reconstructs the finger vein training image according to the latent space representation.

[0016] A further technical solution is to align the comparative learning features and the generated learning features through the alignment loss and the reconstruction loss.

[0017] In a further technical solution, the alignment loss function is expressed as:

[0018]

[0019] in, represents the generative model, represents the contrastive learning model, represents the generated learning features, represents the finger vein training image, Representing and generating learned features Comparative learning features of the same category, Representation and Finger vein images of the same category, Representing and generating learned features Comparative learning features of different categories, Representation and Finger vein images of different categories, Representation and The total number of images of different categories, represents the similarity function, Represents the temperature parameter.

[0020] In a further technical solution, the reconstruction loss function is expressed as:

[0021]

[0022] in, represents the finger vein training image, Represents the reconstructed image.

[0023] In a second aspect, the present invention provides an open set finger vein recognition system based on contrastive learning and generative learning, comprising:

[0024] A data acquisition module, which is configured to: acquire a finger vein training image and a finger vein test image;

[0025] A contrastive learning module is configured to: input the finger vein training image into a contrastive learning model based on channel attention enhancement to perform feature extraction to obtain contrastive learning features, calculate contrastive loss based on the contrastive learning features, and optimize the parameters of the contrastive learning model using the contrastive loss;

[0026] A generative learning module is configured to: input the finger vein training image into a generative model based on a variational autoencoder for feature extraction to obtain generative learning features, and then calculate a reconstructed image; calculate a reconstruction loss based on the finger vein training image and the reconstructed image, calculate a KL divergence based on an output of the generative model encoder, and optimize the parameters of the generative model using the reconstruction loss and the KL divergence;

[0027] A feature alignment module, configured to: calculate an alignment loss based on the contrast learning features and the generative learning features, and further optimize the parameters of the generative model based on the alignment loss and the reconstruction loss;

[0028] The category judgment module is configured to: input the finger vein image to be tested into the optimized generation model, determine whether the reconstruction loss is greater than a preset loss threshold, if so, determine the finger vein image to be tested as an unknown category, otherwise, determine the finger vein image to be tested as a known category; input the image to be tested determined to be a known category into the optimized contrast learning model to pre-extract features and perform category prediction.

[0029] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the open set finger vein recognition method based on contrastive learning and generative learning as described in the first aspect.

[0030] In a fourth aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the open set finger vein recognition method based on contrastive learning and generative learning as described in the first aspect are implemented.

[0031] One or more of the above technical solutions have the following beneficial effects:

[0032] The present invention aims at open-set finger vein recognition and utilizes known category images to train a generation model. The model can reconstruct known category images well, but has a poor reconstruction effect on unknown category images. Therefore, the distinction between known category images and unknown category images is achieved through the reconstruction loss threshold, which can accurately identify whether the object to be tested is a registered user or an unknown user, eliminate the erroneous identification of unknown users, and improve the security of identity authentication.

[0033] The present invention constructs positive sample pairs and negative sample pairs through image augmentation, and designs a contrastive loss optimized contrastive learning model, so that the features extracted by the model are more robust to image deformation, etc. At the same time, a compression-excitation block is introduced to enhance the contrastive learning model's ability to perceive key features in the image.

[0034] The present invention makes full use of contrastive learning models and generative models based on feature alignment. The contrastive learning model focuses on extracting discriminative features, while the generative model focuses on extracting generative features. By combining contrastive learning and generative learning, the features of the image are captured from different angles to obtain a more comprehensive and robust feature representation. In addition, feature alignment improves the distinguishability and consistency of the two features, making images of the same category closer in the feature space, while images of different categories are more dispersed. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0036] Figure 1 is a flowchart of an identification method according to an embodiment of the present invention;

[0037] Figure 2 is a structural schematic diagram of a contrastive learning model based on channel attention enhancement in an embodiment of the present invention;

[0038] Figure 3 Schematic diagram of the feature extraction and alignment process in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0041] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0042] Embodiment 1

[0043] like Figure 1 As shown, this embodiment discloses an open set finger vein recognition method based on contrastive learning and generative learning, and the method includes the following steps:

[0044] S1: Acquire a finger vein training image and a finger vein test image;

[0045] In this embodiment, a finger vein training image and a test image are obtained, wherein the category of the training image is known and is used as training data for the model, and the category of the test image is unknown and needs to be predicted.

[0046] S2: Inputting the finger vein training image into a contrastive learning model based on channel attention enhancement for feature extraction to obtain contrastive learning features, calculating contrastive loss based on the contrastive learning features, and optimizing parameters of the contrastive learning model using contrastive loss;

[0047] like Figure 2 As shown in the figure, the contrastive learning model based on channel attention enhancement uses ResNet50 as the encoder and adds a compression-excitation module after each residual block to enhance the feature representation and further improve the performance of the model.

[0048] The acquired training images are used to construct a training set. Random data augmentation operations are performed on each training image in the training set to generate multiple images of different views. Based on the generated images of different views, positive sample pairs from the same category and negative sample pairs from different categories are formed. The positive sample pairs and negative sample pairs are input into the contrastive learning model to learn image features. Data augmentation operations include random cropping, rotation and translation, Gaussian blur, etc.

[0049] The contrastive learning model includes an improved ResNet50, a global average pooling layer and a projection head in sequence. Among them, the network structure of the improved ResNet50 includes the first stage Stage1, the second stage Stage2, the third stage Stage3, the fourth stage Stage4 and the fifth stage Stage5 in sequence.

[0050] The first stage Stage1 includes a 7×7 convolution layer and a 3×3 maximum pooling layer in sequence; the second stage Stage2 includes a residual block Block1, two residual blocks Block2, and each residual block is followed by a compression-excitation module; the third stage Stage3 includes a residual block Block1, three residual blocks Block2, and each residual block is followed by a compression-excitation module; the fourth stage Stage4 includes a residual block Block1, five residual blocks Block2, and each residual block is followed by a compression-excitation module; the fifth stage Stage5 includes a residual block Block1, two residual blocks Block2, and each residual block is followed by a compression-excitation module.

[0051] The residual block includes three convolutional layers and a compression-excitation module, where the first and third convolutional layers are 1×1 convolutions, and the second convolutional layer is a 3×3 convolution. Residual block Block1 adds the input feature map after passing through a 1×1 convolutional layer to the feature map after passing through three convolutional layers and the compression-excitation module. Residual block Block2 adds the input feature map to the feature map after passing through three convolutional layers and the compression-excitation module.

[0052] The compression-excitation module (SE block) initializes the channel weights through global average pooling, then corrects the channel weights through two fully connected layers and activation functions, and weights the features in the channel dimension based on the corrected weights to strengthen important features and suppress unimportant features.

[0053] The feature map output by the improved ResNet50 is input into a global average pooling layer to convert the feature map into a feature vector, and then the feature vector is input into a projection head to map the feature vector to a lower dimensional space. The projection head includes two fully connected layers and a ReLU activation function between the fully connected layers.

[0054] The contrast loss function is calculated based on positive sample pairs and negative sample pairs, and the effective representation of the finger vein training image is learned by comparing the positive and negative sample pairs to optimize the network parameters. The contrast loss function is defined as follows:

[0055]

[0056] in, represents the contrastive learning model, represents the finger vein training image, Representation and Finger vein images of the same category (i.e. positive samples), Representation and Finger vein images of different categories (i.e. negative samples), Representation and A collection of finger vein images with different categories (i.e., a collection of negative samples), Represents a similarity function, such as cosine similarity, Represents the temperature parameter.

[0057] Using the contrast loss function to optimize the network can make the features of similar images closer and the features of heterogeneous images farther apart.

[0058] S3: Input the finger vein training image into the generative model based on the variational autoencoder for feature extraction to obtain generative learning features, and then calculate the reconstructed image; calculate the reconstruction loss based on the finger vein training image and the reconstructed image, calculate the KL divergence based on the output of the generative model encoder, and use the reconstruction loss and KL divergence to optimize the parameters of the generative model;

[0059] In this embodiment, the generative model based on the variational autoencoder includes an encoder and a decoder, the encoder encodes the input finger vein training image into a latent space representation, and the decoder reconstructs the finger vein training image according to the latent space representation. The feature representation learned by the generative model can capture the distribution information of the image so as to distinguish between images of known categories and images of unknown categories.

[0060] Using images of known categories, we train the network through reconstruction loss and KL divergence to learn the distribution of images and improve the generation ability of the model. The definitions of reconstruction loss and KL divergence are as follows:

[0061]

[0062]

[0063] in, represents the reconstruction loss, represents the finger vein training image, represents the reconstructed image, represents the KL divergence, represents the mean of the encoder output of the generative model, represents the variance of the encoder output, represents the dimension of the latent space, Indicates The dimension of the latent space.

[0064] S4: calculating an alignment loss based on the contrast learning feature and the generative learning feature, and further optimizing parameters of the generative model based on the alignment loss and the reconstruction loss;

[0065] In this embodiment, a total loss function combining alignment loss and reconstruction loss is defined to optimize the parameters of the generative model so that the feature representation extracted by the generative model is aligned with the feature representation extracted by the contrastive learning model. In this step, the parameters of the contrastive learning model are fixed, and the parameters of the generative model are updated using the alignment loss and reconstruction loss. This step only uses images of known categories, so that the generative model learns how to effectively reconstruct training images of known categories.

[0066] like Figure 3As shown, first, the contrastive learning model and the generative model are used to extract features from the training images. The feature representation learned by the contrastive learning model based on channel attention enhancement through the contrastive learning strategy can capture the discriminative information of the image, and the feature representation learned by the generative model through the generative learning strategy can capture the distribution information of the image. Then, the generative model is optimized by combining the alignment loss and the reconstruction loss to achieve the alignment of the two features. In the feature alignment stage, the parameters of the contrastive learning model are fixed, and only the images of known categories are used to update the parameters of the generative model. The alignment loss function is expressed as:

[0067]

[0068] in, represents the alignment loss, represents the generative model, represents the contrastive learning model, represents the generated learning features, represents the finger vein training image, Representing and generating learned features Comparative learning features of the same category, Representation and Finger vein images of the same category, Representing and generating learned features Comparative learning features of different categories, Representation and Finger vein images of different categories, Representation and The total number of images of different categories, Represents a similarity function, such as cosine similarity, Represents the temperature parameter, used to adjust similarity sensitivity.

[0069] The total loss function for feature alignment is:

[0070]

[0071] in, represents the alignment loss, represents the reconstruction loss, , represents the weight parameter that balances the alignment loss and the reconstruction loss.

[0072] By minimizing the alignment loss and reconstruction loss, the generative model can accurately reconstruct the input data. It also makes images of the same category close to each other in the feature space and images of different categories more dispersed, which helps to define clear category boundaries in the feature space, thereby improving the ability to distinguish between known categories and unknown categories in open set recognition.

[0073] S5: Input the finger vein test image into the optimized generation model to determine whether the reconstruction loss is greater than the preset loss threshold. If so, the finger vein test image is determined as an unknown category. Otherwise, the finger vein test image is determined as a known category. Input the test image determined as a known category into the optimized contrast learning model to pre-extract features and perform category prediction.

[0074] In this embodiment, the distinction between known category images and unknown category images is achieved based on a predefined reconstruction loss threshold. The generative model is trained based on images of known categories. Therefore, for images of known categories, the model can generate relatively accurate reconstructions with low reconstruction losses; for images of unknown categories, the model lacks the corresponding data distribution knowledge and cannot generate high-quality reconstructions, and the reconstruction losses are high. Therefore, in open set recognition, reconstruction loss is used to determine whether a test image belongs to a known category. Specifically, a loss threshold is preset, the reconstruction loss of each image to be tested is calculated, and compared with the threshold. If the reconstruction loss is greater than the threshold, the image to be tested is judged to be an unknown category; if the reconstruction loss is less than or equal to the threshold, the image is judged to be a known category. For images of known categories, the encoder of the contrastive learning model is used to extract discriminative features, and category prediction is performed based on Euclidean distance. In detail, the Euclidean distance between the features of the training image of the known category and the image to be tested is calculated, and the image to be tested is classified into the category closest to its features.

[0075] The present invention inputs the finger vein training image into the contrastive learning model for feature extraction, and optimizes the contrastive learning model based on the extracted contrastive learning features; inputs the finger vein training image into the generative model for feature extraction, and then calculates the reconstructed image, calculates the reconstruction loss based on the finger vein image and the reconstructed image, calculates the KL divergence based on the output of the generative model encoder, and optimizes the parameters of the generative model using the reconstruction loss and the KL divergence; calculates the alignment loss based on the contrastive learning features and the generative learning features, and further optimizes the parameters of the generative model based on the alignment loss function and the reconstruction loss function; inputs the finger vein test image into the generative model, and determines whether the reconstruction loss is greater than a preset loss threshold, if so, the finger vein image is determined as an unknown category, otherwise, the test image is determined as a known category; the test image determined as a known category is input into the contrastive learning model for feature pre-extraction, and performs category prediction based on the Euclidean distance. For open set finger vein recognition, the distinction between known category images and unknown category images is achieved by using the reconstruction loss threshold, and the erroneous recognition of unknown users is eliminated, and the contrastive learning model is used to further predict the category of the known category image, so as to correctly identify the registered category image and improve the security of identity authentication.

[0076] Embodiment 2

[0077] This embodiment discloses an open set finger vein recognition system based on contrastive learning and generative learning, including:

[0078] A data acquisition module, which is configured to: acquire a finger vein training image and a finger vein test image;

[0079] A contrastive learning module is configured to: input the finger vein training image into a contrastive learning model based on channel attention enhancement to perform feature extraction to obtain contrastive learning features, calculate contrastive loss based on the contrastive learning features, and optimize the parameters of the contrastive learning model using the contrastive loss;

[0080] A generative learning module is configured to: input the finger vein training image into a generative model based on a variational autoencoder for feature extraction to obtain generative learning features, and then calculate a reconstructed image; calculate a reconstruction loss based on the finger vein training image and the reconstructed image, calculate a KL divergence based on an output of the generative model encoder, and optimize the parameters of the generative model using the reconstruction loss and the KL divergence;

[0081] A feature alignment module, configured to: calculate an alignment loss based on the contrast learning features and the generative learning features, and further optimize the parameters of the generative model based on the alignment loss and the reconstruction loss;

[0082] The category judgment module is configured to: input the finger vein image to be tested into the optimized generation model, determine whether the reconstruction loss is greater than a preset loss threshold, if so, determine the finger vein image to be tested as an unknown category, otherwise, determine the finger vein image to be tested as a known category; input the image to be tested determined to be a known category into the optimized contrast learning model to pre-extract features and perform category prediction.

[0083] Embodiment 3

[0084] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of embodiment 1 when executing the program.

[0085] Embodiment 4

[0086] The purpose of this embodiment is to provide a computer-readable storage medium, a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of the method of embodiment 1 are performed.

[0087] The steps involved in the apparatus of the above embodiments 3 and 4 correspond to the method embodiment 1, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0088] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0090] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. An open-set finger vein recognition method based on contrastive learning and generative learning, characterized in that: include: Acquire a finger vein training image and a finger vein test image; Inputting the finger vein training image into a contrastive learning model based on channel attention enhancement for feature extraction to obtain contrastive learning features, calculating contrastive loss based on the contrastive learning features, and optimizing parameters of the contrastive learning model using contrastive loss; Inputting the finger vein training image into a generative model based on a variational autoencoder to extract features, obtain generative learning features, and then calculate a reconstructed image; The reconstruction loss is calculated based on the finger vein training image and the reconstructed image, the KL divergence is calculated based on the output of the generative model encoder, and the reconstruction loss and KL divergence are used to optimize the parameters of the generative model. Calculating an alignment loss based on the contrastive learning features and the generative learning features, and further optimizing the parameters of the generative model based on the alignment loss and the reconstruction loss; Input the finger vein test image to the optimized generation model, determine whether the reconstruction loss is greater than a preset loss threshold, if so, determine the finger vein test image to be an unknown category, otherwise, determine the finger vein test image to be a known category; input the test image determined to be a known category to the optimized contrast learning model to pre-extract features, and perform category prediction; The alignment loss function is expressed as: in, represents the generative model, represents the contrastive learning model, represents the generated learning features, represents the finger vein training image, Representing and generating learned features Comparative learning features of the same category, Representation and Finger vein images of the same category, Representing and generating learned features Comparative learning features of different categories, Representation and Finger vein images of different categories, Representation and The total number of images of different categories, represents the similarity function, Represents the temperature parameter.

2. The open set finger vein recognition method based on contrastive learning and generative learning as claimed in claim 1, characterized in that: The contrastive learning model based on channel attention enhancement includes an improved ResNet50, a global average pooling layer and a projection head in sequence. In the improved ResNet50, a compression-excitation module is added after each residual block to enhance feature representation.

3. The open set finger vein recognition method based on contrastive learning and generative learning as claimed in claim 2, characterized in that: The compression-excitation module initializes channel weights through global average pooling, then corrects the channel weights through two fully connected layers and an activation function, and weights the features in the channel dimension based on the corrected weights to strengthen important features and suppress unimportant features.

4. The open set finger vein recognition method based on contrastive learning and generative learning as claimed in claim 1, characterized in that: The generative model based on variational autoencoder includes an encoder and a decoder, wherein the encoder encodes an input finger vein training image into a latent space representation, and the decoder reconstructs the finger vein training image according to the latent space representation.

5. The open set finger vein recognition method based on contrastive learning and generative learning as claimed in claim 1, characterized in that: The contrast learning features and the generated learning features are aligned through the alignment loss and the reconstruction loss.

6. The open set finger vein recognition method based on contrastive learning and generative learning as claimed in claim 5, characterized in that: The reconstruction loss function is expressed as: in, represents the finger vein training image, Represents the reconstructed image.

7. An open-set finger vein recognition system based on contrastive learning and generative learning, characterized in that: include: A data acquisition module, which is configured to: acquire a finger vein training image and a finger vein test image; A contrastive learning module is configured to: input the finger vein training image into a contrastive learning model based on channel attention enhancement to perform feature extraction to obtain contrastive learning features, calculate contrastive loss based on the contrastive learning features, and optimize the parameters of the contrastive learning model using the contrastive loss; A generative learning module is configured to: input the finger vein training image into a generative model based on a variational autoencoder for feature extraction to obtain generative learning features, and then calculate a reconstructed image; calculate a reconstruction loss based on the finger vein training image and the reconstructed image, calculate a KL divergence based on an output of the generative model encoder, and optimize the parameters of the generative model using the reconstruction loss and the KL divergence; A feature alignment module, configured to: calculate an alignment loss based on the contrast learning features and the generative learning features, and further optimize the parameters of the generative model based on the alignment loss and the reconstruction loss; The category judgment module is configured to: input the finger vein test image to the optimized generation model, judge whether the reconstruction loss is greater than a preset loss threshold, if so, judge the finger vein test image to be an unknown category, otherwise, judge the finger vein test image to be a known category; input the test image determined to be a known category to the optimized contrast learning model to pre-extract features, and perform category prediction; The alignment loss function is expressed as: in, represents the generative model, represents the contrastive learning model, represents the generated learning features, represents the finger vein training image, Representing and generating learned features Comparative learning features of the same category, Representation and Finger vein images of the same category, Representing and generating learned features Comparative learning features of different categories, Representation and Finger vein images of different categories, Representation and The total number of images of different categories, represents the similarity function, Represents the temperature parameter.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the open set finger vein recognition method based on contrastive learning and generative learning as described in any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the open set finger vein recognition method based on contrastive learning and generative learning are implemented as described in any one of claims 1 to 6.

Images