Multi-mode biological recognition access control system

Abstract

The invention relates to the technical field of access control systems, and discloses a multi-mode biological recognition access control system, a biological recognition module is used for collecting biological characteristic images of a user, and the biological characteristic images at least comprise a fingerprint image, a face image and an iris image; the data processing module is used for performing data processing and analysis on the biological feature image acquired by the biological recognition module, extracting an effective feature value, and obtaining a target feature image; the identity determination module is used for performing multi-mode identification on the target feature image to obtain an identification result, and performing comparison to determine the identity of the user; the control module is used for controlling the opening or closing of the access control according to the comparison result of the identity determination module; the storage module is used for storing a biological characteristic template and access control parameter information of a user; the communication module is used for carrying out communication and linkage with a mobile terminal in a wired communication or wireless communication mode; the safety of the access control system is effectively improved, and the accuracy of identity verification is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of access control systems, and in particular to a multimodal biometric access control system. Background Art

[0002] With the continuous development of science and technology, access control systems are being used more and more widely in various fields. Traditional access control systems mainly use keys, passwords, swipe cards and other methods for identity authentication, but these methods have problems such as easy loss, easy cracking, and low security. In order to improve the security and accuracy of access control systems, biometric technology has gradually been applied to access control systems.

[0003] At present, common biometric technologies include fingerprint recognition, face recognition, iris recognition, etc. These biometric technologies have their own advantages and disadvantages. A single biometric technology is difficult to meet the needs of high-security access control. Therefore, the development of a multimodal biometric access control system has important practical significance. Summary of the invention

[0004] The purpose of the present invention is to solve the above problems and design a multimodal biometric access control system that combines multiple biometric technologies to improve the security and accuracy of the access control system.

[0005] The present invention provides a multimodal biometric access control system, which includes a biometric recognition module, a data processing module, an identity determination module, a control module, a storage module and a communication module, wherein:

[0006] A biometric recognition module, used to collect a user's biometric image, wherein the biometric image includes at least a fingerprint image, a facial image, and an iris image;

[0007] A data processing module, used to process and analyze the biometric feature image collected by the biometric recognition module, extract valid feature values, and obtain a target feature image;

[0008] An identity determination module is used to perform multimodal recognition on the target feature image, obtain a recognition result, and perform a comparison to determine the identity of the user;

[0009] A control module, used to control the opening or closing of the door access according to the comparison result of the identity determination module;

[0010] A storage module, used to store the user's biometric template and access control parameter information;

[0011] The communication module is used to communicate and link with the mobile terminal by wired communication or wireless communication.

[0012] Optionally, in a first implementation of the present invention, data processing and analysis are performed on the biometric feature image collected by the biometric recognition module to extract valid feature values ​​to obtain a target feature image, including:

[0013] Acquire the biometric image output by the biometric recognition module, screen the biometric image by using a k-means clustering algorithm, and classify the biometric image into two categories: usable images and unusable images;

[0014] The LeNet network is used to identify the available images that have been screened out to obtain detailed feature images containing effective features;

[0015] The detail feature image is enhanced by histogram equalization, and the enhanced detail feature image is enhanced by Clahe to obtain a target feature image.

[0016] Optionally, in a second implementation of the present invention, the step of obtaining the biometric image output by the biometric recognition module, screening the biometric image by using a k-means clustering algorithm, and dividing the biometric image into two categories: usable images and unusable images, includes:

[0017] Obtain the biometric image output by the biometric recognition module, and calculate the X of every two data points in the original data set of the biometric image. i , X j The Euclidean distance between them is calculated, and the average distance is calculated based on the Euclidean distance of all points;

[0018] Traverse the Euclidean distance of all points, and based on the average distance, calculate the density p(i) of all data points and the average distance A within the cluster i and parameter λ i :

[0019]

[0020] Select Lambda i The largest data point, the data points with a distance less than the average distance to this data point are added to the new data set, and the corresponding data points are deleted in the original data set, and the iteration is repeated until all points are classified;

[0021] The intra-cluster error square of each data point in the new data set is calculated, and the point with the smallest intra-cluster error square is selected as the initial center point of the cluster. The distance between each data point in the new data set and all the initial center points is calculated, and all data points are classified into k classes. Repeat the iteration until the new cluster center remains unchanged, so as to classify the biometric image into two categories: usable image and unusable image.

[0022] Optionally, in a third implementation of the present invention, the step of using a LeNet network to identify the selected available images to obtain a detail feature image containing effective features includes:

[0023] The first three convolutional pooling layers of the LeNet network learn the features of the input image through different convolutional layers. The following two fully connected layers perform spatial mapping of the convolutional features to high-dimensional features and obtain detailed feature images containing effective features. The convolution kernel of each convolutional layer is 5×5 in size, and each convolutional layer has a 2×2 maximum pooling layer. Nonlinearity is introduced through the ReLU activation function.

[0024] Optionally, in a fourth implementation of the present invention, histogram equalization enhancement is performed on the detail feature image, and Clahe enhancement is performed on the enhanced detail feature image to obtain a target feature image, including:

[0025] Divide the enhanced detail feature image into several sub-blocks of equal size, and make sure there are no overlapping pixels between different sub-blocks;

[0026] Calculate the histogram M of each sub-block and the cropping limit threshold δ:

[0027]

[0028] Where N x and N y represents the number of pixels of the sub-block in the horizontal and vertical directions, and C is the contrast limit coefficient;

[0029] Each sub-block is cropped according to the calculated δ, and the pixels exceeding δ are evenly distributed to the histogram of the sub-block;

[0030] Perform HE and bilinear interpolation operations on the new histogram of the sub-block to obtain the target feature image.

[0031] Optionally, in a fifth implementation of the present invention, performing multimodal recognition on the target feature image to obtain a recognition result, and performing comparison to determine the identity of the user includes:

[0032] Using three networks, namely, a fingerprint recognition network model, a face recognition network model and an iris recognition network model, to extract features of the fingerprint, face and iris of the target feature image in parallel;

[0033] The fingerprint feature vector obtained from the fingerprint recognition network model, the facial feature vector obtained from the facial recognition network model, and the iris feature vector obtained from the iris recognition network model are sequentially arranged and combined;

[0034] Assign a weight to the eigenvector of each mode and multiply the eigenvector of each mode by the corresponding weight:

[0035] Y=αX

[0036] In the formula, α represents the specific weight parameter of the corresponding modal feature, X represents the input of the adaptive weight layer, and Y represents the output of the proposed adaptive weight layer;

[0037] The weighted feature vectors are added together to obtain a fused multimodal feature vector, and the fused multimodal feature vector is converted into a recognition result.

[0038] Optionally, in a sixth implementation of the present invention, the three networks of fingerprint recognition network model, face recognition network model and iris recognition network model are used to extract features of the fingerprint, face and iris of the target feature image in parallel, including:

[0039] The fingerprint recognition network model adopts VGG-FingerprintNet, and extracts features of the fingerprint of the target feature image through multiple convolutional layers. The convolutional layer learns fingerprint texture features of different scales and directions, reduces the resolution of the feature map through the pooling layer, and converts the extracted fingerprint features into fingerprint feature vectors through the fully connected layer;

[0040] The facial recognition network model adopts ResNet-FaceNet, the face of the input target feature image is subjected to feature extraction through a convolutional layer and a residual block, and the facial features are converted into facial feature vectors through a global average pooling layer;

[0041] The iris recognition network model adopts DeepIrisNet, which is composed of multiple convolutional layers and pooling layers. It removes noise and interference from the iris of the input target feature image, gradually extracts the texture features of the iris through the convolutional layers and pooling layers, and finally converts the iris features into iris feature vectors through the fully connected layer.

[0042] Optionally, in a seventh implementation of the present invention, the control module includes a verification pass module and a verification fail module, wherein:

[0043] The verification module is used to control the module to send a door opening signal to the access control actuator to open the door when the user identity verification is passed;

[0044] The verification failure module is used to control the module to send out an alarm signal and record relevant information when the user identity verification fails.

[0045] In the technical solution provided by the present invention, the multimodal biometric access control system includes a biometric module, a data processing module, an identity determination module, a control module, a storage module and a communication module, and combines multiple biometric technologies, such as fingerprint recognition, face recognition, iris recognition, etc. These biometric technologies have the characteristics of uniqueness, stability and non-replicability, and can effectively improve the security of the access control system; the combination of multiple biometric technologies can make up for the shortcomings of a single biometric technology and improve the accuracy of identity authentication; the user only needs to stay briefly in front of the access control system to complete the identity authentication, which improves the use efficiency of the access control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0047] Figure 1 A schematic diagram of the structure of a multimodal biometric access control system provided by an embodiment of the present invention;

[0048] Figure 2 A flowchart of a data processing module provided by an embodiment of the present invention;

[0049] Figure 3 This is a flow chart of an identity determination module provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0051] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 A schematic diagram of the structure of a multimodal biometric access control system provided by an embodiment of the present invention includes a biometric recognition module, a data processing module, an identity determination module, a control module, a storage module and a communication module, wherein:

[0052] A biometric recognition module, used to collect a user's biometric image, wherein the biometric image includes at least a fingerprint image, a facial image, and an iris image;

[0053] The data processing module is used to process and analyze the biometric feature images collected by the biometric recognition module, extract valid feature values, and obtain target feature images;

[0054] The identity determination module is used to perform multimodal recognition on the target feature image, obtain the recognition result, and perform comparison to determine the identity of the user;

[0055] A control module, used to control the opening or closing of the door access according to the comparison result of the identity determination module;

[0056] A storage module, used to store the user's biometric template and access control parameter information;

[0057] The communication module is used to communicate and link with the mobile terminal by wired communication or wireless communication.

[0058] In this embodiment, the control module includes a verification pass module and a verification fail module, wherein:

[0059] The verification module is used to control the module to send a door opening signal to the access control actuator to open the door when the user identity verification is passed;

[0060] The verification failure module is used to control the module to send out an alarm signal and record relevant information when the user identity verification fails.

[0061] See also Figure 2 , a flow chart of a data processing module provided by an embodiment of the present invention, the data processing module includes:

[0062] Obtain the biometric image output by the biometric recognition module, screen the biometric image through the k-means clustering algorithm, and divide the biometric image into two categories: usable images and unusable images;

[0063] The LeNet network is used to identify the available images that have been screened out to obtain detailed feature images containing effective features;

[0064] The detail feature image is enhanced by histogram equalization, and the enhanced detail feature image is enhanced by Clahe to obtain the target feature image.

[0065] In this embodiment, the biometric feature images output by the biometric recognition module are obtained, and the biometric feature images are screened by the k-means clustering algorithm to classify the biometric feature images into two categories: usable images and unusable images, including:

[0066] Get the biometric image output by the biometric recognition module and calculate the X of every two data points in the original data set of the biometric image. i , X j The Euclidean distance between them is calculated, and the average distance is calculated based on the Euclidean distance of all points;

[0067] Traverse the Euclidean distance of all points, and based on the average distance, calculate the density p(i) of all data points and the average distance A within the cluster i and parameter λ i :

[0068]

[0069] Select Lambda i The largest data point, the data points with a distance less than the average distance to this data point are added to the new data set, and the corresponding data points are deleted in the original data set, and the iteration is repeated until all points are classified;

[0070] Calculate the intra-cluster error square of each data point in the new data set, select the point with the smallest intra-cluster error square as the initial center point of the cluster, calculate the distance between each data point in the new data set and all the initial center points, classify all data points into k classes, and repeat the iteration until the new cluster center remains unchanged, so as to classify the biometric images into two categories: usable images and unusable images.

[0071] In this embodiment, the LeNet network is used to identify the filtered available images to obtain detail feature images containing effective features, including:

[0072] The first three convolutional pooling layers of the LeNet network learn the features of the input image through different convolutional layers. The following two fully connected layers perform spatial mapping of the convolutional features to high-dimensional features and obtain detailed feature images containing effective features. The convolution kernel of each convolutional layer is 5×5 in size, and each convolutional layer has a 2×2 maximum pooling layer. Nonlinearity is introduced through the ReLU activation function.

[0073] In this embodiment, histogram equalization is performed on the detail feature image, and Clahe enhancement is performed on the enhanced detail feature image to obtain a target feature image, including:

[0074] Divide the enhanced detail feature image into several sub-blocks of equal size, and make sure there are no overlapping pixels between different sub-blocks;

[0075] Calculate the histogram M of each sub-block and the cropping limit threshold δ:

[0076]

[0077] Where N x and Ny represents the number of pixels of the sub-block in the horizontal and vertical directions, and C is the contrast limit coefficient;

[0078] Each sub-block is cropped according to the calculated δ, and the pixels exceeding δ are evenly distributed to the histogram of the sub-block;

[0079] Perform HE and bilinear interpolation operations on the new histogram of the sub-block to obtain the target feature image.

[0080] See also Figure 3 , a flow chart of an identity determination module provided by an embodiment of the present invention, the identity determination module includes:

[0081] The fingerprint recognition network model, face recognition network model and iris recognition network model are used to extract the fingerprint, face and iris of the target feature image in parallel.

[0082] The fingerprint feature vector obtained from the fingerprint recognition network model, the facial feature vector obtained from the facial recognition network model, and the iris feature vector obtained from the iris recognition network model are sequentially arranged and combined;

[0083] Assign a weight to the eigenvector of each mode and multiply the eigenvector of each mode by the corresponding weight:

[0084] Y=αX

[0085] In the formula, α represents the specific weight parameter of the corresponding modal feature, X represents the input of the adaptive weight layer, and Y represents the output of the proposed adaptive weight layer;

[0086] The weighted feature vectors are added together to obtain a fused multimodal feature vector, and the fused multimodal feature vector is converted into a recognition result.

[0087] In this embodiment, three networks, namely, a fingerprint recognition network model, a facial recognition network model and an iris recognition network model, are used in parallel to extract features of the fingerprint, face and iris of the target feature image, including:

[0088] The fingerprint recognition network model uses VGG-FingerprintNet, which extracts the fingerprint of the input target feature image through multiple convolutional layers. The convolutional layer learns the fingerprint texture features of different scales and directions, reduces the resolution of the feature map through the pooling layer, and converts the extracted fingerprint features into fingerprint feature vectors through the fully connected layer;

[0089] The facial recognition network model uses ResNet-FaceNet. The face of the input target feature image is extracted through the convolution layer and residual block, and the facial features are converted into facial feature vectors through the global average pooling layer;

[0090] The iris recognition network model uses DeepIrisNet, which consists of multiple convolutional layers and pooling layers. It removes noise and interference from the iris of the input target feature image, gradually extracts the texture features of the iris through convolutional layers and pooling layers, and finally converts the iris features into iris feature vectors through a fully connected layer.

[0091] VGG-FingerprintNet combines the characteristics of the VGG (Visual Geometry Group) network architecture to process fingerprint images to achieve efficient fingerprint recognition tasks. The VGG network is known for its deep structure, which helps capture complex features in fingerprint images. This deep structure includes multiple convolutional layers that can gradually extract image features from low-level to high-level. In the VGG architecture, fixed-size convolution kernels (usually 3x3) are used, which helps reduce the number of parameters while maintaining effective feature extraction capabilities for input images. In order to improve the accuracy of fingerprint recognition, VGG-FingerprintNet uses multi-scale feature fusion technology, that is, extracting features from network outputs at different levels and integrating them to enhance the model's sensitivity to fingerprint details. Due to the use of deep learning methods, VGG-FingerprintNet can automatically learn the feature representation of fingerprints through a large amount of training data, thus having strong generalization and adaptability.

[0092] ResNet-FaceNet is a face recognition model that combines ResNet (Residual Network) and FaceNet technologies. It uses the deep network structure of ResNet to extract high-dimensional features of face images, and optimizes the distance metric in the feature space through the triplet loss function of FaceNet, so that the distance between facial features of the same identity is as small as possible, while the distance between facial features of different identities is as large as possible. The core of ResNet is the residual block, which solves the gradient vanishing or exploding problem that occurs during the training of deep neural networks by introducing skip connections. This structure allows information to be directly passed from one layer to another, thereby improving the training efficiency and performance of the network; FaceNet uses the triplet loss function to train the model. Each triplet contains an anchor sample (Anchor), a positive sample (Positive, a different picture of the same person as the anchor) and a negative sample (Negative, a picture of a different person from the anchor), with the goal of minimizing the distance between the anchor and the positive sample, while maximizing the distance between the anchor and the negative sample.

[0093] As a high-precision biometric technology, iris recognition has been widely used in many fields due to its unique stability and security, such as secure access control, identity authentication, and personal identification. DeepIrisNet uses the powerful ability of deep learning to improve the accuracy and efficiency of iris recognition. DeepIrisNet is built on a deep convolutional neural network (CNN), which can automatically learn and extract complex feature representations from iris images. This automatic feature learning capability reduces the need for manual feature engineering and improves the generalization ability of the model;: The model supports an end-to-end learning process, that is, learning directly from the original iris image to the final recognition result without the need for manual feature selection or preprocessing steps in the middle; In order to better capture the detailed information in the iris image, DeepIrisNet uses multi-scale feature extraction technology, that is, analyzing the iris image at different scales to obtain a more comprehensive feature representation; In order to increase the robustness and generalization ability of the model, DeepIrisNet may apply a variety of data augmentation techniques during the training process, such as rotation, scaling, translation, etc., so that the model can correctly recognize the iris under more diverse conditions.

[0094] Through the implementation of the above scheme, the multimodal biometric access control system combines multiple biometric technologies, such as fingerprint recognition, face recognition, iris recognition, etc. These biometric technologies have the characteristics of uniqueness, stability and non-replicability, which can effectively improve the security of the access control system; the combination of multiple biometric technologies can make up for the shortcomings of a single biometric technology and improve the accuracy of identity authentication. For example, fingerprint recognition may be affected by factors such as finger moisture and stains, while face recognition may be affected by factors such as light and angle, and iris recognition is relatively stable. Through multimodal biometrics, various factors can be comprehensively considered to improve the accuracy of identity authentication; the multimodal biometric access control system adopts advanced algorithms and hardware equipment, which can achieve rapid identification. Users only need to stay in front of the access control system for a short time to complete identity authentication, which improves the efficiency of the access control system; the multimodal biometric access control system can adapt to different environments and user needs. For example, for some users with disabled or worn fingers, face recognition or iris recognition can be used for identity authentication; for some places with dim light, infrared face recognition and other technologies can be used.

[0095] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A multimodal biometric access control system, characterized in that: The multimodal biometric access control system includes a biometric module, a data processing module, an identity determination module, a control module, a storage module and a communication module, wherein: A biometric recognition module, used to collect a user's biometric image, wherein the biometric image includes at least a fingerprint image, a facial image, and an iris image; A data processing module, used to process and analyze the biometric feature image collected by the biometric recognition module, extract valid feature values, and obtain a target feature image; An identity determination module is used to perform multimodal recognition on the target feature image, obtain a recognition result, and perform a comparison to determine the identity of the user; A control module, used to control the opening or closing of the door access according to the comparison result of the identity determination module; A storage module, used to store the user's biometric template and access control parameter information; The communication module is used to communicate and link with the mobile terminal by wired communication or wireless communication.

2. A steep slope surface slope detection method based on image processing as claimed in claim 1, characterized in that: Processing and analyzing the biometric image collected by the biometric recognition module, extracting effective feature values, and obtaining a target feature image, including: Acquire the biometric image output by the biometric recognition module, screen the biometric image by using a k-means clustering algorithm, and classify the biometric image into two categories: usable images and unusable images; The LeNet network is used to identify the available images that have been screened out to obtain detailed feature images containing effective features; The detail feature image is enhanced by histogram equalization, and the enhanced detail feature image is enhanced by Clahe to obtain a target feature image.

3. A steep slope surface slope detection method based on image processing as claimed in claim 2, characterized in that: The step of obtaining the biometric image output by the biometric recognition module, screening the biometric image by using a k-means clustering algorithm, and classifying the biometric image into two categories: usable images and unusable images, includes: Obtain the biometric image output by the biometric recognition module, and calculate the X of every two data points in the original data set of the biometric image. i , X j The Euclidean distance between them is calculated, and the average distance is calculated based on the Euclidean distance of all points; Traverse the Euclidean distance of all points, and based on the average distance, calculate the density pi of all data points and the average distance A within the cluster i and parameter λ i : Select Lambda i The largest data point, the data points with a distance less than the average distance to this data point are added to the new data set, and the corresponding data points are deleted in the original data set, and the iteration is repeated until all points are classified; The intra-cluster error square of each data point in the new data set is calculated, and the point with the smallest intra-cluster error square is selected as the initial center point of the cluster. The distance between each data point in the new data set and all the initial center points is calculated, and all data points are classified into k classes. Repeat the iteration until the new cluster center remains unchanged, so as to classify the biometric image into two categories: usable image and unusable image.

4. A steep slope surface slope detection method based on image processing as claimed in claim 2, characterized in that: The LeNet network is used to identify the screened available images to obtain a detail feature image containing effective features, including: The first three convolutional pooling layers of the LeNet network learn the features of the input image through different convolutional layers. The following two fully connected layers perform spatial mapping of the convolutional features to high-dimensional features and obtain detailed feature images containing effective features. The convolution kernel of each convolutional layer is 5×5 in size, and each convolutional layer has a 2×2 maximum pooling layer. Nonlinearity is introduced through the ReLU activation function.

5. The method for detecting the steep slope surface based on image processing according to claim 2, characterized in that: Performing histogram equalization enhancement on the detail feature image, and performing Clahe enhancement on the enhanced detail feature image to obtain a target feature image, including: Divide the enhanced detail feature image into several sub-blocks of equal size, and make sure there are no overlapping pixels between different sub-blocks; Calculate the histogram M of each sub-block and the cropping limit threshold δ: Where N x and N y represents the number of pixels of the sub-block in the horizontal and vertical directions, and C is the contrast limit coefficient; Each sub-block is cropped according to the calculated δ, and the pixels exceeding δ are evenly distributed to the histogram of the sub-block; Perform HE and bilinear interpolation operations on the new histogram of the sub-block to obtain the target feature image.

6. The method for detecting the steep slope surface based on image processing according to claim 1, characterized in that: The performing multimodal recognition on the target feature image to obtain a recognition result, and performing a comparison to determine the identity of the user, includes: Using three networks, namely, a fingerprint recognition network model, a face recognition network model and an iris recognition network model, to extract features of the fingerprint, face and iris of the target feature image in parallel; The fingerprint feature vector obtained from the fingerprint recognition network model, the facial feature vector obtained from the facial recognition network model, and the iris feature vector obtained from the iris recognition network model are sequentially arranged and combined; Assign a weight to the eigenvector of each mode and multiply the eigenvector of each mode by the corresponding weight: Y=αX In the formula, α represents the specific weight parameter of the corresponding modal feature, X represents the input of the adaptive weight layer, and Y represents the output of the proposed adaptive weight layer; The weighted feature vectors are added together to obtain a fused multimodal feature vector, and the fused multimodal feature vector is converted into a recognition result.

7. A steep slope surface slope detection method based on image processing as claimed in claim 6, characterized in that: The method of using the fingerprint recognition network model, the face recognition network model and the iris recognition network model to extract features of the fingerprint, face and iris of the target feature image in parallel includes: The fingerprint recognition network model adopts VGG-FingerprintNet, and extracts features of the fingerprint of the target feature image through multiple convolutional layers. The convolutional layer learns fingerprint texture features of different scales and directions, reduces the resolution of the feature map through the pooling layer, and converts the extracted fingerprint features into fingerprint feature vectors through the fully connected layer; The facial recognition network model adopts ResNet-FaceNet, the face of the input target feature image is subjected to feature extraction through a convolutional layer and a residual block, and the facial features are converted into facial feature vectors through a global average pooling layer; The iris recognition network model adopts DeepIrisNet, which is composed of multiple convolutional layers and pooling layers. It removes noise and interference from the iris of the input target feature image, gradually extracts the texture features of the iris through the convolutional layers and pooling layers, and finally converts the iris features into iris feature vectors through the fully connected layer.

8. The method for detecting the steep slope surface based on image processing according to claim 1, characterized in that: The control module includes a verification pass module and a verification fail module, wherein: The verification module is used to control the module to send a door opening signal to the access control actuator to open the door when the user identity verification is passed; The verification failure module is used to control the module to send out an alarm signal and record relevant information when the user identity verification fails.

Images