Partial finger vein recognition method and system based on feature interactive fusion network

Abstract

The present invention relates to the field of biometric feature recognition technology, specifically to a partial finger vein recognition method and system based on a feature interaction fusion network, which obtains vein images of partial fingers and extracts features of different scales; generates information interaction matrices for features of different scales using features of at least two other scales, and performs feature interaction with the information interaction matrix based on matrix multiplication operations to obtain interactive features of corresponding scales, wherein the interactive features emphasize distinguishing channel information based on a channel attention mechanism; performs feature weighted fusion based on a spatial weight matrix and interactive features of different scales to obtain fused features; and obtains partial finger vein recognition results based on the fused features based on cross entropy loss and center loss. Under the premise that the image acquisition window size is limited, the features of different scales are fully mined and fused to improve the performance of partial finger vein recognition.

Description

Technical Field

[0001] The present invention relates to the field of biometric feature recognition technology, and in particular to a method and system for partial finger vein recognition based on a feature interactive fusion network. Background Art

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

[0003] Identity authentication technologies (e.g., common passwords and ID cards) are widely used in daily life. However, simple passwords are at risk of being stolen, complex passwords are easily forgotten by users, and ID cards are easily lost or stolen because they are physical entities. Identity authentication using human physiological or behavioral characteristics can address the above problems, such as biometric recognition technologies such as fingerprint recognition and face recognition.

[0004] Among various biometric features, the veins inside the finger are not easy to be stolen, nor are they easily affected by skin damage, and only living fingers can be used for identity authentication. Currently, finger vein recognition is mostly aimed at near-full finger vein images, that is, the image contains the veins of the entire finger or close to the veins of the entire finger. In general, the more veins collected in the image, the more distinguishing information in the image, and the more conducive to identification.

[0005] However, when the finger vein acquisition module or finger vein recognition module is embedded in small electronic devices, such as smart door locks and attendance machines, the size of the small electronic devices limits the size of the finger vein image acquisition window. Some acquisition devices can only acquire partial finger vein images, that is, the image only contains part of the veins of a finger. Traditional finger vein recognition methods do not consider partial finger vein images, and most recognition methods cannot extract rich distinguishing information from partial finger vein images, resulting in unsatisfactory performance of partial finger vein recognition. Summary of the invention

[0006] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a partial finger vein recognition method and system based on a feature interaction fusion network. Under the premise that the image acquisition window size is limited, partial finger vein images are utilized, multi-scale feature interaction based on the information interaction matrix and feature fusion based on spatial position information are used to fully mine and fuse features of different scales, thereby improving the performance of partial finger vein recognition.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A first aspect of the present invention provides a method for partial finger vein recognition based on a feature interactive fusion network, comprising the following steps:

[0009] Obtain partial finger vein images and extract features at different scales;

[0010] For features of different scales, information interaction matrices are generated using features of at least two other scales, and features are interacted with the information interaction matrix based on matrix multiplication operations to obtain interactive features of corresponding scales. The interactive features emphasize discriminative channel information based on the channel attention mechanism.

[0011] Based on the spatial weight matrix and the features after interaction at different scales, feature weighted fusion is performed to obtain fused features;

[0012] The fused features are based on cross entropy loss and center loss to obtain partial finger vein recognition results.

[0013] Furthermore, features of different scales are extracted, specifically: network branches with different expansion rates are used to extract features of different scales.

[0014] Furthermore, for features of different scales, information interaction matrices are generated using features of at least two other scales, specifically:

[0015] Reshape the feature F1∈C*H*W to generate a new feature F1′∈C*N (N=H*W), where C is the number of channels, H is the height, and W is the width.

[0016] Reshape and transpose the feature F2∈C*H*W to generate a new feature F2′∈N*C (N=H*W);

[0017] Reshape the feature F3∈C*H*W to generate a new feature F3′∈C*N (N=H*W);

[0018] Perform matrix multiplication on features F2′ and F3′ to obtain the information interaction matrix IIM∈N*N.

[0019] Furthermore, based on the matrix multiplication operation, the reshaped feature F1′ is interacted with the information interaction matrix to obtain the interactive features of the corresponding scale, specifically:

[0020] Reshape the product matrix of feature F1′ and information interaction matrix into the original dimension F1″ (C*H*W);

[0021] The features F1 and F1″ are summed to obtain the interactive feature FI1 corresponding to the feature F1.

[0022] The above interaction operations are also performed on F2 and F3 respectively to obtain the corresponding interactive features FI2 and FI3.

[0023] Furthermore, based on the spatial weight matrix and the features after interaction at different scales, feature weighted fusion is performed to obtain the fused features, specifically:

[0024] The interacted features are summed, and the obtained results are average pooled and max pooled along the X-axis and Y-axis directions respectively to obtain new features FSX and FSY. After dimension exchange, FSY is concatenated with FSX along the channel direction to generate feature FS′;

[0025] Perform point-by-point convolution on the generated feature FS′ to obtain FS″, and split it into FSX′ and FSY′. Use point-by-point convolution on the two split features to restore the number of feature channels. The two features after channel number restoration use Sigmoid activation function to obtain the weight W1 in the X-axis direction and the weight W2 in the Y-axis direction. Multiply the two weights to get the spatial position weight W.

[0026] The spatial position weights are multiplied with the interacted features respectively, and all the product matrices are summed to obtain the fused feature FF.

[0027] Furthermore, the fused features are based on cross entropy loss and center loss to obtain partial finger vein recognition results, specifically:

[0028] The cross entropy loss and center loss are weighted and fused to obtain the total loss function as shown below:

[0029] L = a1*L1+(1-a1)*L2;

[0030] Among them, L1 and L2 represent cross entropy loss and center loss respectively, and a1 is the weighted weight.

[0031] A second aspect of the present invention provides a partial finger vein recognition system based on a feature interaction fusion network, comprising:

[0032] The feature extraction module is configured to: obtain a portion of the finger vein image, and extract features of different scales from the portion of the finger vein image;

[0033] The feature interaction module is configured to: generate an information interaction matrix using features of at least two other scales for features of different scales, and perform feature interaction with the information interaction matrix based on matrix multiplication operation to obtain interactive features of corresponding scales, wherein the interactive features emphasize discriminative channel information based on a channel attention mechanism;

[0034] The feature fusion module is configured to: perform weighted fusion of features based on the spatial weight matrix and the interacted features of different scales to obtain fused features;

[0035] The identity recognition module is configured as follows: the fused features are based on cross entropy loss and center loss to obtain partial finger vein recognition results.

[0036] A third aspect of 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 above-mentioned partial finger vein recognition method.

[0037] A fourth aspect of 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 the processor implements some of the steps in the above-mentioned finger vein recognition method when executing the program.

[0038] Compared with the prior art, one or more of the above technical solutions have the following beneficial effects:

[0039] 1. Existing finger vein recognition methods are mainly aimed at recognition tasks based on nearly full finger vein images. This scheme fully considers the problem of partial finger vein image recognition caused by the limited acquisition window size of the imaging device, and studies the extraction and mining of discriminative information of partial finger vein images to improve the performance of partial finger vein recognition.

[0040] 2. In the traditional finger vein recognition process, the features extracted from different branches are fused in series, ignoring the interaction problem between features from different branches. The present invention uses the information interaction matrix obtained from features of different scales to perform feature interaction, and cooperates with the channel attention mechanism and feature fusion based on spatial position information to fully mine and fuse features of different scales.

[0041] 3. Regarding the selection of loss function, most finger vein recognition methods only use cross entropy loss to train the network, focusing only on the differences between classes and ignoring the consistency within the class. The present invention combines cross entropy loss and center loss to comprehensively consider the differences between classes and the consistency within the class, thereby improving the recognition performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] 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.

[0043] Figure 1 is a schematic diagram of a portion of a finger vein recognition process provided by one or more embodiments of the present invention;

[0044] Figure 2 is a schematic diagram of a three-branch convolutional network with different dilation rates provided by one or more embodiments of the present invention;

[0045] Figure 3is a schematic diagram of a multi-scale feature interaction module provided by one or more embodiments of the present invention;

[0046] Figure 4 is a schematic diagram of calculating spatial position weights provided by one or more embodiments of the present invention;

[0047] Figure 5 is a schematic diagram of feature fusion at different scales provided by one or more embodiments of the present invention;

[0048] Figure 6 It is a schematic diagram of a feature interaction fusion network provided by one or more embodiments of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed descriptions are exemplary and 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.

[0051] It should be noted that the terms herein are only intended to describe 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 intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0052] As introduced in the background technology, the traditional finger vein recognition algorithm does not take partial finger vein images into consideration, and most recognition methods cannot extract rich distinguishing information from partial finger vein images, resulting in unsatisfactory performance of partial finger vein recognition.

[0053] Therefore, the following embodiments provide a partial finger vein recognition method and system, which implements partial finger vein recognition based on a feature interaction fusion network, which includes a feature extraction module, a feature interaction module, and a feature fusion module. Based on a three-branch convolutional network with different scales, three different scale features are extracted from the partial finger vein image, and the feature interaction module is used to realize information interaction between features of different scales, and the feature fusion module is used to fuse the different scale features after interaction together. In addition, the cross entropy loss and center loss are used together to optimize the feature interaction fusion network to further improve the performance of partial finger vein recognition.

[0054] Embodiment 1:

[0055] like Figure 1As shown, a partial finger vein recognition method includes the following steps:

[0056] For some finger vein images, three different scales of features are extracted based on a three-branch convolutional network with different scales to capture the local information and contextual information of the finger vein images at different scales. Among them, the three network branches all contain a series of convolutional layers and pooling layers, but use different expansion rates, which are 1, 2, and 3 respectively. Due to the different expansion rates, the receptive field size of the convolution kernel is different, which allows the network to focus on features of different scales.

[0057] For features of three different scales, a multi-scale feature interaction module based on information interaction matrix and channel attention mechanism is proposed to interact between features of a certain scale and features of the other two scales. For features of a certain scale, the information interaction matrix is ​​generated using features of the other two scales, and feature interaction is performed with the information interaction matrix based on matrix multiplication operation. In addition, the channel attention mechanism is used to emphasize the discriminative channel information for the interacted features.

[0058] For the three different scale features after interaction, a feature fusion module based on spatial position information is designed to effectively fuse the features of different scales. The spatial weight matrix is ​​calculated based on the sum matrix of the three different scale features, and the spatial weight matrix is ​​applied to the three different scale features, and the three different scale features are weighted and fused to obtain the fused features.

[0059] The network model is learned by combining cross entropy loss and center loss. Cross entropy loss is used to measure the difference between the model output and the true label, helping the network to make better probability predictions. Center loss can help reduce the difference between different images of the same category, so that different images of the same category are more closely clustered together in the feature space. This helps improve the model to reduce intra-class differences, thereby improving classification accuracy.

[0060] Existing finger vein recognition methods are mainly aimed at recognition tasks based on nearly full finger vein images. This scheme fully considers the problem of partial finger vein image recognition caused by the limited acquisition window size of the imaging device, and studies the extraction and mining of discriminative information of partial finger vein images to improve the performance of partial finger vein recognition.

[0061] The existing finger vein recognition method based on multiple network branches fuses the features extracted by different branches in series, ignoring the interaction between features from different branches, which is not conducive to the mining of discriminative information and the improvement of recognition performance. To address this problem, this scheme uses a multi-scale feature interaction module based on information interaction matrix and a feature fusion module based on spatial position information to fully mine and fuse the discriminative information of features at different scales.

[0062] In addition, regarding the selection of loss functions, most finger vein recognition methods only use cross entropy loss to train the network, focusing only on the differences between classes and ignoring the consistency within the class. However, this solution combines the loss functions of cross entropy loss and center loss, which can comprehensively consider the differences between classes and the consistency within classes, and improve the recognition performance of the network model.

[0063] For the above scheme, a specific introduction is given:

[0064] Step S01: feature extraction module;

[0065] In this step, a feature extraction module is designed to extract rich distinguishing features from part of the finger vein image. The designed feature extraction module has three network branches with the same structure but different expansion rates (the structure of this module is as follows Figure 2 As shown in Figure 2). Through these three branches, features of three different scales can be extracted. All three network branches have three convolutional layers and three pooling layers, where each convolutional layer has batchnorm and relu activation functions. The expansion rates of the three network branches are different. The expansion rate of all convolutional layers in the first branch is 1, and image features are extracted in a small-scale receptive field; the expansion rate of all convolutional layers in the second branch is 2, and image features are extracted in a medium-scale receptive field; the expansion rate of all convolutional layers in the third branch is 3, and image features are extracted in a large-scale receptive field.

[0066] like Figure 2 As shown in the figure, for a given partial finger vein image F, three features of different scales F1, F2 and F3 are extracted through a three-branch convolutional network. The calculation formula is:

[0067] F1=branch1(F); F2=branch2(F); F3=branch3(F) (1)

[0068] Among them, branch1 is network branch 1 with an expansion rate of 1, branch2 is network branch 2 with an expansion rate of 2, and branch3 is network branch 3 with an expansion rate of 3.

[0069] Step S02: feature interaction module;

[0070] In this step, a multi-scale feature interaction module based on information interaction matrix and channel attention mechanism is designed to realize the interaction between features of different scales and improve the distinguishability of some finger vein image features. The feature interaction module is introduced by taking the interaction between F1 and other two scale features as an example. Figure 3 shown.

[0071] Given feature F1, the interaction matrix is ​​calculated based on the other two scale features.

[0072] Reshape the feature F1∈C*H*W to generate a new feature F1′∈C*N (N=H*W);

[0073] Reshape the feature F2∈C*H*W and transpose it to generate a new feature F2′∈N*C (N=H*W);

[0074] Reshape the feature F3∈C*H*W to generate a new feature F3′∈C*N (N=H*W);

[0075] Perform matrix multiplication on features F2′ and F3′ to obtain the information interaction matrix IIM∈N*N;

[0076] Reshape the product matrix of feature F1′ and information interaction matrix into the original dimension F1″ (C*H*W);

[0077] The interaction between features is completed by summing features F1 and F1″.

[0078] In order to make the features after information interaction more discriminative, a channel attention mechanism is introduced to the features after interaction to emphasize the discriminative channel information. The above operations are also performed on F2 and F3 respectively.

[0079] The interaction formula between features of three different scales is:

[0080] FI1=CA(F1+R(R(F1)*(R(F2)′*R(F3)))) (2)

[0081] FI2=CA(F2+R(R(F2)*(R(F1)′*R(F3)))) (3)

[0082] FI3=CA(F3+R(R(F3)*(R(F2)′*R(F1)))) (4)

[0083] Among them, R represents the reshape operation and CA represents channel attention.

[0084] Step S03: feature fusion module;

[0085] In order to effectively fuse the three interacted features FI1, FI2 and FI3, a feature fusion module based on spatial position information is proposed, such as Figure 4 and Figure 5 shown.

[0086] The new feature FS∈C*H*W is generated by summing FI1, FI2 and FI3. Average pooling and maximum pooling are performed on FS along the X-axis and Y-axis directions respectively to obtain new features FSX∈C*H*1 and FSY∈C*1*W. After exchanging the dimension of FSY, a concatenation operation is performed with FSX along the channel direction to generate feature FS′, as shown in formulas (5) and (6):

[0087] FS=FI1+FI2+FI3 (5)

[0088] FS′=[XAP(FS);YMP(FS)′] (6)

[0089] Among them, XAP represents the average pooling operation in the X-axis direction, YMP represents the maximum pooling operation in the Y-axis direction, and ' represents the exchange dimension operation.

[0090] Use point-by-point convolution on FS′ to make features at different positions interact, and reduce the number of channels to half of the original, to obtain FS″. Split FS″ into FSX′ and FSY′, and use point-by-point convolution on the two features to restore the number of channels of the two features. Use the Sigmoid activation function on the two features after the number of channels is restored to obtain the weight W1 in the X-axis direction and the weight W2 in the Y-axis direction. Multiply the two weights to get the spatial position weight W.

[0091] Multiply the spatial position weights by the interactive features FI1, FI2, and FI3 respectively, and sum the three product matrices to obtain the fusion feature FF. The specific calculation formula of FF is:

[0092] FS″=Conv(FS′) (7)

[0093] W1=sigmoid(Conv(SP1(FS″))) (8)

[0094] W2=sigmoid(Conv(SP2(FS″))) (9)

[0095] W=W1*W2 (10)

[0096] FF=FI1*W+FI2*W+FI3*W (11)

[0097] Among them, Conv represents the point-by-point convolution operation, SP1 and SP2 represent taking out the partial features corresponding to FSX and the partial features corresponding to FSY from FS″ respectively, and W represents the spatial position weight.

[0098] Step S04: joint loss function;

[0099] In order to make the network have good recognition performance, the network model is trained by combining cross entropy loss and center loss. The output of the last layer of the network is used to calculate the cross entropy loss and center loss. The total loss function is obtained by weighted fusion of the cross entropy loss and the center loss, as shown below:

[0100] L=a1*L1+(1-a1)*L2 (12)

[0101] Among them, L1 and L2 represent cross entropy loss and center loss respectively, a1 is the weighted weight, which is initially set to 0.5.

[0102] During training, branches with different expansion rates are subjected to feature fusion after passing through an interaction module, and are classified after passing through a fully connected layer (e.g., two fully connected layers).

[0103] Embodiment 2:

[0104] Some finger vein recognition systems include:

[0105] The feature extraction module is configured to: obtain a partial finger vein image and extract features of different scales;

[0106] The feature interaction module is configured to: generate an information interaction matrix for features of all scales respectively using features of at least two other scales, and perform feature interaction with the information interaction matrix based on matrix multiplication operation to obtain interactive features of corresponding scales, wherein the interactive features emphasize discriminative channel information based on a channel attention mechanism;

[0107] The feature fusion module is configured to: perform weighted fusion of features based on the spatial weight matrix and the interacted features of different scales to obtain fused features;

[0108] The identity recognition module is configured as follows: the fused features are based on cross entropy loss and center loss to obtain partial finger vein recognition results.

[0109] Embodiment three:

[0110] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the method for partial finger vein recognition as described in the first embodiment are implemented.

[0111] Embodiment 4:

[0112] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the partial finger vein recognition method described in the first embodiment are implemented.

[0113] The steps involved in the above embodiments 2 to 4 correspond to those in embodiment 1. For the specific implementation, please refer to the relevant description 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.

[0114] 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.

Claims

1. A partial finger vein recognition method based on a feature interactive fusion network, characterized in that: The steps include: Obtain partial finger vein images and extract features at different scales; For features of different scales, information interaction matrices are generated using features of at least two other scales, and features are interacted with the information interaction matrix based on matrix multiplication operations to obtain interactive features of corresponding scales. The interactive features emphasize discriminative channel information based on the channel attention mechanism. Based on the spatial weight matrix and the interactive features of different scales, the feature weighted fusion is performed to obtain the fused features, specifically: The interacted features are summed, and the results are average pooled and max pooled along the X-axis and Y-axis directions respectively to obtain new features FSX and FSY. After dimension exchange, FSY is concatenated with FSX along the channel direction to generate features. Perform point-by-point convolution on the generated features to obtain and split them into and. Use point-by-point convolution on the two split features to restore the number of feature channels. The two features after the number of channels is restored use the activation function to obtain the weight W1 in the X-axis direction and the weight W2 in the Y-axis direction. The two weights are multiplied to obtain the spatial position weight W. The spatial position weights are multiplied by the interacted features respectively, and all the product matrices are summed to obtain the fused features; The fused features are based on cross entropy loss and center loss to obtain partial finger vein recognition results.

2. The method for partial finger vein recognition based on feature interactive fusion network as claimed in claim 1, characterized in that: Extract features of different scales, specifically: extract features of different scales based on network branches with different expansion rates.

3. The method for partial finger vein recognition based on feature interactive fusion network as claimed in claim 1, characterized in that: For features of different scales, information interaction matrices are generated using features of at least two other scales, specifically: Reshape the feature F1∈C*H*W to generate a new feature F1'∈C*N (N=H*W), where C is the number of channels, H is the height, and W is the width; Reshape and transpose the feature F2∈C*H*W to generate a new feature F2'∈N*C (N=H*W); Reshape the feature F3∈C*H*W to generate a new feature F3'∈C*N (N=H*W); Perform matrix multiplication on features F2' and F3' to obtain the information interaction matrix IIM∈N*N.

4. The method for partial finger vein recognition based on feature interactive fusion network as claimed in claim 1, characterized in that: Based on the matrix multiplication operation and the information interaction matrix, the feature interaction is performed to obtain the interactive features of the corresponding scale, which are as follows: Reshape the product matrix of feature F1' and information interaction matrix into the original dimension F1" (C*H*W); Sum the features F1 and F1' to obtain the interactive feature FI1 corresponding to the feature F1.

5. The method for partial finger vein recognition based on feature interactive fusion network as claimed in claim 1, characterized in that: The fused features are based on cross entropy loss and center loss to obtain partial finger vein recognition results. Specifically, the cross entropy loss and center loss are weightedly fused to obtain the total loss function.

6. The method for partial finger vein recognition based on feature interactive fusion network as claimed in claim 5, characterized in that: The cross entropy loss and center loss are weighted and fused to obtain the total loss function, as shown in the following formula: L = a1*L1+(1-a1)*L2; Among them, L1 and L2 represent cross entropy loss and center loss respectively, and a1 is the weighted weight.

7. A partial finger vein recognition system based on a feature interactive fusion network, characterized in that: include: The feature extraction module is configured to: obtain a partial finger vein image and extract features of different scales; The feature interaction module is configured to: generate an information interaction matrix for features of all scales respectively using features of at least two other scales, and perform feature interaction with the information interaction matrix based on matrix multiplication operation to obtain interactive features of corresponding scales, wherein the interactive features emphasize discriminative channel information based on a channel attention mechanism; The feature fusion module is configured to perform weighted fusion of features based on the spatial weight matrix and the features after interaction at different scales to obtain fused features, specifically: The interacted features are summed, and the results are average pooled and max pooled along the X-axis and Y-axis directions respectively to obtain new features FSX and FSY. After dimension exchange, FSY is concatenated with FSX along the channel direction to generate features. Perform point-by-point convolution on the generated features to obtain and split them into and. Use point-by-point convolution on the two split features to restore the number of feature channels. The two features after the number of channels is restored use the activation function to obtain the weight W1 in the X-axis direction and the weight W2 in the Y-axis direction. The two weights are multiplied to obtain the spatial position weight W. The spatial position weights are multiplied by the interacted features respectively, and all the product matrices are summed to obtain the fused features; The identity recognition module is configured as follows: the fused features are based on cross entropy loss and center loss to obtain partial finger vein recognition results.

8. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps in the partial finger vein recognition method based on a feature interactive fusion network as described in any one of claims 1 to 6 are implemented.

9. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for partial finger vein recognition based on a feature interactive fusion network as described in any one of claims 1 to 6 are implemented.

Images