Wind farm authority management system and method

Abstract

The application relates to the field of wind farm intelligent management, and particularly discloses a wind farm permission management system and a method thereof, which extracts features of a multi-channel image fused by a face image, a local binary pattern image and a Canny edge detection image of a user to be assigned with permission by using a convolutional neural network model with a double attention mechanism, so as to respectively focus on feature content and feature position in the image, complement each other to a certain extent, improve feature extraction effect of the network, and in the process, also homogenize a comparison search space in the same direction to constrain feature representation of a feature map obtained from the convolutional neural network model with the double attention mechanism to an isotropic and distinguishable representation space, and then enhance feature distribution continuity of high-dimensional feature expression after fusion, so that the predetermined permission assigned to the user can be accurately judged.

Description

Technical Field

[0001] This invention relates to the field of intelligent management of wind farms, and more specifically, to a wind farm access control system and method thereof. Background Technology

[0002] Wind power companies are increasingly emphasizing data-driven software management. For example, software can be used to collect and store data generated during the operation of wind turbine systems, and to organize and analyze this data. To facilitate software management, different permissions are typically configured for different accounts on the wind power management software to prevent accidental data deletion.

[0003] However, since account and password information can be leaked, a permission management system is needed to improve security by performing secondary verification of user identity when configuring permissions for corresponding users. Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a wind farm access control system and method. This system uses a convolutional neural network model with a dual attention mechanism to extract features from a multi-channel image fused from the user's face image, local binary pattern image, and Canny edge detection image. This focuses on both the feature content and feature location within the image, complementing each other to a certain extent and improving the network's feature extraction performance. Furthermore, by comparing the search space for homogenization, the feature representation of the feature map obtained from the dual attention mechanism convolutional neural network model is constrained to an isotropic and discriminative representation space, thereby enhancing the continuity of the feature distribution in the fused high-dimensional feature representation. This enables accurate determination of the predetermined permissions assigned to a user.

[0005] According to one aspect of this application, a wind farm access control system is provided, comprising:

[0006] The face image acquisition module is used to acquire the face images of users to be assigned permissions;

[0007] The image processing module is used to perform local binarization processing and Canny edge detection on the face image of the user to be assigned permissions to obtain a local binary pattern map and a Canny edge detection map.

[0008] The input extension module is used to arrange the face image, the local binary pattern map, and the Canny edge detection map into a multi-channel image;

[0009] The first convolutional coding module is used to obtain a first feature map from the multi-channel image by using a first convolutional neural network model with a spatial attention mechanism.

[0010] The second convolutional coding module is used to obtain a second feature map from the multi-channel image by using a second convolutional neural network model with a channel attention mechanism.

[0011] The first correction module is used to correct the feature values ​​at each position in the first feature map based on the second feature map to obtain the corrected first feature map;

[0012] The second correction module is used to correct the feature values ​​at each position in the second feature map based on the first feature map to obtain the corrected second feature map;

[0013] A fusion module is used to fuse the corrected first feature map and the corrected second feature map to obtain a classification feature map;

[0014] A face recognition module is used to pass the classification feature map through a classifier to obtain a classification result, wherein the classification result is used to represent the object label to which the face image of the user to be assigned permissions belongs; and

[0015] The permission management module is used to determine whether to assign predetermined permissions to the user based on the classification results.

[0016] In the aforementioned wind farm access control system, the first convolutional coding module is further configured to: perform the following operations on the input data during the forward propagation of the first convolutional neural network model: perform convolution processing on the input data based on a two-dimensional convolutional kernel to generate a convolutional feature map; perform pooling processing on the convolutional feature map to generate a pooled feature map; perform activation processing on the pooled feature map to generate an activation feature map; perform global average pooling along the channel dimension on the activation feature map to obtain a spatial feature matrix; perform convolution processing and activation processing on the spatial feature matrix to generate a weight vector; and weight each feature matrix of the activation feature map with the weight values ​​at each position in the weight vector to obtain a generated feature map; wherein the generated feature map output by the last layer of the first convolutional neural network model is the first feature map.

[0017] In the aforementioned wind farm access control system, the second convolutional coding module is further configured to: perform the following operations on the input data during the forward propagation of each layer of the second convolutional neural network model: perform convolution processing on the input data based on a two-dimensional convolutional kernel to generate a convolutional feature map; perform pooling processing on the convolutional feature map to generate a pooled feature map; perform activation processing on the pooled feature map to generate an activated feature map; calculate the global mean of each feature matrix along the channel dimension of the activated feature map to obtain a channel feature vector; calculate the ratio of the feature value at each position in the channel feature vector to the weighted sum of the feature values ​​at all positions in the channel feature vector to obtain a channel weighted feature vector; and perform dot product on the feature matrix along the channel dimension of the activated feature map using the feature values ​​at each position of the channel weighted feature vector as weights to obtain a generated feature map; wherein, the generated feature map output by the last layer of the second convolutional neural network model is the second feature map.

[0018] In the aforementioned wind farm access control system, the first correction module is further configured to: based on the second feature map, correct the feature values ​​at each position in the first feature map using the following formula to obtain the corrected first feature map;

[0019] The formula is as follows:

[0020]

[0021] Where f 1i and f 2j These are the feature values ​​of the first feature map and the second feature map, respectively, and ρ is a control hyperparameter, and d(f 1i ,f 2j ) represents the distance between eigenvalues.

[0022] In the aforementioned wind farm access control system, the second correction module is further configured to: based on the first feature map, correct the feature values ​​at each position in the second feature map using the following formula to obtain the corrected second feature map;

[0023] The formula is as follows:

[0024]

[0025] Where f 1i and f 2j These are the feature values ​​of the first feature map and the second feature map, respectively, and ρ is a control hyperparameter, and d(f 1i ,f 2j ) represents the distance between eigenvalues.

[0026] In the aforementioned wind farm access control system, the fusion module is further configured to: fuse the corrected first feature map and the corrected second feature map using the following formula to obtain the classification feature map;

[0027] The formula is as follows:

[0028] F s =αF1+βF2

[0029] Among them, F s F1 is the corrected first feature map, F2 is the corrected second feature map, "+" indicates that the elements at corresponding positions of the corrected first feature map and the corrected second feature map are added together, and α and β are weighting parameters used to control the balance between the corrected first feature map and the corrected second feature map in the classification feature map.

[0030] In the aforementioned wind farm access control system, the face recognition module is further configured to: process the classification feature map using the following formula to generate a classification result, wherein the formula is: softmax{(W n B n ):…:(W1,B1)|Project(F)}, where Project(F) represents projecting the classification feature map into a vector, W1 to W n Here are the weight matrices for each fully connected layer, B1 to B... n This represents the bias matrix of each fully connected layer.

[0031] According to another aspect of this application, a wind farm access control method includes:

[0032] Obtain the facial image of the user to be assigned permissions;

[0033] Local binarization and Canny edge detection are performed on the face images of the users whose permissions are to be assigned to obtain a local binary pattern map and a Canny edge detection map.

[0034] The face image, the local binary pattern image, and the Canny edge detection image are arranged into a multi-channel image;

[0035] The multi-channel image is used to obtain a first feature map by using a first convolutional neural network model with a spatial attention mechanism;

[0036] The multi-channel image is used to obtain a second feature map by employing a second convolutional neural network model with a channel attention mechanism.

[0037] Based on the second feature map, the feature values ​​at each position in the first feature map are corrected to obtain the corrected first feature map;

[0038] Based on the first feature map, the feature values ​​at each position in the second feature map are corrected to obtain the corrected second feature map;

[0039] The corrected first feature map and the corrected second feature map are fused to obtain a classification feature map;

[0040] The classification feature map is passed through a classifier to obtain a classification result, which is used to represent the object label to which the face image of the user to be assigned permissions belongs; and

[0041] Based on the classification results, determine whether to assign predetermined permissions to the user.

[0042] In the aforementioned wind farm access control method, the multi-channel image is used to obtain a first feature map by employing a first convolutional neural network model with a spatial attention mechanism. This includes: each layer of the first convolutional neural network model performing the following operations on the input data during forward propagation: performing convolution processing on the input data based on a two-dimensional convolutional kernel to generate a convolutional feature map; performing pooling processing on the convolutional feature map to generate a pooled feature map; performing activation processing on the pooled feature map to generate an activated feature map; performing global average pooling along the channel dimension on the activated feature map to obtain a spatial feature matrix; performing convolution and activation processing on the spatial feature matrix to generate a weight vector; and weighting each feature matrix of the activated feature map with the weight values ​​at each position in the weight vector to obtain a generated feature map; wherein the generated feature map output by the last layer of the first convolutional neural network model is the first feature map.

[0043] In the aforementioned wind farm access control method, the multi-channel image is used to obtain a second feature map by employing a second convolutional neural network model with a channel attention mechanism. This includes: each layer of the second convolutional neural network model performing the following operations on the input data during forward propagation: performing convolution processing on the input data based on a two-dimensional convolutional kernel to generate a convolutional feature map; performing pooling processing on the convolutional feature map to generate a pooled feature map; performing activation processing on the pooled feature map to generate an activated feature map; calculating the global mean of each feature matrix along the channel dimension of the activated feature map to obtain a channel feature vector; calculating the ratio of the feature value at each position in the channel feature vector to the weighted sum of the feature values ​​at all positions in the channel feature vector to obtain a channel-weighted feature vector; and performing a dot product on the feature matrix along the channel dimension of the activated feature map using the feature values ​​at each position of the channel-weighted feature vector as weights to obtain a generated feature map; wherein the generated feature map output by the last layer of the second convolutional neural network model is the second feature map.

[0044] In the above-mentioned wind farm access control method, based on the second feature map, the feature values ​​at each position in the first feature map are corrected to obtain the corrected first feature map, including: based on the second feature map, the feature values ​​at each position in the first feature map are corrected using the following formula to obtain the corrected first feature map;

[0045] The formula is as follows:

[0046]

[0047] Where f 1i and f 2j These are the feature values ​​of the first feature map and the second feature map, respectively, and ρ is a control hyperparameter, and d(f 1i ,f 2j ) represents the distance between eigenvalues.

[0048] In the above-mentioned wind farm access control method, based on the first feature map, the feature values ​​at each position in the second feature map are corrected to obtain the corrected second feature map, including: based on the first feature map, the feature values ​​at each position in the second feature map are corrected using the following formula to obtain the corrected second feature map;

[0049] The formula is as follows:

[0050]

[0051] Where f 1i and f 2jThese are the feature values ​​of the first feature map and the second feature map, respectively, and ρ is a control hyperparameter, and d(f 1i ,f 2j ) represents the distance between eigenvalues.

[0052] In the above-mentioned wind farm access control method, fusing the corrected first feature map and the corrected second feature map to obtain a classification feature map includes: fusing the corrected first feature map and the corrected second feature map to obtain the classification feature map using the following formula;

[0053] The formula is as follows:

[0054] F s =αF1+βF2

[0055] Among them, F s F1 is the corrected first feature map, F2 is the corrected second feature map, "+" indicates that the elements at corresponding positions of the corrected first feature map and the corrected second feature map are added together, and α and β are weighting parameters used to control the balance between the corrected first feature map and the corrected second feature map in the classification feature map.

[0056] In the aforementioned wind farm access control method, the classification feature map is processed by a classifier to obtain a classification result, including: the classifier processes the classification feature map using the following formula to generate the classification result, wherein the formula is: softmax{(W n B n ):…:(W1,B1)|Project(F)}, where Project(F) represents projecting the classification feature map into a vector, W1 to W n Here are the weight matrices for each fully connected layer, B1 to B... n This represents the bias matrix of each fully connected layer.

[0057] Compared with existing technologies, the wind farm access control system and method provided in this application extract features from a multi-channel image fused from the user's face image, local binary pattern image, and Canny edge detection image using a convolutional neural network model with a dual attention mechanism. This focuses on both the feature content and feature location in the image, complementing each other to a certain extent and improving the feature extraction effect of the network. Furthermore, by contrastive search space homogenization, the feature representation of the feature map obtained from the dual attention mechanism convolutional neural network model is constrained to an isotropic and discriminative representation space, thereby enhancing the continuity of the feature distribution of the fused high-dimensional feature representation. In this way, the predetermined permissions assigned to the user can be accurately determined. Attached Figure Description

[0058] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0059] Figure 1 This is a block diagram of a wind farm access control system according to an embodiment of this application.

[0060] Figure 2 This is a flowchart of a wind farm access control method according to an embodiment of this application.

[0061] Figure 3 This is a schematic diagram of the architecture of the wind farm access control method according to an embodiment of this application. Detailed Implementation

[0062] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0063] Scene Overview

[0064] As mentioned earlier, wind power companies are increasingly emphasizing data-driven software management. For example, software can be used to collect and store data generated during the operation of wind turbine systems, and to organize and analyze this data. To facilitate software management, different permissions are typically configured for different accounts on the wind power management software to prevent accidental data deletion.

[0065] However, since account and password information can be leaked, a permission management system is needed to improve security by performing secondary verification of user identity when configuring permissions for corresponding users.

[0066] Currently, deep learning and neural networks have been widely applied in fields such as computer vision, natural language processing, and speech signal processing. Furthermore, deep learning and neural networks have demonstrated near-human or even superior performance in areas such as image classification, object detection, semantic segmentation, and text translation.

[0067] The development of deep learning and neural networks has provided new ideas and solutions for user permission management.

[0068] Accordingly, considering that relying solely on the facial image of the user to be assigned permissions is not accurate enough for user permission management, as it may lead to errors due to interference from the external environment and insufficient detection accuracy, the inventors of this application have proposed merging the local binary pattern image and the Canny edge detection feature image with the RGB image into a 5-channel input to the network to expand the data width at the network input end. This allows the network to learn and express more information, which is beneficial for improving accuracy.

[0069] Specifically, in the technical solution of this application, firstly, the face image of the user to be assigned permissions is acquired by a camera. It should be understood that local binary patterns are a very effective texture description feature in the field of computer vision, possessing advantages such as rotation invariance, translation invariance, and the ability to eliminate illumination changes. Specifically, using a 3×3 window unit, if the surrounding pixel value is greater than the center pixel value, the pixel is marked as 1; otherwise, it is marked as 0. Then, the neighboring pixels are binaryized, and the resulting values ​​are multiplied by the corresponding binary sequence and summed to obtain the LBP value of the center pixel. The Canny operator has three specifications: a low probability of false positives for edge points, the detected edge points being located as close as possible to the center of the true edge, and only one response on each side. Therefore, in the technical solution of this application, the face image of the user to be assigned permissions is further subjected to local binarization processing and Canny edge detection to obtain a local binary pattern map and a Canny edge detection map.

[0070] In this way, the face image, the local binary pattern map, and the Canny edge detection map are arranged into a multi-channel image. The local binary pattern map and the Canny edge detection feature map are then merged with the RGB image into a 5-channel input to the network. This expands the data width at the network input end, allowing the network to learn and express more information, which is beneficial for improving accuracy.

[0071] Considering the correlation between the face image, the local binary pattern map, and the Canny edge detection map for the user's face image to be assigned permissions, a convolutional neural network model with a dual attention mechanism is used to process the multi-channel image to obtain a first feature map and a second feature map, respectively, in order to improve the extraction effect of the multi-channel image. Specifically, the dual attention mechanism here is a spatial attention mechanism and a channel attention mechanism. It should be understood that the channel attention and the spatial attention can focus on the feature content and feature location in the image, respectively, complementing each other to a certain extent and improving the feature extraction effect of the network.

[0072] It is understandable that, since the first convolutional neural network using spatial attention focuses on the spatial scale of the image in relation to the multi-channel image, while the second convolutional neural network using channel attention focuses on the dimensional scale of the image in relation to the multi-channel image, the first and second feature maps exhibit anisotropy due to their distributed representation. In the high-dimensional feature space, their feature representations reside in a narrow subset of the entire high-dimensional feature space. This results in a lack of continuity in the fused high-dimensional feature representation, which degrades the solution space of the classifier and affects the fitting and classification accuracy of the classifier during training and inference.

[0073] Therefore, in the technical solution of this application, the search space is further aligned based on the feature values ​​of the first feature map and the second feature map, that is:

[0074]

[0075]

[0076] Where f 1i and f 2j These are the feature values ​​of the first feature map and the second feature map, respectively, and ρ is a control hyperparameter, for example, initially set to the distance between the first feature map and the second feature map, and d(f 1i ,f 2j () represents the distance between eigenvalues, such as absolute distance.

[0077] In this way, by comparing the search space to be homogenized, the feature representations of the first feature map and the second feature map can be constrained to an isotropic and discriminative representation space, thereby enhancing the continuity of the feature distribution of the fused high-dimensional feature expression, improving the fitting degree of the classifier during training and the classification accuracy during inference, and thus enabling accurate judgment on whether the object label of the face image of the user to be assigned permissions is to assign predetermined permissions to the user.

[0078] Based on this, this application proposes a wind farm access control system, comprising: a face image acquisition module for acquiring face images of users to be assigned access; an image processing module for performing local binarization processing and Canny edge detection on the face images of the users to be assigned access to obtain a local binary pattern map and a Canny edge detection map; an input extension module for arranging the face image, the local binary pattern map, and the Canny edge detection map into a multi-channel image; a first convolutional coding module for applying a first convolutional neural network model with a spatial attention mechanism to the multi-channel image to obtain a first feature map; and a second convolutional coding module for applying a second convolutional neural network model with a channel attention mechanism to the multi-channel image. The system comprises: a network model to obtain a second feature map; a first correction module to correct the feature values ​​at each position in the first feature map based on the second feature map to obtain a corrected first feature map; a second correction module to correct the feature values ​​at each position in the second feature map based on the first feature map to obtain a corrected second feature map; a fusion module to fuse the corrected first feature map and the corrected second feature map to obtain a classification feature map; a face recognition module to pass the classification feature map through a classifier to obtain a classification result, the classification result being used to represent the object label to which the face image of the user to be assigned permissions belongs; and a permission management module to determine whether to assign predetermined permissions to the user based on the classification result.

[0079] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0080] Exemplary System

[0081] Figure 1 The diagram illustrates a block diagram of a wind farm access control system according to an embodiment of this application. Figure 1As shown, the wind farm access control system 200 according to an embodiment of this application includes: a face image acquisition module 210, used to acquire the face image of a user to be assigned access; an image processing module 220, used to perform local binarization processing and Canny edge detection on the face image of the user to be assigned access to obtain a local binary pattern map and a Canny edge detection map; an input extension module 230, used to arrange the face image, the local binary pattern map and the Canny edge detection map into a multi-channel image; a first convolutional encoding module 240, used to obtain a first feature map by using a first convolutional neural network model with spatial attention mechanism on the multi-channel image; and a second convolutional encoding module 250, used to obtain a first feature map by using a second convolutional neural network model with channel attention mechanism on the multi-channel image. A neural network model is used to obtain a second feature map; a first correction module 260 is used to correct the feature values ​​at each position in the first feature map based on the second feature map to obtain a corrected first feature map; a second correction module 270 is used to correct the feature values ​​at each position in the second feature map based on the first feature map to obtain a corrected second feature map; a fusion module 280 is used to fuse the corrected first feature map and the corrected second feature map to obtain a classification feature map; a face recognition module 290 is used to pass the classification feature map through a classifier to obtain a classification result, the classification result being used to represent the object label to which the face image of the user to be assigned permissions belongs; and a permission management module 300 is used to determine whether to assign predetermined permissions to the user based on the classification result.

[0082] Specifically, in this embodiment, the face image acquisition module 210 and the image processing module 220 are used to acquire the face image of the user to be assigned permissions, and perform local binarization processing and Canny edge detection on the face image of the user to be assigned permissions to obtain a local binary pattern map and a Canny edge detection map. It should be understood that, considering user permission management, relying solely on the face image of the user to be assigned permissions is not accurate enough, as it may lead to errors due to interference from the external environment and insufficient detection accuracy. Therefore, in the technical solution of this application, to improve the accuracy of permission management, the local binary pattern map and the Canny edge detection feature map are merged with the RGB image into a 5-channel input to the network, thereby expanding the data width at the network input end, allowing the network to learn and express more information, which is beneficial to improving accuracy.

[0083] Specifically, in the technical solution of this application, firstly, the face image of the user to be assigned permissions is acquired through a camera. It should be understood that local binary patterns are a very effective texture description feature in computer vision, possessing advantages such as rotation invariance, translation invariance, and the ability to eliminate illumination changes. The specific principle is to use a 3×3 window unit; if the surrounding pixel value is greater than the center pixel value, the pixel is marked as 1; otherwise, it is marked as 0. Then, the neighboring pixels are binaryized, and the resulting values ​​are multiplied by the corresponding binary sequence and summed to obtain the LBP value of the center pixel. The Canny operator has three specifications: a low probability of false positives for edge points, the detected edge points being located as close as possible to the center of the true edge, and only one response on each side. Therefore, in the technical solution of this application, the face image of the user to be assigned permissions is further subjected to local binarization processing and Canny edge detection to obtain a local binary pattern map and a Canny edge detection map.

[0084] Specifically, in this embodiment, the input extension module 230 is used to arrange the face image, the local binary pattern map, and the Canny edge detection map into a multi-channel image. That is, in this technical solution, the face image, the local binary pattern map, and the Canny edge detection map are further arranged into a multi-channel image to merge the local binary pattern map and the Canny edge detection feature map with the RGB image into a 5-channel input to the network. This expands the data width of the network input, allowing the network to learn and express more information, which is beneficial for improving accuracy.

[0085] Specifically, in this embodiment, the first convolutional coding module 240 and the second convolutional coding module 250 are used to obtain a first feature map from the multi-channel image by using a first convolutional neural network model with a spatial attention mechanism, and to obtain a second feature map from the multi-channel image by using a second convolutional neural network model with a channel attention mechanism. It should be understood that, considering the correlation between the face image, the local binary pattern map, and the Canny edge detection map for the user's face image to be assigned permissions, in order to improve the extraction effect of the multi-channel image, this application uses a convolutional neural network model with a dual attention mechanism to process the multi-channel image to obtain the first and second feature maps respectively. Specifically, here, the dual attention mechanism is a spatial attention mechanism and a channel attention mechanism. It should be understood that the channel attention and the spatial attention can focus on the feature content and feature location in the image respectively, complementing each other to a certain extent and improving the feature extraction effect of the network.

[0086] More specifically, in this embodiment, the first convolutional coding module is further configured to: perform the following operations on the input data during the forward propagation of each layer of the first convolutional neural network model: perform convolution processing on the input data based on a two-dimensional convolutional kernel to generate a convolutional feature map; perform pooling processing on the convolutional feature map to generate a pooled feature map; perform activation processing on the pooled feature map to generate an activation feature map; perform global average pooling along the channel dimension on the activation feature map to obtain a spatial feature matrix; perform convolution processing and activation processing on the spatial feature matrix to generate a weight vector; and weight each feature matrix of the activation feature map with the weight values ​​at each position in the weight vector to obtain a generated feature map; wherein the generated feature map output by the last layer of the first convolutional neural network model is the first feature map.

[0087] More specifically, in this embodiment, the second convolutional encoding module is further configured to: perform the following operations on the input data during the forward propagation of each layer of the second convolutional neural network model: perform convolution processing on the input data based on a two-dimensional convolutional kernel to generate a convolutional feature map; perform pooling processing on the convolutional feature map to generate a pooled feature map; perform activation processing on the pooled feature map to generate an activation feature map; calculate the global mean of each feature matrix along the channel dimension of the activation feature map to obtain a channel feature vector; calculate the ratio of the feature value at each position in the channel feature vector to the weighted sum of the feature values ​​at all positions in the channel feature vector to obtain a channel-weighted feature vector; and perform dot product on the feature matrix along the channel dimension of the activation feature map using the feature values ​​at each position of the channel-weighted feature vector as weights to obtain a generated feature map; wherein the generated feature map output by the last layer of the second convolutional neural network model is the second feature map.

[0088] Specifically, in this embodiment, the first correction module 260 and the second correction module 270 are used to correct the feature values ​​at each position in the first feature map based on the second feature map to obtain a corrected first feature map, and to correct the feature values ​​at each position in the second feature map based on the first feature map to obtain a corrected second feature map. It should be understood that because the first convolutional neural network using spatial attention focuses on the spatial scale of the image relative to the multi-channel image, while the second convolutional neural network using channel attention focuses on the dimensional scale of the image relative to the multi-channel image, the first and second feature maps exhibit anisotropy due to their distributed representation. In the high-dimensional feature space, their feature representations reside in a narrow subset of the entire high-dimensional feature space. This results in a lack of continuity in the fused high-dimensional feature representation, leading to a degradation of the classifier's solution space and affecting the classifier's fitting and classification accuracy during training and inference. Therefore, in the technical solution of this application, a comparative search space is performed based on the feature values ​​of the first and second feature maps to achieve homogenization.

[0089] It should be understood that by comparing the search space to be homogenized, the feature representations of the first feature map and the second feature map can be constrained to an isotropic and discriminative representation space, thereby enhancing the continuity of the feature distribution of the fused high-dimensional feature expression, improving the fitting degree of the classifier during training and the classification accuracy during inference, and thus enabling accurate judgment on whether the object label to which the face image of the user to be assigned permission belongs to the user to be assigned predetermined permission.

[0090] More specifically, in this embodiment of the application, the first correction module is further configured to: based on the second feature map, correct the feature values ​​at each position in the first feature map using the following formula to obtain the corrected first feature map;

[0091] The formula is as follows:

[0092]

[0093] Where f 1i and f 2j These are the feature values ​​of the first feature map and the second feature map, respectively, and ρ is a control hyperparameter, for example, initially set to the distance between the first feature map and the second feature map, and d(f 1i ,f 2j () represents the distance between eigenvalues, such as absolute distance.

[0094] More specifically, in this embodiment of the application, the second correction module is further configured to: based on the first feature map, correct the feature values ​​at each position in the second feature map using the following formula to obtain the corrected second feature map;

[0095] The formula is as follows:

[0096]

[0097] Where f 1i and f 2j These are the feature values ​​of the first feature map and the second feature map, respectively, and ρ is a control hyperparameter, for example, initially set to the distance between the first feature map and the second feature map, and d(f 1i ,f 2j () represents the distance between eigenvalues, such as absolute distance.

[0098] Specifically, in this embodiment, the fusion module 280 is used to fuse the corrected first feature map and the corrected second feature map to obtain a classification feature map. That is, in the technical solution of this application, after obtaining the corrected first feature map and the corrected second feature map, the feature information of the two is further fused to obtain a classification feature map.

[0099] More specifically, in the embodiments of this application, the corrected first feature map and the corrected second feature map are fused using the following formula to obtain the classification feature map;

[0100] The formula is as follows:

[0101] F s =αF1+βF2

[0102] Among them, F s F1 is the corrected first feature map, F2 is the corrected second feature map, "+" indicates that the elements at corresponding positions of the corrected first feature map and the corrected second feature map are added together, and α and β are weighting parameters used to control the balance between the corrected first feature map and the corrected second feature map in the classification feature map.

[0103] Specifically, in this embodiment, the face recognition module 290 and the permission management module 300 are used to process the classification feature map through a classifier to obtain a classification result. The classification result represents the object label to which the face image of the user to be assigned permissions belongs, and based on the classification result, determine whether to assign predetermined permissions to the user. That is, in the technical solution of this application, after obtaining the classification feature map, the classification feature map is processed through a classifier to obtain a classification result representing the object label to which the face image of the user to be assigned permissions belongs. Then, based on the classification result, an accurate determination is made as to whether the object label to which the face image of the user to be assigned permissions belongs warrants the assignment of predetermined permissions to the user.

[0104] More specifically, in embodiments of this application, the face recognition module is further configured to: process the classification feature map using the following formula to generate a classification result, wherein the formula is: softmax{(W n B n ):…:(W1,B1)|Project(F)}, where Project(F) represents projecting the classification feature map into a vector, W1 to W n Here are the weight matrices for each fully connected layer, B1 to B... n This represents the bias matrix of each fully connected layer.

[0105] In summary, the wind farm access control system 200 described in this application embodiment is explained. It uses a convolutional neural network model with a dual attention mechanism to extract features from a multi-channel image fused from the user's face image, local binary pattern image, and Canny edge detection image. This focuses on both the feature content and feature location within the image, complementing each other to a certain extent and improving the network's feature extraction performance. Furthermore, by contrastive search space homogenization, the feature representation of the feature map obtained from the dual attention mechanism convolutional neural network model is constrained to an isotropic and discriminative representation space, thereby enhancing the continuity of the feature distribution in the fused high-dimensional feature representation. In this way, the predetermined permissions assigned to the user can be accurately determined.

[0106] As described above, the wind farm access control system 200 according to the embodiments of this application can be implemented in various terminal devices, such as servers for wind farm access control algorithms. In one example, the wind farm access control system 200 according to the embodiments of this application can be integrated into a terminal device as a software module and / or a hardware module. For example, the wind farm access control system 200 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the wind farm access control system 200 can also be one of many hardware modules of the terminal device.

[0107] Alternatively, in another example, the wind farm access control system 200 and the terminal device can also be separate devices, and the wind farm access control system 200 can connect to the terminal device via wired and / or wireless networks and transmit interactive information in accordance with an agreed data format.

[0108] Exemplary methods

[0109] Figure 2 The flowchart illustrates the wind farm access control method. For example... Figure 2 As shown, the wind farm access control method according to an embodiment of this application includes the following steps: S110, acquiring a face image of a user to be assigned access; S120, performing local binarization and Canny edge detection on the face image of the user to be assigned access to obtain a local binary pattern map and a Canny edge detection map; S130, arranging the face image, the local binary pattern map, and the Canny edge detection map into a multi-channel image; S140, using a first convolutional neural network model with a spatial attention mechanism to obtain a first feature map from the multi-channel image; S150, using a second convolutional neural network model with a channel attention mechanism to obtain a first feature map from the multi-channel image. S160: Based on the second feature map, the feature values ​​at each position in the first feature map are corrected to obtain a corrected first feature map; S170: Based on the first feature map, the feature values ​​at each position in the second feature map are corrected to obtain a corrected second feature map; S180: The corrected first feature map and the corrected second feature map are fused to obtain a classification feature map; S190: The classification feature map is passed through a classifier to obtain a classification result, the classification result being used to represent the object label to which the face image of the user to be assigned permissions belongs; and S200: Based on the classification result, it is determined whether to assign predetermined permissions to the user.

[0110] Figure 3 The diagram illustrates the architecture of a wind farm access control method according to an embodiment of this application. Figure 3As shown, in the network architecture of the wind farm access control method, firstly, the facial images of the users whose permissions are to be assigned (e.g., as shown in the image) are processed. Figure 3 The P1 shown is subjected to local binarization and Canny edge detection to obtain a local binary pattern map (e.g., as shown in the figure). Figure 3 The Q1 shown is intended) and the Canny edge detection map (e.g., as shown) Figure 3 (as shown in Q2); then, the face image, the local binary pattern map, and the Canny edge detection map are arranged into a multi-channel image (e.g., as shown in Q2); Figure 3 (as shown in IN); then, the multi-channel image is processed using a first convolutional neural network model with a spatial attention mechanism (e.g., such as...). Figure 3 The CNN1 shown is used to obtain the first feature map (e.g., as shown in the diagram). Figure 3 The F1 signal is shown in the diagram; then, the multi-channel image is processed using a second convolutional neural network model with a channel attention mechanism (e.g., such as...). Figure 3 The CNN2 shown is used to obtain the second feature map (e.g., as shown in the diagram). Figure 3 (as shown in F2); then, based on the second feature map, the feature values ​​at each position in the first feature map are corrected to obtain the corrected first feature map (e.g., as shown in F2); Figure 3 (as shown in FC1); then, based on the first feature map, the feature values ​​at each position in the second feature map are corrected to obtain the corrected second feature map (e.g., as shown in FC1); Figure 3 The intended FC2); then, the corrected first feature map and the corrected second feature map are fused to obtain a classification feature map (e.g., as shown in the image). Figure 3 The FC shown in the diagram); then, the classification feature map is passed through a classifier (e.g., such as...). Figure 3 The classifier shown is used to obtain a classification result, which is used to represent the object label to which the face image of the user to be assigned permissions belongs; and finally, based on the classification result, it is determined whether to assign the predetermined permissions to the user.

[0111] More specifically, in steps S110 and S120, the face image of the user to be assigned permissions is acquired, and local binarization and Canny edge detection are performed on the face image to obtain a local binary pattern map and a Canny edge detection map. It should be understood that, considering user permission management, relying solely on the face image of the user to be assigned permissions is not accurate enough, as it may lead to errors due to interference from the external environment and insufficient detection precision. Therefore, in the technical solution of this application, to improve the accuracy of permission management, the local binary pattern map and the Canny edge detection feature map are merged with the RGB image into a 5-channel input to the network, thereby expanding the data width at the network input end, allowing the network to learn and express more information, which is beneficial for improving accuracy.

[0112] Specifically, in the technical solution of this application, firstly, the face image of the user to be assigned permissions is acquired through a camera. It should be understood that local binary patterns are a very effective texture description feature in computer vision, possessing advantages such as rotation invariance, translation invariance, and the ability to eliminate illumination changes. The specific principle is to use a 3×3 window unit; if the surrounding pixel value is greater than the center pixel value, the pixel is marked as 1; otherwise, it is marked as 0. Then, the neighboring pixels are binaryized, and the resulting values ​​are multiplied by the corresponding binary sequence and summed to obtain the LBP value of the center pixel. The Canny operator has three specifications: a low probability of false positives for edge points, the detected edge points being located as close as possible to the center of the true edge, and only one response on each side. Therefore, in the technical solution of this application, the face image of the user to be assigned permissions is further subjected to local binarization processing and Canny edge detection to obtain a local binary pattern map and a Canny edge detection map.

[0113] More specifically, in step S130, the face image, the local binary pattern map, and the Canny edge detection map are arranged into a multi-channel image. That is, in the technical solution of this application, the face image, the local binary pattern map, and the Canny edge detection map are further arranged into a multi-channel image to merge the local binary pattern map and the Canny edge detection feature map with the RGB image into a 5-channel input to the network. This expands the data width at the network input, allowing the network to learn and express more information, which is beneficial for improving accuracy.

[0114] More specifically, in steps S140 and S150, the multi-channel image is processed using a first convolutional neural network model with a spatial attention mechanism to obtain a first feature map, and the multi-channel image is processed using a second convolutional neural network model with a channel attention mechanism to obtain a second feature map. It should be understood that, considering the correlation between the face image, the local binary pattern map, and the Canny edge detection map for the user's face image to be assigned permissions, in order to improve the extraction effect of the multi-channel image, the technical solution of this application uses a convolutional neural network model with a dual attention mechanism to process the multi-channel image to obtain the first and second feature maps respectively. In particular, here, the dual attention mechanism is a spatial attention mechanism and a channel attention mechanism. It should be understood that the channel attention and the spatial attention can respectively focus on the feature content and feature location in the image, complementing each other to a certain extent and improving the feature extraction effect of the network.

[0115] More specifically, in steps S160 and S170, based on the second feature map, the feature values ​​at each position in the first feature map are corrected to obtain a corrected first feature map, and based on the first feature map, the feature values ​​at each position in the second feature map are corrected to obtain a corrected second feature map. It should be understood that because the first convolutional neural network using spatial attention focuses on the spatial scale of the image relative to the multi-channel image, while the second convolutional neural network using channel attention focuses on the dimensional scale of the images relative to the multi-channel image, the first and second feature maps exhibit anisotropy due to their distributed representation. In the high-dimensional feature space, their feature representations reside in a narrow subset of the entire high-dimensional feature space. This results in a lack of continuity in the fused high-dimensional feature representation, leading to a degradation of the classifier's solution space and affecting the classifier's fitting during training and classification accuracy during inference. Therefore, in the technical solution of this application, a comparative search for spatial homogenization is further performed based on the feature values ​​of the first and second feature maps.

[0116] In this way, by comparing the search space to be homogenized, the feature representations of the first feature map and the second feature map can be constrained to an isotropic and discriminative representation space, thereby enhancing the continuity of the feature distribution of the fused high-dimensional feature expression, improving the fitting degree of the classifier during training and the classification accuracy during inference, and thus enabling accurate judgment on whether the object label of the face image of the user to be assigned permissions is to assign predetermined permissions to the user.

[0117] More specifically, in steps S180, S190, and S200, the corrected first feature map and the corrected second feature map are fused to obtain a classification feature map. The classification feature map is then passed through a classifier to obtain a classification result. This classification result represents the object label to which the face image of the user to be assigned permissions belongs. Based on the classification result, it is determined whether to assign predetermined permissions to the user. That is, in the technical solution of this application, after obtaining the corrected first feature map and the corrected second feature map, the feature information of these two is further fused to obtain a classification feature map. Then, the classification feature map is passed through a classifier for classification processing to obtain a classification result representing the object label to which the face image of the user to be assigned permissions belongs. Furthermore, based on the classification result, an accurate determination is made as to whether the object label to which the face image of the user to be assigned permissions belongs warrants the assignment of predetermined permissions to the user.

[0118] In summary, the wind farm access control method described in this application is explained. It uses a convolutional neural network model with a dual attention mechanism to extract features from a multi-channel image fused from the user's face image, local binary pattern image, and Canny edge detection image. This focuses on both the feature content and feature location within the image, complementing each other to a certain extent and improving the network's feature extraction performance. Furthermore, by contrastive search space homogenization, the feature representation of the feature map obtained from the dual attention mechanism convolutional neural network model is constrained to an isotropic and discriminative representation space, thereby enhancing the continuity of the feature distribution in the fused high-dimensional feature representation. This allows for accurate determination of the predetermined permissions assigned to the user.

[0119] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0120] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0121] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0122] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0123] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A wind farm access control system, characterized in that, include: The face image acquisition module is used to acquire the face images of users to be assigned permissions; The image processing module is used to perform local binarization processing and Canny edge detection on the face image of the user to be assigned permissions to obtain a local binary pattern map and a Canny edge detection map. The input extension module is used to arrange the face image, the local binary pattern map, and the Canny edge detection map into a multi-channel image; The first convolutional coding module is used to obtain a first feature map from the multi-channel image by using a first convolutional neural network model with a spatial attention mechanism. The second convolutional coding module is used to obtain a second feature map from the multi-channel image by using a second convolutional neural network model with a channel attention mechanism. The first correction module is used to: based on the second feature map, correct the feature values ​​at each position in the first feature map using the following formula to obtain the corrected first feature map; The formula is as follows: in and These are the feature values ​​of the first feature map and the second feature map, respectively. To control hyperparameters, and Indicates the distance between eigenvalues; The second correction module is used to: based on the first feature map, correct the feature values ​​at each position in the second feature map using the following formula to obtain the corrected second feature map; The formula is as follows: in and These are the feature values ​​of the first feature map and the second feature map, respectively. To control hyperparameters, and Indicates the distance between eigenvalues; A fusion module is used to fuse the corrected first feature map and the corrected second feature map to obtain a classification feature map; A face recognition module is used to pass the classification feature map through a classifier to obtain a classification result, wherein the classification result is used to represent the object label to which the face image of the user to be assigned permissions belongs; and The permission management module is used to determine whether to assign predetermined permissions to the user based on the classification results.

2. The wind farm access control system according to claim 1, characterized in that, The first convolutional coding module is further configured to: process the input data in the forward propagation of each layer of the first convolutional neural network model: The input data is subjected to convolution processing based on two-dimensional convolution kernels to generate a convolutional feature map; The convolutional feature map is subjected to pooling to generate a pooled feature map; The pooled feature map is activated to generate an activated feature map; The activated feature map is subjected to global average pooling along the channel dimension to obtain a spatial feature matrix; The spatial feature matrix is ​​subjected to convolution and activation processing to generate a weight vector; and The feature matrices of the activated feature map are weighted by the weight values ​​at each position in the weight vector to obtain the generated feature map. The generated feature map output from the last layer of the first convolutional neural network model is the first feature map.

3. The wind farm access control system according to claim 2, characterized in that, The second convolutional coding module is further configured to: process the input data in the forward propagation of each layer of the second convolutional neural network model as follows: The input data is subjected to convolution processing based on two-dimensional convolution kernels to generate a convolutional feature map; The convolutional feature map is subjected to pooling to generate a pooled feature map; The pooled feature map is activated to generate an activated feature map; The global mean of each feature matrix along the channel dimension of the activated feature map is calculated to obtain the channel feature vector; Calculate the ratio of the feature value at each position in the channel feature vector to the weighted sum of the feature values ​​at all positions in the channel feature vector to obtain the channel weighted feature vector; and The feature map is generated by multiplying the feature matrix along the channel dimension of the activated feature map by the feature values ​​at each position of the channel-weighted feature vector as weights. The generated feature map output from the last layer of the second convolutional neural network model is the second feature map.

4. The wind farm access control system according to claim 3, characterized in that, The fusion module is further configured to: fuse the corrected first feature map and the corrected second feature map using the following formula to obtain the classification feature map; The formula is as follows: in, For the classification feature map, This is the corrected first feature map. For the corrected second feature map, "" indicates that the elements at corresponding positions in the corrected first feature map and the corrected second feature map are added together. The weighting parameter is used to control the balance between the corrected first feature map and the corrected second feature map in the classification feature map.

5. The wind farm access control system according to claim 4, characterized in that, The face recognition module is further configured to: process the classification feature map using the classifier to generate a classification result using the following formula, wherein the formula is: ,in This indicates that the classification feature map is projected into a vector. to Here are the weight matrices for each fully connected layer. to This represents the bias matrix of each fully connected layer.

6. A wind farm access control method, characterized in that, include: Obtain the facial image of the user to be assigned permissions; Local binarization and Canny edge detection are performed on the face images of the users whose permissions are to be assigned to obtain a local binary pattern map and a Canny edge detection map. The face image, the local binary pattern image, and the Canny edge detection image are arranged into a multi-channel image; The multi-channel image is used to obtain a first feature map by using a first convolutional neural network model with a spatial attention mechanism; The multi-channel image is used to obtain a second feature map by employing a second convolutional neural network model with a channel attention mechanism. Based on the second feature map, the feature values ​​at each position in the first feature map are corrected to obtain the corrected first feature map, including: Based on the second feature map, the feature values ​​at each position in the first feature map are corrected using the following formula to obtain the corrected first feature map; The formula is as follows: in and These are the feature values ​​of the first feature map and the second feature map, respectively. To control hyperparameters, and Indicates the distance between eigenvalues; Based on the first feature map, the feature values ​​at each position in the second feature map are corrected to obtain the corrected second feature map, including: based on the first feature map, the feature values ​​at each position in the second feature map are corrected using the following formula to obtain the corrected second feature map; The formula is as follows: in and These are the feature values ​​of the first feature map and the second feature map, respectively. To control hyperparameters, and Indicates the distance between eigenvalues; The corrected first feature map and the corrected second feature map are fused to obtain a classification feature map; The classification feature map is passed through a classifier to obtain a classification result, which is used to represent the object label to which the face image of the user to be assigned permissions belongs; and Based on the classification results, determine whether to assign predetermined permissions to the user.

7. The wind farm access control method according to claim 6, characterized in that, The step of obtaining a first feature map from the multi-channel image using a first convolutional neural network model with a spatial attention mechanism includes: each layer of the first convolutional neural network model processes the input data during the forward propagation of the layer. The input data is subjected to convolution processing based on two-dimensional convolution kernels to generate a convolutional feature map; The convolutional feature map is subjected to pooling to generate a pooled feature map; The pooled feature map is activated to generate an activated feature map; The activated feature map is subjected to global average pooling along the channel dimension to obtain a spatial feature matrix; The spatial feature matrix is ​​subjected to convolution and activation processing to generate a weight vector; and The feature matrices of the activated feature map are weighted by the weight values ​​at each position in the weight vector to obtain the generated feature map. The generated feature map output from the last layer of the first convolutional neural network model is the first feature map.

Images