Multi-feature finger fusion biological recognition method and system

Abstract

The invention discloses a multi-feature finger fusion biological recognition method and system, and relates to the technical field of biological recognition, and the method comprises the steps: 1, collecting a multi-finger vein image, including the front or back of a plurality of fingers; 2, performing finger segmentation on the multi-finger vein image to obtain position and boundary information of each finger, and independently separating a vein image area of each finger; 3, finger vein feature extraction is carried out, the vein feature of each finger is obtained, feature fusion is carried out, and multi-finger fusion features are obtained; and step 4, carrying out feature matching, identifying the identity, and driving the equipment to be started to complete the whole biological identification. The system comprises a finger groove, a vein image acquisition unit, a core processing unit and a lock driving unit.

Description

Technical Field

[0001] The present invention relates to the field of biometric identification technology, and in particular to a multi-feature finger fusion biometric identification method and system. Background Art

[0002] In the digital age, traditional mechanical keys and password locks can no longer meet people's needs, and people have increasingly higher requirements for the security and convenience of identity authentication.

[0003] Single finger vein recognition and finger back vein recognition technology, hereinafter referred to as "finger vein", requires users to place their fingers in a specific posture so that the device can accurately obtain images of the finger veins. Users do not need to carry keys or remember complex passwords. They only need to place their fingers in the collection area of ​​the door lock to complete the unlocking operation. This is very convenient for the elderly, children, and users with both hands full of items. The fast recognition speed also greatly improves the efficiency of users' entry and exit. In addition, the finger vein features are not affected by the surface conditions of the fingers. For example, fingerprints may be difficult to recognize due to factors such as moisture and wear of the fingers, while the finger vein features can still maintain a stable recognition effect under these conditions.

[0004] Although the uniqueness and stability of finger vein characteristics make this biometric technology difficult to forge, there are still people who try to imitate the vein characteristics of a single finger, so that the existing finger vein technology still has a certain risk of being forged and imitated.

[0005] Therefore, technicians in this field are committed to developing a new biometric identification method and system to solve the above-mentioned defects in the prior art. Summary of the invention

[0006] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is how to maintain a stable recognition effect while further reducing the risk of finger vein technology being forged and imitated.

[0007] To achieve the above object, the present invention provides a biometric identification method of multi-feature finger fusion, comprising the following steps: Step 1, collecting multiple finger vein images, including the front or back of multiple fingers; Step 2: segmenting the multiple finger vein images to obtain the position and boundary information of each finger, and separating the vein image area of ​​each finger; Step 3: Extract finger vein features, obtain the vein features of each finger, and then perform feature fusion to obtain multi-finger fusion features; Step 4: Perform feature matching, identify the identity, and drive the device to turn on to complete the entire biometric identification.

[0008] Furthermore, the step 1 includes the following sub-steps: Step 1.1, use the finger groove to guide the identification object to place the finger at the pre-set collection position; Step 1.2, surround the collection area with several infrared lamps to form a near-infrared light source to illuminate the finger; Step 1.3: Use a CMOS sensor and a camera with FOV to collect the multiple finger vein images.

[0009] Furthermore, the shooting scheme adopted in step 1.3 is a single-camera scheme or a multi-camera scheme, wherein the multi-camera scheme uses two or more cameras to capture images of the front or back of multiple fingers from different angles.

[0010] Furthermore, the multi-camera solution is provided with an image stitching method, comprising the following sub-steps: Step 1.3.1, extract feature points from different images respectively; Step 1.3.2, find the feature points corresponding to the same physical location on different images and perform feature matching; Step 1.3.3, calculate the relative position relationship between different images based on the matched feature points, and align the different images in space through coordinate transformation; Step 1.3.4: Fusing the overlapping areas of different images to obtain the fused multi-finger vein image.

[0011] Furthermore, in step 2, a pre-trained finger segmentation model is used to perform finger segmentation; the finger segmentation model is a neural network model built using the PyTorch deep learning framework, and the labeled training data is input into the finger segmentation model for training to learn the characteristics and segmentation patterns of the fingers; the preparation of the training data includes: collecting multiple finger vein image data from different individuals, annotating the vein areas of the thumb, index finger, middle finger, ring finger and little finger, and annotating 4 to 6 key position points for each finger, wherein the key position points include but are not limited to the joints, fingertips, or bases of the fingers.

[0012] Furthermore, in step 3, a pre-trained feature extraction model is used to extract the vein features of each finger; the feature extraction model is a deep learning model, including multiple convolutional layers, pooling layers, transformer layers and fully connected layers. The deep learning model is trained using a training set, and the loss function between the predicted results and the true annotations is calculated. The parameters of the feature extraction model are updated according to the gradient back propagation of the loss function, and the training is repeated until the feature extraction model converges or reaches a preset number of iterations.

[0013] Furthermore, the vein feature of each finger extracted in step 3 corresponds to a feature vector, and the feature vectors of each finger are spliced ​​in a head-to-tail splicing or weighted splicing manner; Assume that Fingers, The feature vector of each finger is expressed as , then the feature vector after concatenation is expressed as: The feature vector after weighted concatenation is expressed as: in, is the weight coefficient.

[0014] Furthermore, in step 4, the matching result of a single finger is first obtained, and then the matching results are fused by weighted summation; Among them, the matching of a single finger adopts feature matching based on distance measurement or feature matching based on deep learning model; The feature matching based on distance measurement calculates the distance between the concatenated feature vector obtained in step 3 and the stored known feature vector. If the calculated distance is less than a preset threshold, it is considered that the identity is successfully identified, wherein the distance measurement method includes Euclidean distance or Manhattan distance; The feature matching based on the deep learning model will first use a large amount of multi-finger vein feature vector data to train the model, and then input the spliced ​​feature vector obtained in step 3 into the trained model to output the matching result.

[0015] The present invention also provides a biometric identification system with multi-feature finger fusion, including a finger groove, a vein image acquisition unit, a core processing unit, and a lock driving unit; The finger groove guides the identification subject to place the finger at a preset collection position; The vein image acquisition unit is connected to the finger groove, illuminates the finger through a near-infrared light source, and uses a camera to acquire the vein images of multiple fingers; The core processing unit is connected to the vein image acquisition unit, uses a pre-trained finger segmentation model to perform finger segmentation and a pre-trained feature extraction model to extract vein features of each finger, then performs feature matching, identifies the user identity, and sends a start signal; The lock driving unit is connected to the core processing unit, receives the opening signal, drives the corresponding lock or device to open, and completes the entire biometric identification process; The system also includes a biometric identification method of multi-feature finger fusion as described in any of the above items.

[0016] Furthermore, the system also includes a touch awakening device, and when the identification object contacts the touch awakening device, the system is awakened from the low power consumption state.

[0017] The biometric identification method and system for multi-feature finger fusion provided by the present invention have at least the following technical effects: 1. Compared with the traditional biometric recognition system which often only recognizes a single finger or a single part, the technical solution provided by the present invention uses multiple finger veins or multiple back of hand veins for recognition, which expands the collection scope of biometric features and provides richer recognition information; 2. The technical solution provided by the present invention greatly increases the accuracy of recognition by integrating the vein features of multiple fingers and multiple backs of hands. The vein features of different fingers are independent and complementary. Even if the features of a certain part are disturbed or damaged to a certain extent, the features of other parts can still provide strong support for accurate recognition. At the same time, multi-finger and multi-back vein recognition is based on the physiological characteristics of the human body and is difficult to forge. Compared with traditional biometric recognition methods (such as fingerprint recognition, passwords, etc.), it has higher security and can effectively prevent illegal intrusion and identity counterfeiting.

[0018] 3. The technical solution provided by the present invention is more stable and reliable in the face of various complex situations. For example, under different environmental conditions (such as changes in light and temperature), the recognition of multiple fingers / multiple backs of hands can reduce the impact that single part recognition may be affected, and improve the adaptability and stability of the system.

[0019] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a single camera solution of a preferred embodiment of the present invention; Figure 2 is a multi-camera solution flow chart of a preferred embodiment of the present invention; Figure 3 This is a schematic diagram of collecting multiple finger vein images according to a preferred embodiment of the present invention; Figure 4 It is a biological system structure diagram of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following describes several preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0022] Example 1 like Figure 1 , a biometric identification method for multi-feature finger fusion provided by an embodiment of the present invention, comprising the following steps: Step 1, collecting multiple finger vein images, including the front or back of multiple fingers; Step 2: Segment the multi-finger vein images to obtain the position and boundary information of each finger, and separate the vein image area of ​​each finger; Step 3: Extract finger vein features, obtain the vein features of each finger, and then perform feature fusion to obtain multi-finger fusion features; Step 4: Perform feature matching, identify the identity, and drive the device to turn on to complete the entire biometric identification.

[0023] Example 2 Based on Example 1, step 1 includes the following sub-steps: Figure 3 As shown: Step 1.1, use the finger groove to guide the identification object to place the finger at the pre-set collection position; Step 1.2, surround the collection area with several infrared lamps to form a near-infrared light source to illuminate the finger; Step 1.3, collect multiple finger vein images using a CMOS sensor and a camera with FOV.

[0024] In particular, the finger groove in step 1.1 can be made of plastic, silicone and other materials, which have a certain degree of elasticity and softness to ensure that the user feels comfortable when placing the finger. Finger grooves of different sizes can be designed according to the size and shape of different fingers. For example, the size of the thumb groove can be slightly larger, while the size of the little finger groove can be slightly smaller. The working principle and function of the finger groove are as follows: the shape and size of the finger groove match the shape of the finger, which can guide the user to place the finger in the correct position. When the user puts the finger into the groove, the groove will limit the movement range of the finger to ensure that the finger maintains a stable position and angle during the acquisition process. The finger groove can also protect the finger and prevent the finger from being hit and damaged by the outside world during the acquisition process. At the same time, the design of the finger groove can make the contact between the finger and the acquisition device closer, improving the quality and stability of image acquisition.

[0025] In particular, the shooting scheme adopted in step 1.3 is a single-camera scheme or a multi-camera scheme, wherein the multi-camera scheme uses two or more cameras to capture images of the front or back sides of multiple fingers from different angles.

[0026] In particular, the collected multi-finger vein images are preprocessed, specifically including: 1) Noise removal: Due to the influence of the acquisition environment and the device itself, noise may exist in the original image. Median filtering, Gaussian filtering and other algorithms are used to remove noise and improve the image quality. For example, for each pixel, median filtering will use the median of the surrounding pixel values as the new value of this pixel, thus effectively removing salt-and-pepper noise, etc.

[0027] 2) Image correction: The collected finger vein images may have problems such as position offset and angle tilt, and the images need to be corrected. The position and angle of the image are adjusted through algorithms to meet the requirements of subsequent processing and comparison, and to ensure the accuracy of vein features.

[0028] 3) Contrast enhancement: The contrast of the image is enhanced by methods such as histogram equalization, making the vein patterns clearer and distinguishable. Histogram equalization is a method of stretching or compressing the gray value distribution of the image, so as to improve the overall contrast of the image.

[0029] 4) Image normalization: The processed images are normalized so that the images under different acquisition conditions have a unified size and gray scale range. For example, the images are uniformly adjusted to a size of 256x256 pixels, and the gray value range is mapped to between [0, 255].

[0030] In particular, as Figure 2 shown, the multi-camera scheme is provided with an image stitching method, including the following sub-steps: Step 1.3.1: Extract feature points from different images respectively; for example, for two images with overlapping parts, labeled data can be used for feature point matching, or feature point detection algorithms (such as SIFT, SURF or ORB, etc.) can be used to extract feature points on the two images respectively. These feature points can reflect the local feature information of the image. For example, more feature points will be extracted at the edges of the finger and in areas with obvious texture changes. Taking the middle finger and the ring finger as an example, feature points will be detected at the positions of the vein patterns and joints of these fingers in the image; Step 1.3.2: Find the feature points corresponding to the same physical position on different images and perform feature matching; specifically, find the corresponding feature points in the two images, that is, the feature points representing the same physical position in the two images. For the overlapping part of the middle finger and the ring finger, a certain feature point on the ring finger in the camera 1 image should be able to find a matching point at the corresponding position of the ring finger in the camera 2 image. Methods such as brute-force matching or ratio test based on the nearest neighbor distance can be used for matching; Step 1.3.3: Calculate the relative position relationship between different images according to the matched feature points, and align different images in spatial positions through coordinate transformation; Step 1.3.4: Fuse the overlapping areas of different images to obtain a fused multi-finger vein image. After the images are aligned, the overlapping areas need to be fused, and the weighted average method can be used. The weighted average method is to assign different weights according to the distance from the pixel to the image boundary in the overlapping area, and perform weighted summation on the values ​​of the corresponding pixels in the two images.

[0031] Example 3 On the basis of Example 1 or 2, a pre-trained finger segmentation model is used in step 2 to perform finger segmentation; the finger segmentation model is a neural network model built using the PyTorch deep learning framework, and the labeled training data is input into the finger segmentation model for training to learn the characteristics and segmentation patterns of the fingers; during the training process, these labeled position points can be regarded as representative key positions on the fingers, and the model will gradually understand the relationship between these key points and the overall shape, position and boundary of the fingers during the learning process. By identifying and locating these key points, it is helpful to more accurately depict the contours and areas of the fingers. By continuously adjusting the parameters of the model, the prediction results of the model are made as close as possible to the labeled data to improve the accuracy of the model. In particular, the neural network model is a convolutional neural network (CNN) or a fully convolutional network (FCN).

[0032] Preparation of training data includes: collecting multiple finger vein image data of different individuals, using special image annotation tools to annotate the vein areas of the thumb, index finger, middle finger, ring finger and little finger, clearly distinguishing the area of ​​each finger, and annotating 4 to 6 key position points for each finger to assist in subsequent feature extraction and model training, where key position points include but are not limited to finger joints, fingertips, or finger roots.

[0033] Specifically, during the data collection phase, a large amount of multi-finger vein image data of different individuals is collected. These data should cover different ages, genders, skin colors, and different lighting, shooting angles, tilt angles, etc., to ensure that the model has good generalization ability. For example, thousands of finger vein image samples of different people are collected in various environments (such as indoors, outdoors, strong light, weak light, etc.). Then, the identity is associated, that is, the corresponding individual identity information is annotated for each group of multi-finger vein images. For example, if the finger vein images of 2,000 people are collected, a unique identity identifier (such as ID numbers from 1 to 2,000) is assigned to each person's image data. Ensure the accuracy and uniqueness of the identity information to avoid the situation where different people are annotated with the same identity or the same person is annotated with different identities. In addition to the identity identifier, some other information related to the individual can also be annotated, such as age, gender, skin color, etc. This information can be used as auxiliary information to help the model better adapt to the characteristic differences of different people during the training process. For example, when annotating, record the information that a person is 35 years old, male, and medium skin color. In order to further enrich the data set, data enhancement technology can be used. For example, more training samples can be generated through operations such as rotation, flipping, scaling, and noise addition to increase the diversity and robustness of the data.

[0034] Example 4 Based on Examples 1, 2 or 3, in step 3, a pre-trained feature extraction model is used to extract the vein features of each finger; the feature extraction model is a deep learning model, including multiple convolutional layers, pooling layers, transformer layers and fully connected layers. The deep learning model is trained using a training set, and the loss function between the predicted results and the true annotations is calculated. The parameters of the feature extraction model are updated according to the gradient back propagation of the loss function, and the training is repeated until the feature extraction model converges or reaches a preset number of iterations.

[0035] In particular, based on common image extraction models, such as convolutional neural networks (CNN), deep belief networks (DBN), VGGNet, ResNet, etc., appropriate adjustments and optimizations can be made according to the characteristics of finger vein images, including determining the number of network layers, the number of neurons in each layer, activation functions, convolution kernel size and other parameters. Parameters of the constructed deep learning model can be set, including learning rate, batch size, number of iterations, etc.

[0036] In particular, the vein feature of each finger extracted in step 3 corresponds to a feature vector, which can be used as a feature representation of the finger vein for subsequent identity recognition, matching and other tasks; the feature vectors of each finger are spliced ​​in a head-to-tail splicing or weighted splicing manner; In particular, multi-finger feature fusion is to fuse the dorsal vein features of multiple fingers. Assume that Fingers, The feature vector of each finger is expressed as , then the feature vector after concatenation is expressed as: The feature vector after weighted concatenation is expressed as: in, is the weight coefficient.

[0037] Example 5 Based on embodiments 1, 2, 3 or 4, in step 4, the matching result of a single finger is first obtained, and then the matching results are fused by weighted summation; generally speaking, each ID stores 4 vein features, and the matching conditions of multiple fingers need to be comprehensively considered.

[0038] Among them, the matching of a single finger adopts feature matching based on distance measurement or feature matching based on deep learning model; The feature matching based on distance measurement calculates the distance between the concatenated feature vector obtained in step 3 and the stored known feature vector. If the calculated distance is less than a preset threshold, the identification is considered successful. The distance measurement method includes Euclidean distance or Manhattan distance. The feature matching based on the deep learning model will first use a large amount of multi-finger vein feature vector data for model training, and then input the spliced ​​feature vector obtained in step 3 into the trained model to output the matching result.

[0039] In the decision-making mechanism part, it includes: 1) Identity confirmation: When the feature match is successful, the system will record the user's recognition time, recognition results and other information for subsequent query and statistics.

[0040] 2) Authority judgment: Based on the user's identity information, determine whether the user has the right to unlock the door. For example, in a home environment, different family members may have different unlocking permissions. For example, the owner can unlock the door at any time, while the visitor has the right to unlock the door only during a specific time period.

[0041] 3) Exception handling: If feature matching fails or other abnormal situations occur, such as multiple recognition failures, low quality of collected images, etc., the system will trigger the corresponding exception handling mechanism. For example, sound an alarm, send an alarm message to the user's mobile terminal, etc.

[0042] Example 6 The embodiment of the present invention also provides a biometric identification system with multi-feature finger fusion, including a finger groove, a vein image acquisition unit, a core processing unit, and a lock driving unit; The finger groove guides the identification subject to place the finger at the pre-set collection position; The vein image acquisition unit is connected to the finger groove, illuminates the finger through a near-infrared light source, and uses a camera to collect multiple finger vein images; The core processing unit is connected to the vein image acquisition unit, and uses a pre-trained finger segmentation model to perform finger segmentation and a pre-trained feature extraction model to extract the vein features of each finger, and then performs feature matching to identify the user and send out a start signal; The lock drive unit is connected to the core processing unit, receives the opening signal, drives the corresponding lock or device to open, and completes the entire biometric identification process; The system also includes the multi-feature finger fusion biometric recognition method described in any one of Examples 1 to 5.

[0043] In particular, the near-infrared light source is a near-infrared light source of a specific wavelength, specifically 850nm or 940nm, as an illumination source. Because the hemoglobin in the human vein under the irradiation of near-infrared light has different light absorption characteristics of the wavelength from the surrounding tissue, the vein pattern can be clearly displayed. A near-infrared LED lamp with moderate power and stable light emission can be selected, and multiple LED lamps can be selected and arranged to ensure that the light is evenly irradiated on the back of the finger. Generally speaking, in order to ensure uniform illumination, multiple infrared lamps may be required. If the acquisition area is small, 3 to 5 infrared lamps may meet the needs; but if the acquisition area is large or higher illumination uniformity is required, more infrared lamps may be required, such as 6 to 8 or even more. However, too many infrared lamps may increase the cost and power consumption of the equipment, so it is necessary to balance the illumination uniformity and cost and power consumption. The optimal number of infrared lamps can be determined through experiments and simulations. For example, different numbers of infrared lamps are used to irradiate the sample, collect vein images and analyze the image quality to determine the most appropriate number of infrared lamps. In order to achieve uniform illumination, the positions of the infrared lamps should be distributed around the acquisition areas of the fingers and the back of the hand. It can be arranged in a ring or matrix. In addition, you can also consider using reflective materials or lenses to optimize the light distribution. For example, set up reflective plates around the infrared lamp to reflect the light to the collection area to improve the utilization rate of the light; or use lenses to focus the light to make the light more concentrated and uniform.

[0044] In particular, the image sensor on the camera uses a high-pixel, high-sensitivity image sensor to capture vein images. For example, a CMOS sensor with a resolution of more than 2 million pixels is selected to ensure that the subtle features of the veins can be clearly captured. The optical lens system is equipped with high-quality optical lenses, including multi-layer coated lenses, to reduce light reflection and scattering, and improve image clarity and contrast. The focal length and aperture size of the lens are carefully designed to meet the acquisition requirements of different finger sizes and distances. 1) Select a camera with a suitable FOV according to the acquisition requirements. The FOV of the camera should be able to adapt to the finger sizes of different users, while avoiding the need for users to position their fingers too precisely when placing them. 2) Since the distance of shooting and collecting fingers is too close, using a single camera cannot meet the requirements of collecting multiple fingers at the same time. You can consider using a combination of multiple cameras to achieve a larger FOV or higher resolution. For example, use two or more cameras to collect images of fingers and backs of hands from different angles, and then combine these images into a complete image through image stitching technology. 3) When using two / more cameras, you need to ensure compatibility between hardware. Different camera models may have different interfaces, resolutions, frame rates and other parameters. If the hardware is incompatible, it may cause image acquisition failure or problems in the stitching process. In order to ensure the accuracy of stitching, the two cameras need to synchronize the acquisition of images in time. If there is a problem with synchronization, the acquired images may be deviated in time, causing the stitched images to appear dynamic blur or discontinuous.

[0045] In particular, the system also includes a touch wake-up device, and when the identification object contacts the touch wake-up device, the system is awakened from a low power state. The touch wake-up device usually works based on the principle of capacitive sensing or resistive sensing. The output pin of the touch sensor chip is connected to an interrupt input pin of the main control chip of the multi-finger vein recognition system so that the system can be woken up in time when a touch event is detected. When a human body touches the sensing area of ​​the device, the touch wake-up device determines whether a touch event occurs by detecting changes in capacitance or resistance. The multi-finger vein recognition system is activated or awakened by the touch wake-up device. When no touch is detected, the recognition system enters a low power state.

[0046] In particular, the touch detection and wake-up process: 1) When the system is in low power state, the interrupt input pin of the main control chip is in a listening state, waiting for the signal from the touch sensor.

[0047] 2) Once a touch event is detected, the touch sensor chip outputs an interrupt signal to the main control chip. The main control chip immediately wakes up from low power mode and enters normal working mode.

[0048] 3) The main control chip reads the status register of the touch sensor chip to confirm the validity of the touch event. If it is a valid touch event, it starts to start each module of the multi-finger vein recognition system and prepares for finger vein recognition.

[0049] 4) After the multi-finger vein recognition system completes a recognition operation, if no new touch event is detected within a certain period of time, the main control chip switches the system back to low power consumption mode.

[0050] The multi-feature finger fusion biometric recognition method and system provided by the embodiment of the present invention can be applied to car unlocking, smart door locks, access control systems, attendance systems, safes, etc.

[0051] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A biometric identification method for multi-feature finger fusion, characterized in that: The method comprises the following steps: Step 1, collecting multiple finger vein images, including the front or back of multiple fingers; Step 2: segmenting the multiple finger vein images to obtain the position and boundary information of each finger, and separating the vein image area of ​​each finger; Step 3: Extract finger vein features, obtain the vein features of each finger, and then perform feature fusion to obtain multi-finger fusion features; Step 4: Perform feature matching, identify the identity, and drive the device to turn on to complete the entire biometric identification.

2. The biometric identification method of multi-feature finger fusion as claimed in claim 1, characterized in that: The step 1 includes the following sub-steps: Step 1.1, use the finger groove to guide the identification object to place the finger at the pre-set collection position; Step 1.2, surround the collection area with several infrared lamps to form a near-infrared light source to illuminate the finger; Step 1.3: Use a CMOS sensor and a camera with FOV to collect the multiple finger vein images.

3. The biometric identification method of multi-feature finger fusion as claimed in claim 2, characterized in that: The shooting scheme adopted in step 1.3 is a single-camera scheme or a multi-camera scheme, wherein the multi-camera scheme uses two or more cameras to capture images of the front or back of multiple fingers from different angles.

4. The biometric identification method of multi-feature finger fusion as claimed in claim 3, characterized in that: The multi-camera solution is provided with an image stitching method, comprising the following sub-steps: Step 1.3.1, extract feature points from different images respectively; Step 1.3.2, find the feature points corresponding to the same physical location on different images and perform feature matching; Step 1.3.3, calculate the relative position relationship between different images based on the matched feature points, and align the different images in space through coordinate transformation; Step 1.3.4: Fusing the overlapping areas of different images to obtain the fused multi-finger vein image.

5. The biometric identification method of multi-feature finger fusion as claimed in claim 1, characterized in that: In step 2, a pre-trained finger segmentation model is used to perform finger segmentation; the finger segmentation model is a neural network model built using the PyTorch deep learning framework, and the labeled training data is input into the finger segmentation model for training to learn the characteristics and segmentation patterns of the fingers; the preparation of the training data includes: collecting multiple finger vein image data from different individuals, annotating the vein areas of the thumb, index finger, middle finger, ring finger and little finger, and annotating 4 to 6 key position points for each finger, wherein the key position points include but are not limited to the joints, fingertips, or bases of the fingers.

6. The biometric identification method of multi-feature finger fusion as claimed in claim 1, characterized in that: In step 3, a pre-trained feature extraction model is used to extract the vein features of each finger; the feature extraction model is a deep learning model, including multiple convolutional layers, pooling layers, transformer layers and fully connected layers. The deep learning model is trained using a training set, and the loss function between the predicted result and the true annotation is calculated. The parameters of the feature extraction model are updated according to the gradient back propagation of the loss function, and the training is repeated until the feature extraction model converges or reaches a preset number of iterations.

7. The biometric identification method of multi-feature finger fusion as claimed in claim 6, characterized in that: The vein feature of each finger extracted in step 3 corresponds to a feature vector, and the feature vectors of each finger are spliced ​​in a head-to-tail splicing or weighted splicing manner; Assume that Fingers, The feature vector of each finger is expressed as , then the feature vector after concatenation is expressed as: The feature vector after weighted concatenation is expressed as: in, is the weight coefficient.

8. The biometric identification method of multi-feature finger fusion as claimed in claim 7, characterized in that: In step 4, the matching results of a single finger are first obtained, and then the matching results are merged by weighted summation; Among them, the matching of a single finger adopts feature matching based on distance measurement or feature matching based on deep learning model; The feature matching based on distance measurement calculates the distance between the concatenated feature vector obtained in step 3 and the stored known feature vector. If the calculated distance is less than a preset threshold, it is considered that the identity is successfully identified, wherein the distance measurement method includes Euclidean distance or Manhattan distance; The feature matching based on the deep learning model will first use a large amount of multi-finger vein feature vector data to train the model, and then input the spliced ​​feature vector obtained in step 3 into the trained model to output the matching result.

9. A biometric recognition system with multi-feature finger fusion, characterized in that: The system includes a finger groove, a vein image acquisition unit, a core processing unit, and a lock drive unit; The finger groove guides the identification subject to place the finger at a preset collection position; The vein image acquisition unit is connected to the finger groove, illuminates the finger through a near-infrared light source, and uses a camera to acquire the vein images of multiple fingers; The core processing unit is connected to the vein image acquisition unit, uses a pre-trained finger segmentation model to perform finger segmentation and a pre-trained feature extraction model to extract vein features of each finger, then performs feature matching, identifies the user identity, and sends a start signal; The lock driving unit is connected to the core processing unit, receives the opening signal, drives the corresponding lock or device to open, and completes the entire biometric identification process; The system also includes the multi-feature finger fusion biometric recognition method as described in any one of claims 1 to 8.

10. The multi-feature finger fusion biometric recognition system as claimed in claim 9, characterized in that: The system further comprises a touch awakening device, and when an identification object contacts the touch awakening device, the system is awakened from a low power consumption state.

Images