A method, device and storage medium for detecting vegetation intrusion on transmission lines based on improved B-CNN algorithm

Through the improved B-CNN algorithm and GLCM feature extraction, the problem of low accuracy and efficiency of vegetation intrusion detection in the prior art is solved, and efficient classification and labeling of vegetation intrusions in transmission lines is realized, and the detection accuracy is improved.

CN114596504BActive Publication Date: 2024-11-05STATE GRID JIANGSU ELECTRIC POWER CO LTD RESEARCH INSTITUTE +2
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202210214637.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-04
Publication Date
2024-11-05
Estimated Expiration
2042-03-04

AI Technical Summary

Technical Problem

Existing vegetation intrusion detection methods rely on a large amount of manual labeling operations and lack of available stereoscopic image data at the appropriate resolution, resulting in inaccuracy and inefficiency of detection.

Method used

The improved B-CNN algorithm is used to pre-process the transmission line environment image to be detected, and GLCM features are extracted, and vegetation invasive areas are classified using the B-CNN network to simplify sample marking operations and improve detection accuracy.

Benefits of technology

By utilizing the fine-grained classification capabilities and GLCM feature extraction of the B-CNN network, the accuracy of vegetation intrusion detection is significantly improved, manual labeling operations are reduced, and detection efficiency is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114596504B_ABST
    Figure CN114596504B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, device and storage medium for detecting vegetation intrusion on power lines based on an improved B‑CNN algorithm, and belongs to the technical field of image detection of electric power facilities. The method comprises: obtaining an environmental image of a power line to be detected; preprocessing the environmental image of the power line; extracting GLCM features from the preprocessed image; inputting the preprocessed environmental image of the power line and the extracted GLCM features of the image into a pre-trained power line vegetation intrusion detection model to obtain a classification result output by the model; and outputting images of vegetation intrusion areas and power line areas respectively marked according to the classification result of the power line vegetation intrusion detection model. The present invention utilizes the texture characteristics of visible images, and on the basis of extracting image texture features using the GLCM algorithm, uses the B‑CNN algorithm to detect and classify vegetation areas outside power lines, which can not only simplify the sample marking operation, but also improve the detection accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image detection technology, and in particular to a method, a device and a storage medium for detecting vegetation intrusion on a power transmission line based on an improved B-CNN algorithm. Background Art

[0002] The process of delivering electricity to users consists of three steps: generation, transmission, and distribution. Transmission lines are the backbone infrastructure of the transmission process. There are many environmental factors that can affect the transmission process, such as forest fires, storms, and vegetation degradation. A major challenge in the installation, operation, and maintenance of transmission lines in areas with high vegetation density is intrusion. When branches come into contact with transmission lines, overgrowth can cause flashes. There are currently many different methods available for monitoring vegetation encroachment, such as regular visits by inspectors to patrol designated areas, but this requires a large number of inspectors to conduct inspections, and although the data collection process is effective, it is very expensive relative to the entire area. High-resolution image acquisition methods can significantly overcome the high cost limitations and can provide extensive geographic information.

[0003] The vegetation index method is a method that uses visible images to monitor vegetation activity along a strong light path. It detects plants of different densities based on the calculation of vegetation indices, including the calculation of the Normalized Difference Vegetation Index (NDVI), the Enhanced Vegetation Index (EVI), and the Outermost Vegetation Index (ARVI). However, this type of detection depends on the availability of data at the target location.

[0004] The Digital Assessment Model approach creates a dual elevation model (DEM) map from visible images using images with different viewing angles. This approach is useful for estimating the height of objects around transmission line corridors, but the lack of available stereo image data of appropriate resolution limits this approach. Summary of the invention

[0005] The purpose of the present invention is to provide a method, device and storage medium for detecting vegetation intrusion on power lines based on an improved B-CNN algorithm, which utilizes the texture characteristics of visible images and uses the B-CNN algorithm to classify vegetation areas outside power lines, simplify sample marking operations, and improve detection accuracy. The technical solution adopted by the present invention is as follows.

[0006] On the one hand, the present invention provides a method for detecting vegetation intrusion on power lines based on an improved B-CNN algorithm, comprising:

[0007] Acquire an environmental image of a transmission line to be detected;

[0008] Preprocessing the transmission line environment image;

[0009] Extract GLCM features from the preprocessed image;

[0010] The preprocessed transmission line environment image and the extracted image GLCM features are input into the pre-trained transmission line vegetation intrusion detection model to obtain the classification result output by the model;

[0011] According to the classification results of the transmission line vegetation intrusion detection model, images of vegetation intrusion areas and transmission line areas are respectively marked.

[0012] Optionally, the preprocessing of the power line environment image further includes dividing the image into blocks, each image block has n×n pixels, and the power line environment image input to the power line vegetation intrusion detection model is the image after block processing. The process of dividing the image into small blocks is intended to reduce processing time. However, the detection time is proportional to the number of image blocks, so the more images there are, the longer the execution time required. The small blocks have a reasonable size.

[0013] Optionally, the transmission line vegetation intrusion detection model adopts a B-CNN network, which includes an input layer, a convolution layer, a bilinear pooling layer and a Softmax normalization layer arranged in sequence, the convolution layer includes a convolution network A and a convolution network B respectively connected to the input layer, and the outputs of the two are subjected to outer product processing and pooling processing by the bilinear pooling layer to obtain a bilinear vector;

[0014] The B-CNN network is expressed as the following formula:

[0015] F=(f A ,f B ,P,C)

[0016] Where F represents the output classification of the B-CNN network; f A and f B Respectively represent the feature extraction functions of convolutional network A and convolutional network B, which are the mapping process functions f:L×I→R c×D , which means that the image I and the location area L are mapped into a vector of c×D dimensions through the set mapping relationship; P represents the pooling function, and C represents the bidirectional classification recognition vector through the Softmax normalization layer;

[0017] The outer product processing formula is expressed as:

[0018] bilinear=f A (l,I) T f B (l,I)

[0019] where f A (l,I) and f B (l, I) are the features output by convolutional network A and convolutional network B at position l of image I respectively;

[0020] The pooling function P is expressed as:

[0021]

[0022] Where φ(I) is the pooling result of the convolution output of image I after the outer product processing. It can be seen that the role of the pooling function P is to integrate the features at different positions to obtain a global image.

[0023] Convolutional network A and convolutional network B can be two symmetric networks or two asymmetric networks. For example, VGG-16 and VGG-19 networks can be used respectively. Using two different convolutional neural networks can obtain better performance than two CNNs of the same model.

[0024] Optionally, in the transmission line vegetation intrusion detection model B-CNN network, the output ends of the convolutional network A and the convolutional network B are respectively connected to a channel attention module and a spatial attention module;

[0025] The feature calculation formula of the channel attention module is:

[0026]

[0027] The feature calculation formula of the spatial attention module is:

[0028]

[0029] In the formula, F represents the features output by the convolutional network, F′ represents the features output by the channel attention module, and F″ represents the features output by the spatial attention module; and They represent the results of average pooling and maximum pooling of feature F, respectively. W0 and W1 represent the weights of different layers in the multilayer perceptron MLP. σ represents the sigmoid function. represents element-by-element multiplication, f 7×7 represents a convolution operation with a filter size of 7*7; AvgPool(·) represents the average pooling function, and MaxPool(·) represents the maximum pooling function.

[0030] Optionally, extracting GLCM features from the preprocessed image includes:

[0031] The preprocessed transmission line environment image is linearly converted into a one-dimensional vector and the statistical moment is calculated;

[0032] Calculate the angular second moment, entropy ENT, contrast Con, inverse difference matrix IDM and correlation Correlation;

[0033] The statistical moments include:

[0034] Mean μ, standard deviation σ, variance σ 2 , third-order center distance μ3, skewness And the kurtosis moment Kurt, calculated as:

[0035]

[0036]

[0037]

[0038] μ3=E(p i -μ) 3

[0039]

[0040]

[0041] In the above formula, E represents the averaging function;

[0042] The calculation formulas for the angular second moment, entropy ENT, contrast Con, inverse difference matrix IDM and correlation Correlation are:

[0043]

[0044]

[0045]

[0046]

[0047]

[0048] In the formula, μ x represents the average value of the elements in the i-th row of the GLCM feature matrix, σ x represents the standard deviation of the elements in the i-th row, μ y represents the average value of the elements in the jth column, σ y represents the standard deviation of the elements in the jth column.

[0049] Optionally, the transmission line vegetation intrusion detection model training method comprises:

[0050] Obtain image samples from different transmission line environments and their vegetation invasion area annotation information;

[0051] Divide the image samples into training set samples and test set samples;

[0052] Preprocess the image samples to determine the annotation information of the vegetation invasion area and the transmission line area;

[0053] Perform GLCM feature extraction on the preprocessed image samples;

[0054] The preprocessed training set samples, sample annotation information, and extracted sample GLCM information are input into the pre-built B-CNN network to train the B-CNN network;

[0055] The classification accuracy of the trained B-CNN network is tested through the test set samples. If the classification accuracy reaches the preset accuracy requirement, the training is stopped. Otherwise, the B-CNN network training is continued through the training set samples.

[0056] Optionally, the image samples have multiple resolutions and ratios, including images containing high-density vegetation areas and images containing low-density vegetation areas.

[0057] In a second aspect, the present invention provides a transmission line vegetation intrusion detection device based on an improved B-CNN algorithm, comprising:

[0058] An image acquisition module is configured to acquire an image of a power transmission line environment to be detected;

[0059] A preprocessing module is configured to preprocess the transmission line environment image;

[0060] A GLCM feature extraction module is configured to extract GLCM features from the preprocessed image;

[0061] The vegetation intrusion detection module is configured to input the pre-processed transmission line environment image and the extracted image GLCM features into a pre-trained transmission line vegetation intrusion detection model to obtain a classification result output by the model;

[0062] And, the detection result output module is configured to output images in which the vegetation intrusion area and the transmission line area are marked respectively according to the classification result of the transmission line vegetation intrusion detection model.

[0063] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for detecting vegetation intrusion on a transmission line based on the improved B-CNN algorithm as described in the first aspect is implemented.

[0064] Beneficial Effects

[0065] The present invention utilizes the texture characteristics of visible images. First, the GLCM algorithm is used to extract the GLCM features of the image. Then, based on the image itself and the extracted GLCM features, the B-CNN algorithm is used to classify the vegetation area outside the power lines. The advantages of the B-CNN network in fine-grained classification are fully utilized, and a large number of manual labeling operations in traditional methods are avoided. At the same time, the application of the GLCM algorithm significantly increases the effectiveness of feature extraction, which can greatly improve the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 Shown is a schematic diagram of an application scenario of the present invention;

[0067] Figure 2 Shown is a schematic diagram of the B-CNN network architecture used in the method of the present invention;

[0068] Figure 3 Shown is a schematic diagram of the effect of using the method of the present invention to detect vegetation invasion (actually output color distinction). DETAILED DESCRIPTION

[0069] The invention is further described below with reference to the accompanying drawings and specific embodiments.

[0070] Example 1

[0071] This embodiment introduces a method for detecting vegetation intrusion on power lines based on an improved B-CNN algorithm, including:

[0072] Acquire an environmental image of a transmission line to be detected;

[0073] Preprocessing the transmission line environment image;

[0074] Extract GLCM features from the preprocessed image;

[0075] The preprocessed transmission line environment image and the extracted image GLCM features are input into the pre-trained transmission line vegetation intrusion detection model to obtain the classification result output by the model;

[0076] According to the classification results of the transmission line vegetation intrusion detection model, images of vegetation intrusion areas and transmission line areas are respectively marked.

[0077] The method of this embodiment mainly involves the following contents.

[0078] 1. Construction of Transmission Line Vegetation Intrusion Detection Model

[0079] In this embodiment, the transmission line vegetation intrusion detection model adopts an improved B-CNN network, such as Figure 2As shown in the figure, the network includes an input layer, a convolution layer, a bilinear pooling layer and a Softmax normalization layer arranged in sequence. The convolution layer includes a convolution network A and a convolution network B respectively connected to the input layer. The outputs of the two networks are respectively connected to a channel attention module and a spatial attention module. The features of the two channel outputs are processed by the bilinear pooling layer for outer product and pooling to obtain a bilinear vector. Finally, the softmax layer is used to realize classification, that is, target recognition.

[0080] The B-CNN network without the attention mechanism is expressed as follows:

[0081] F=(f A ,f B ,P,C)

[0082] Where F represents the output classification of the B-CNN network; f A and f B Respectively represent the feature extraction functions of convolutional network A and convolutional network B, which are the mapping process functions f:L×I→R c×D , which means that the image I and the location area L are mapped into a vector of c×D dimensions through the set mapping relationship; P represents the pooling function, and C represents the bidirectional classification recognition vector through the Softmax normalization layer;

[0083] The outer product processing formula is expressed as:

[0084] bilinear=f A (l,I) T f B (l,I)

[0085] where f A (l,I) and f B (l, I) are the features output by convolutional network A and convolutional network B at position l of image I respectively;

[0086] The pooling function P is expressed as:

[0087]

[0088] Where φ(I) is the pooling result of the convolution output of image I after the outer product processing. It can be seen that the role of the pooling function P is to integrate the features at different positions to obtain a global image.

[0089] Convolutional network A and convolutional network B can be two symmetric networks or two asymmetric networks. For example, VGG-16 and VGG-19 networks can be used respectively. Using two different convolutional neural networks can obtain better performance than two CNNs of the same model.

[0090] The transmission line vegetation intrusion detection model of this embodiment incorporates an attention mechanism.

[0091] The attention mechanism is a visual information processing mechanism unique to humans. It allows the network to dynamically learn significant features according to needs, and can effectively improve the network's ability to learn key information. In the attention mechanism, there are attention feature calculation methods that focus on channel dimensions and spatial dimensions.

[0092] refer to Figure 2 As shown, in this embodiment, the main idea of ​​the channel attention module is to generate a channel attention feature map by changing the weights, aggregate the spatial information of the feature map using average pooling and maximum pooling operations, generate two descriptors, and forward the two descriptors to a multi-layer perceptron (MLP) with a hidden layer, and then obtain the channel attention feature through the sigmoid function, and obtain the final attention feature map F′ by multiplying the feature map F output by the convolutional network and the channel attention feature in the corresponding element-by-element manner;

[0093] The feature calculation formula of the channel attention module is:

[0094]

[0095] The spatial attention module focuses on the spatial relationship information between features. In order to calculate the spatial attention, it is necessary to apply average pooling and maximum pooling operations along the channel dimension and connect them to generate a valid feature descriptor. Then, it is output through a standard convolutional layer and the spatial attention map is obtained through the sigmoid function. Finally, the spatial attention map and the channel attention feature map are multiplied by the corresponding elements to obtain the feature map F″ that combines the channel and spatial attention.

[0096] The feature calculation formula of the spatial attention module is:

[0097]

[0098] In the above formula, F represents the feature output by the convolutional network, F′ represents the feature output by the channel attention module, and F″ represents the feature output by the spatial attention module; and They represent the results of average pooling and maximum pooling of feature F, respectively. W0 and W1 represent the weights of different layers in the multilayer perceptron MLP. σ represents the sigmoid function. represents element-by-element multiplication, f 7×7 represents a convolution operation with a filter size of 7*7; AvgPool(·) represents the average pooling function, and MaxPool(·) represents the maximum pooling function.

[0099] 2. Training of Transmission Line Vegetation Intrusion Detection Model

[0100] 2.1 Sample data collection

[0101] In order to train the above improved B-CNN network for power line vegetation intrusion recognition, this embodiment first manually collects training data from image sources in different power line environments. The collected images have different resolutions and scales. The reason for collecting data from different sources is to understand the behavior of the extracted features under different conditions, which helps to find a general vegetation density classification solution.

[0102] In this embodiment, among the 12344 training samples collected, 6172 samples contain high-density vegetation areas, and the remaining samples contain low-density vegetation areas. The data set is divided into an 80% and 20% ratio, where 80% of the data set is used for training and 20% is used for testing. Vegetation invasion area information and transmission line area information are marked in the sample images.

[0103] 2.2 Image Segmentation

[0104] In order to reduce the processing time of the neural network when recognizing an image, this embodiment first divides the image into blocks.

[0105] The input image is divided into small blocks of n×n pixels, where the total number of small blocks L represents the total number of segments. If the input image is square, the width w and height h are equal. The total number of small blocks can be calculated as The offset between blocks is controlled by s. If the input image is not a square, the number of small blocks can be used calculate.

[0106] The detection time is proportional to the number of sliced ​​patch images, so more images will take longer to execute, and the tiles should be of reasonable size.

[0107] 2.3 GLCM feature extraction

[0108] In order to obtain more effective features and more accurate recognition results in subsequent network recognition, this embodiment extracts GLCM features from the sample image before inputting it into the neural network.

[0109] The texture-based features are composed of GLCM (Grey-Level Co-occurrence Matrix) features, which are: angular second moment, entropy, contrast, inverse difference matrix and correlation. After the patch image is linearly converted into a one-dimensional vector, the statistical moments are calculated: mean μ, standard deviation σ, variance σ 2 , third-order center distance μ3, skewness and kurtosis moment Kurt.

[0110]

[0111]

[0112]

[0113] μ3=E(p i -μ) 3

[0114]

[0115]

[0116] Among them, N is the length of a one-dimensional vector representing the number of gray levels, p i The vector element represents the gray value corresponding to gray level i. E represents the averaging function;

[0117] GLCM describes the values ​​that frequently appear in a specific direction. This method gives an impression of the relationship between adjacent pixel values. We extract zero degrees as a fixed angle of GLCM. The elements of GLCM are recorded as p(i,j), which represents the joint conditional probability density p(i,j|d,θ) between gray levels, that is, the probability of gray level j (column) appearing with i as the starting point (row) given the spatial distance d and direction θ. The order M of GLCM is equal to the number of gray levels, so N = 2 M . GLCM features are extracted from GLCM. In this work, the angular second moment ASM, entropy ENT, contrast Con, inverse difference matrix IDM and correlation Correlation are extracted:

[0118]

[0119]

[0120]

[0121]

[0122]

[0123] In the formula, μ x represents the average value of the elements in the i-th row of the GLCM feature matrix, σ x represents the standard deviation of the elements in the i-th row, μ y represents the average value of the elements in the jth column, σ y represents the standard deviation of the elements in the jth column.

[0124] 2.4 Neural Network Training

[0125] The image samples after block processing, sample annotation information and the extracted sample GLCM information are input into the constructed improved B-CNN network, and the improved B-CNN network is trained until convergence.

[0126] 3. Practical Application

[0127] like Figure 1 As shown in the figure, in the power line corridor, the transmission lines are invaded in low-density vegetation areas and high-density vegetation areas. In practical applications, the environmental image of the transmission line to be detected is collected, and preprocessing including image segmentation and GLCM feature extraction is performed. Then, the image and GLCM features are input into the trained improved B-CNN model to obtain the recognition result output by the model, and then the image of the vegetation invasion area is output. The labeled image is output, as shown in Figure 1. Figure 3 shown.

[0128] Example 2

[0129] Based on the same inventive concept as Example 1, this embodiment introduces a transmission line vegetation intrusion detection device based on an improved B-CNN algorithm, including:

[0130] An image acquisition module is configured to acquire an image of a power transmission line environment to be detected;

[0131] A preprocessing module is configured to preprocess the transmission line environment image;

[0132] A GLCM feature extraction module is configured to extract GLCM features from the preprocessed image;

[0133] The vegetation intrusion detection module is configured to input the pre-processed transmission line environment image and the extracted image GLCM features into a pre-trained transmission line vegetation intrusion detection model to obtain a classification result output by the model;

[0134] And, the detection result output module is configured to output images in which the vegetation intrusion area and the transmission line area are marked respectively according to the classification result of the transmission line vegetation intrusion detection model.

[0135] The specific functional implementation of the above-mentioned functional modules, as well as the construction, training and image processing principles of the transmission line vegetation intrusion detection model, can all be referred to the relevant contents in Example 1.

[0136] Example 3

[0137] Based on the same inventive concept as Example 1 and Example 2, this embodiment introduces a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the transmission line vegetation intrusion detection method based on the improved B-CNN algorithm as described in the embodiment is implemented.

[0138] In summary, the present invention can realize the recognition of low-density and high-density vegetation areas invading near the power transmission line in the image, and its detection effect is comparable to that of the original image. Figure 3 As shown, the detection efficiency and accuracy are greatly improved compared with traditional technologies.

[0139] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0140] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0143] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for detecting vegetation intrusion on power lines based on an improved B-CNN algorithm, characterized in that: include: Acquire an environmental image of a transmission line to be detected; Preprocessing the transmission line environment image; Extract GLCM features from the preprocessed image; The preprocessed transmission line environment image and the extracted image GLCM features are input into the pre-trained transmission line vegetation intrusion detection model to obtain the classification result output by the model; Output images in which vegetation invasion areas and transmission line areas are marked respectively according to the classification results of the transmission line vegetation invasion detection model; The transmission line vegetation intrusion detection model adopts a B-CNN network, which includes an input layer, a convolution layer, a bilinear pooling layer and a Softmax normalization layer arranged in sequence. The convolution layer includes a convolution network A and a convolution network B respectively connected to the input layer. The outputs of the two are processed by the bilinear pooling layer for outer product and pooling to obtain a bilinear vector. The B-CNN network is expressed as the following formula: ; In the formula, Represents the output classification of the B-CNN network; and Respectively represent the feature extraction functions of convolutional network A and convolutional network B, and are the mapping process functions , which means that the image and location area Through the set mapping relationship, it is mapped into a A vector of dimension size; represents the pooling function, Represents the bidirectional classification recognition vector through the Softmax normalization layer; The outer product processing formula is expressed as: ; in and Respectively, images Location The features output by convolutional network A and convolutional network B; The pooling function It is expressed as: ; in For images The pooling result of the convolution output after the outer product processing.

2. The method according to claim 1, characterized in that: The preprocessing of the transmission line environment image also includes dividing the image into blocks, each image block has n×n pixels, and the transmission line environment image input to the transmission line vegetation intrusion detection model is the image after block processing.

3. The method according to claim 1, characterized in that: In the B-CNN network of the transmission line vegetation intrusion detection model, the output ends of convolutional network A and convolutional network B are connected to the channel attention module and the spatial attention module respectively; The feature calculation formula of the channel attention module is: ; The feature calculation formula of the spatial attention module is: ; In the formula, represents the features of the convolutional network output, is the feature output by the channel attention module, Output features of the spatial attention module; and Respectively represent the characteristics The results of average pooling and maximum pooling are: , represents the weights of different layers in the multilayer perceptron MLP, represents the sigmoid function, represents element-wise multiplication, Indicates a convolution operation with a filter size of 7*7; represents the average pooling function, Represents the maximum pooling function.

4. The method according to claim 1, characterized in that: The extracting GLCM features from the pre-processed image comprises: The preprocessed transmission line environment image is linearly converted into a one-dimensional vector and the statistical moment is calculated; Calculate the angular second moment, entropy ENT, contrast Con, inverse difference matrix IDM and correlation Correlation; The statistical moments include: average value , standard deviation ,variance , third-order center distance , skewness Kurtosis moment , the calculation formula is: ; ; ; ; ; ; In the above formula, represents the averaging function; The calculation formulas for the angular second moment, entropy ENT, contrast Con, inverse difference matrix IDM and correlation Correlation are: ; ; ; ; ; In the formula, Indicates the first The average value of the row elements, Indicates The standard deviation of the row elements, Indicates The average value of the column elements, Indicates Standard deviation of the column elements.

5. The method according to claim 1, characterized in that: The transmission line vegetation intrusion detection model training method comprises: Obtain image samples from different transmission line environments and their vegetation invasion area annotation information; Divide the image samples into training set samples and test set samples; Preprocess the image samples to determine the annotation information of the vegetation invasion area and the transmission line area; Perform GLCM feature extraction on the preprocessed image samples; The preprocessed training set samples, sample annotation information, and extracted sample GLCM information are input into the pre-built B-CNN network to train the B-CNN network; The classification accuracy of the trained B-CNN network is tested through the test set samples. If the classification accuracy reaches the preset accuracy requirement, the training is stopped. Otherwise, the B-CNN network training is continued through the training set samples.

6. The method according to claim 5, characterized in that: The image samples have various resolutions and ratios, including images containing high-density vegetation areas and images containing low-density vegetation areas.

7. A transmission line vegetation intrusion detection device based on an improved B-CNN algorithm, characterized in that: include: An image acquisition module is configured to acquire an image of a power transmission line environment to be detected; A preprocessing module is configured to preprocess the transmission line environment image; A GLCM feature extraction module is configured to extract GLCM features from the preprocessed image; The vegetation intrusion detection module is configured to input the pre-processed transmission line environment image and the extracted image GLCM features into a pre-trained transmission line vegetation intrusion detection model to obtain a classification result output by the model; and, a detection result output module configured to output images in which the vegetation intrusion area and the transmission line area are respectively marked according to the classification result of the transmission line vegetation intrusion detection model; The transmission line vegetation intrusion detection model adopts a B-CNN network, which includes an input layer, a convolution layer, a bilinear pooling layer and a Softmax normalization layer arranged in sequence. The convolution layer includes a convolution network A and a convolution network B respectively connected to the input layer. The outputs of the two are processed by the bilinear pooling layer for outer product and pooling to obtain a bilinear vector. The B-CNN network is expressed as the following formula: ; In the formula, Represents the output classification of the B-CNN network; and Respectively represent the feature extraction functions of convolutional network A and convolutional network B, and are the mapping process functions , which means that the image and location area Through the set mapping relationship, it is mapped into a A vector of dimension size; represents the pooling function, Represents the bidirectional classification recognition vector through the Softmax normalization layer; The outer product processing formula is expressed as: ; in and Respectively, images Location The features output by convolutional network A and convolutional network B; The pooling function It is expressed as: ; in For images The pooling result of the convolution output after the outer product processing.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting vegetation intrusion on a transmission line based on an improved B-CNN algorithm as described in any one of claims 1 to 6 is implemented.

Citation Information

Patent Citations

  • Intrusion detection model establishing method, device and data processing device

    CN109460787A

  • Crop planting structure monthly scale dynamic extraction method

    CN110321861A

  • Underwater building crack detection method based on multi-scale features and stacked full convolutional network

    CN111257341A

  • Intelligent power grid foreign matter intrusion detection method and system based on visual perception

    CN112464851A

  • Image privacy perception method based on deep learning

    US20210224586A1