Power transmission line fault identification method and system based on deep learning

Abstract

The invention belongs to the technical field of power line fault identification, and discloses a power transmission line fault identification method and system based on deep learning, and the method comprises the steps: obtaining a monitoring image of a power transmission line, and carrying out the preprocessing of the monitoring image of the power transmission line, and obtaining a target image; performing light value solving on the target image based on the enhanced light value estimation model to obtain an atmospheric light value of the target image; performing multi-level feature extraction on the target image based on the dense pyramid network to obtain a transmissivity graph of the target image; inputting the atmospheric light value of the target image and the transmissivity graph of the target image into an atmospheric scattering model to obtain a defogged image; and inputting the defogged image into a pre-constructed fault identification model, and outputting the fault type of the power transmission line. According to the invention, through accurate calculation of the atmospheric light value and the transmissivity graph of the target image, a high-quality defogged image can be obtained; by performing fault identification on the high-quality defogged image, the accuracy of line fault identification can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power line fault identification, and specifically relates to a method and system for identifying power transmission line faults based on deep learning. Background Art

[0002] China's demand for electricity continues to grow with its rapid economic development. China is carrying out power grid construction across the country to meet the needs of industrial and civilian electricity. With the continuous development of hydropower energy in southwest my country and southeast Tibet, a number of ultra-high voltage transmission projects have been put into operation as the "artery" for hydropower transmission. However, due to the existence of a large number of nonlinear devices and uncertain factors in the power system, the system stability is reduced and the accident rate is increased. Power lines that are outdoors for a long time are not only disturbed by natural environmental factors, but also bear the load of power systems and mechanical equipment. Natural factors such as lightning, hurricanes, and heavy rains can cause negative effects such as corrosion and oxidation of power line equipment. For the maintenance of power lines, failure to take preventive measures can easily lead to the emergence of corresponding problems.

[0003] In the past, the inspection method of power transmission lines mainly relied on manual inspection, relying on the naked eye and manual experience to identify defects. Often, only obvious defects such as the self-explosion of some tower base insulators can be detected, and it is difficult to detect faults such as hardware corrosion, loose bolts, and pin shedding. However, this method has a long detection cycle, high labor intensity for workers, and high maintenance costs. At the same time, during the detection process, the degree of fault in the transmission line depends on the judgment of the workers based on their own experience and handheld devices, which makes the professional ability and technical requirements of the workers high. When the workers themselves lack experience and there are errors in the field of view of manual detection, the accuracy of line inspection will be greatly reduced.

[0004] With the development of drone technology, the use of drone inspection has saved more manpower and financial resources and has been widely used. Drones use computer vision systems to capture images of transmission lines. By performing target detection and defect recognition on inspection images, intelligent inspection of transmission lines can be achieved. Since drone inspections are mostly carried out in the field, the image quality is affected by fog and haze. Haze weather will increase the noise of transmission line inspection images, reduce brightness and contrast, and lose detail information, significantly reducing image quality. Due to the serious loss of feature information in foggy images, the accuracy of target detection algorithms decreases, which reduces the accuracy of defect recognition in transmission line images. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for identifying power transmission line faults based on deep learning, so as to solve the problem that the accuracy of target detection algorithm decreases due to serious loss of image feature information in fog, thereby reducing the accuracy of defect recognition in transmission line images.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for identifying power transmission line faults based on deep learning, the method comprising: Acquire a monitoring image of the transmission line, pre-process the monitoring image of the transmission line, and obtain a target image; Solve the light value of the target image based on the enhanced light value estimation model to obtain the atmospheric light value of the target image; Perform multi-level feature extraction on the target image based on a dense pyramid network to obtain a transmittance map of the target image; The atmospheric light value of the target image and the transmittance map of the target image are input into the atmospheric scattering model to obtain a defogging image; The dehazed image is input into the pre-built fault identification model, and the fault type of the power transmission line is output.

[0007] Preferably, the dense pyramid network comprises: a first encoder and a first decoder, wherein the first encoder and the first decoder are both internally composed of densely connected dense blocks and multi-layer pyramid pooling modules.

[0008] Preferably, the enhanced light value estimation model comprises: an enhanced U-net network, an attention network and a fusion network; The enhanced U-net network is used to: extract features from the target image to obtain a light value feature matrix; The attention network is used to: perform feature transformation on the target image based on the attention mechanism to obtain an initial feature matrix; extract high-frequency information from the initial feature matrix to obtain a high-frequency feature matrix; fuse the high-frequency feature matrix with the initial feature matrix to obtain an attention feature matrix; The fusion network is used to fuse the attention feature matrix and the light value feature matrix to obtain an atmospheric light value feature map, and use the atmospheric light value feature map as the atmospheric light value of the target image.

[0009] Preferably, high-frequency information is extracted from the initial feature matrix to obtain a high-frequency feature matrix, including: Constructing a plurality of first residual modules connected in sequence; wherein the output of the mth first residual module is used as the input of the m+1th first residual module, where m is a positive integer; In each first residual module, convolving the input matrix of the first residual module to obtain a first matrix; Performing global average pooling on the first matrix to obtain a first low-channel matrix; Deconvolve the first low channel matrix to obtain a first low channel weight matrix; The first low channel weight matrix is ​​multiplied by the first low channel matrix to obtain a second matrix, and the second matrix is ​​used as the output matrix of the first residual module; wherein the initial feature matrix is ​​used as the input matrix of the first first residual module, and the second matrix output by the last first residual module is used as the high-frequency feature matrix.

[0010] Preferably, the enhanced U-net network comprises: a second encoder, a feature recovery module and a second decoder; Perform feature extraction on the target image to obtain the light value feature matrix, including: Perform multi-level downsampling on the target image based on the second encoder to obtain a downsampling feature matrix; Based on the feature recovery module, multi-level feature recovery is performed on the downsampled feature matrix to obtain the recovered features; Based on the second decoder, multi-level upsampling and feature enhancement are performed on the restored features to obtain a light value feature matrix.

[0011] Preferably, the restored features are subjected to multi-level upsampling and feature enhancement based on the second decoder to obtain a light value feature matrix, including: Initialize the input sampling matrix of each level of upsampling, deconvolve the output of the j-th level upsampling as the input of the j-1-th level upsampling, and deconvolve the restored features as the input of the highest level upsampling, where j is a positive integer greater than or equal to 2; Deconvolve the j-th level upsampled input to obtain the j-th level sampling features; The down-sampled feature matrix with the same feature dimension as the j-th level sampling feature and the j-th level sampling feature are fused and enhanced to obtain the j-th level enhanced feature; Convolve the j-th level enhancement feature to obtain the j-th level enhancement feature after convolution, and use the j-th level enhancement feature after convolution as the output of the j-th level upsampling; use the first-level enhancement feature after convolution output of the first-level upsampling as the light value feature matrix.

[0012] Preferably, the feature recovery module comprises a plurality of connected second residual modules, and the output of the nth second residual module is used as the input of the n+1th second residual module, where n is a positive integer; Based on the feature recovery module, the downsampled feature matrix is ​​restored at multiple levels to obtain the restored features, including: In each second residual module, convolving the input matrix of the second residual module to obtain a third matrix; Perform global average pooling on the third matrix to obtain the second low-channel matrix; Deconvolve the second low channel matrix to obtain the second low channel weight matrix; The second low channel weight matrix is ​​multiplied by the second low channel matrix to obtain a fourth matrix, and the fourth matrix is ​​used as the output matrix of the second residual module; wherein the downsampled feature matrix is ​​used as the input matrix of the first second residual module, and the fourth matrix output by the last second residual module is used as the restored feature.

[0013] Preferably, the method further comprises: Based on the discriminator, the atmospheric light value, transmittance map and dehazed image are optimized by game theory to obtain the optimized dehazed image.

[0014] Preferably, the method further comprises: constructing a fault identification model, comprising: Acquire power transmission line sample data, preprocess the power transmission line sample data, and obtain preprocessed power transmission line sample data; The convolutional neural network is trained based on the preprocessed power transmission line sample data until the convolutional neural network converges to obtain a trained convolutional neural network, and the trained convolutional neural network is used as a fault recognition model.

[0015] In a second aspect, the present invention provides a power transmission line fault identification system based on deep learning, the system is used to implement the above-mentioned power transmission line fault identification method based on deep learning, and the system includes: A data acquisition module is used to acquire a monitoring image of the power transmission line, pre-process the monitoring image of the power transmission line, and obtain a target image; A light value solving module is used to solve the light value of the target image based on the enhanced light value estimation model to obtain the atmospheric light value of the target image; A feature extraction module is used to perform multi-level feature extraction on the target image based on a dense pyramid network to obtain a transmittance map of the target image; An image defogging module is used to input the atmospheric light value of the target image and the transmittance map of the target image into the atmospheric scattering model to obtain a defogging image; The fault identification module is used to input the dehazed image into a pre-built fault identification model and output the fault type of the power transmission line.

[0016] Beneficial effects: The present invention can accurately calculate the atmospheric light value of the target image through the strong light value estimation model, and simultaneously use the dense pyramid network to perform multi-level feature extraction on the target image, so as to obtain an accurate transmittance map of the target image; then the atmospheric light value and the transmittance map are introduced into the atmospheric scattering model, so as to solve the defogging image and obtain a high-quality defogging image; and by performing fault identification on the high-quality defogging image, the accuracy of line fault identification can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings: Figure 1 It is a flow chart of a method for identifying power transmission line faults based on deep learning provided by an embodiment of the present invention; Figure 2 It is a block diagram of a power transmission line fault identification system based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0019] Embodiment 1 Figure 1 is a flow chart of a method for identifying power transmission line faults based on deep learning provided by an embodiment of the present invention, such as Figure 1 As shown, this embodiment provides a method for identifying power transmission line faults based on deep learning, the method comprising: Step S10: Acquire a monitoring image of the power transmission line, and pre-process the monitoring image of the power transmission line to obtain a target image.

[0020] In this embodiment, a UAV is used to collect images of the transmission line, so that monitoring images of the transmission line can be obtained. The preprocessing includes grayscale processing, image flipping, image cropping, geometric correction and other processing.

[0021] Step S20: solving the light value of the target image based on the enhanced light value estimation model to obtain the atmospheric light value of the target image.

[0022] In this embodiment, the existing atmospheric light values ​​often use empirical values. Since the specific values ​​of the atmospheric light values ​​in different scenes may be different, using the empirical values ​​as the atmospheric light values ​​of the target image for dehazing processing will reduce the dehazing effect. Therefore, in order to improve the accuracy of the atmospheric light values, the enhanced light value estimation model is used to solve the atmospheric light values ​​of the target image to obtain accurate atmospheric light values.

[0023] Specifically, the enhanced light value estimation model includes: an enhanced U-net network, an attention network and a fusion network; The enhanced U-net network is used to: extract features from the target image to obtain a light value feature matrix; The attention network is used to: perform feature transformation on the target image based on the attention mechanism to obtain an initial feature matrix; extract high-frequency information from the initial feature matrix to obtain a high-frequency feature matrix; fuse the high-frequency feature matrix with the initial feature matrix to obtain an attention feature matrix; The fusion network is used to fuse the attention feature matrix and the light value feature matrix to obtain an atmospheric light value feature map, and use the atmospheric light value feature map as the atmospheric light value of the target image.

[0024] As a further optimization of this embodiment, high-frequency information is extracted from the initial feature matrix to obtain a high-frequency feature matrix, including: Step a201: construct multiple first residual modules connected in sequence; wherein the output of the mth first residual module is used as the input of the (m+1)th first residual module, where m is a positive integer; Step a202: in each first residual module, convolve the input matrix of the first residual module to obtain a first matrix; Step a203: performing global average pooling processing on the first matrix to obtain a first low-channel matrix; Step a204: performing deconvolution on the first low channel matrix to obtain a first low channel weight matrix; Step a205: multiply the first low channel weight matrix by the first low channel matrix to obtain a second matrix, and use the second matrix as the output matrix of the first residual module; wherein the initial feature matrix is ​​used as the input matrix of the first first residual module, and the second matrix output by the last first residual module is used as the high-frequency feature matrix.

[0025] In this embodiment, through steps a202 to a205, multiple first residual modules can redistribute the features of different channels, and then combine with the initial feature matrix obtained by the attention mechanism to enable the attention network to have a stronger fitting ability, retain more high-frequency feature information, and at the same time speed up the training speed of the network.

[0026] Since the attention network has a strong fitting ability, it is easy to over-fit, resulting in reduced accuracy of the atmospheric light feature map. In order to prevent over-fitting, in this embodiment, the enhanced U-net network includes: a second encoder, a feature recovery module and a second decoder. As a further optimization of this embodiment, feature extraction is performed on the target image to obtain a light value feature matrix, including: Step b201: Perform multi-level downsampling on the target image based on the second encoder to obtain a downsampling feature matrix.

[0027] Step b202: Perform multi-level feature recovery on the downsampled feature matrix based on the feature recovery module to obtain restored features.

[0028] In this embodiment, the feature recovery module includes a plurality of connected second residual modules, and the output of the nth second residual module is used as the input of the n+1th second residual module, where n is a positive integer; then step b02 specifically includes the following steps: A1: In each second residual module, convolve the input matrix of the second residual module to obtain a third matrix; A2: Perform global average pooling on the third matrix to obtain the second low-channel matrix; A3: Deconvolve the second lowest channel matrix to obtain the second lowest channel weight matrix; A4: Multiply the second low channel weight matrix by the second low channel matrix to obtain a fourth matrix, and use the fourth matrix as the output matrix of the second residual module; wherein the downsampled feature matrix is ​​used as the input matrix of the first second residual module, and the fourth matrix output by the last second residual module is used as the restored feature.

[0029] In this embodiment, the feature recovery module composed of multiple second residual modules, steps A1 to A4 can retain as many effective features as possible, so that the extracted feature information has good robustness, and the enhanced light value estimation model can have a certain generalization effect.

[0030] Step b203: Perform multi-level upsampling and feature enhancement on the restored features based on the second decoder to obtain a light value feature matrix.

[0031] Specifically, based on the second decoder, the restored features are multi-level up-sampled and feature enhanced to obtain a light value feature matrix, including: B1: Initialize the input sampling matrix of each level of upsampling, deconvolve the output of the j-th level upsampling as the input of the j-1-th level upsampling, and deconvolve the restored features as the input of the highest level upsampling, where j is a positive integer greater than or equal to 2; B2: Deconvolve the j-th level upsampled input to obtain the j-th level sampling features; B3: The down-sampled feature matrix with the same feature dimension as the j-th level sampling feature is fused and enhanced with the j-th level sampling feature to obtain the j-th level enhanced feature; B4: Convolve the j-th level enhancement feature to obtain the j-th level enhancement feature after convolution, and use the j-th level enhancement feature after convolution as the output of the j-th level upsampling; use the first-level enhancement feature after convolution output by the first-level upsampling as the light value feature matrix.

[0032] In this embodiment, during the upsampling process, the downsampling feature matrix in the downsampling is added and fused with the corresponding sampling features to enhance the upsampling sampling features, thereby enhancing the recovery of features in the decoder.

[0033] This embodiment downsamples and upsamples the target image by enhancing the U-net network, retaining the important feature information extracted in the sampling process to the maximum extent, and then combines the strong fitting ability of the attention network to further improve the robustness of the enhanced light value estimation model. The enhanced light value estimation model can output a high-precision atmospheric light map.

[0034] Step S30: performing multi-level feature extraction on the target image based on a dense pyramid network to obtain a transmittance map of the target image.

[0035] In this embodiment, the dense pyramid network includes: a first encoder and a first decoder, and the first encoder and the first decoder are both composed of densely connected dense blocks and multi-layer pyramid pooling modules.

[0036] In this embodiment, for the dense pyramid network, firstly, the dense blocks are constructed based on the CNN network (Convolutional Neural Networks), and the dense blocks can effectively extract the multi-level features of the image; secondly, the dense blocks can make the entire encoding and decoding model converge quickly. The encoder and decoder contain a multi-layer pyramid pooling module, which completes the estimation of the transmittance map based on the global transmittance information of the image, effectively avoiding the disadvantage of ignoring global information due to excessive concentration on local detail information, and can obtain a more accurate transmittance map.

[0037] Among them, the first encoder is composed of a traditional convolution block connected to four dense blocks, and the output size is 1 / 32 of the original input image. The first decoder after the first encoder is composed of four dense blocks connected to a convolution block, and the output size is the same as the original input image. The corresponding modules between the first encoder and the first decoder have a direct connection. Since the feature rate map is not generated by multiple types of features extracted from images of different scales, although the densely connected encoder-decoder combines different features of the network, the transmittance map actually generated still does not have the global structural information of multi-scale and multi-features. Considering the above situation, the embodiment adds a multi-level pyramid pooling module after the encoder-decoder, and the different layers of the feature pyramid network fully extract the feature information of the target object in the image, and these feature information will be used for accurate estimation of the transmittance map. Specifically, this embodiment constructs a pyramid pooling module with four-level pooling operations, and the result sizes of the output of each level of pooling operation are 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the original image size, respectively. In order to establish a connection between the transmittance map and the original image, it is necessary to upsample the four different sizes of transmittance maps one by one to the original image size, so as to obtain a more accurate transmittance map.

[0038] Step S40: inputting the atmospheric light value of the target image and the transmittance map of the target image into the atmospheric scattering model to obtain a defogging image.

[0039] In this embodiment, the expression of the atmospheric scattering model is: ; In the formula, For an image containing fog, For a fog-free image, is the atmospheric light value, is the transmittance map, and z is the pixel position in the image.

[0040] Therefore, after calculating the atmospheric light value and transmittance map of the target image through steps S20 and S30, the atmospheric scattering model can be used to calculate the fog-free image. , the fog-free image as a dehazed image.

[0041] As a further optimization of this embodiment, the method also includes: performing game optimization on the atmospheric light value, the transmittance map and the dehazed image based on the discriminator to obtain an optimized dehazed image; in this embodiment, through game optimization, the transmittance map and dehazed image finally generated are as close to the real data as possible, thereby improving the dehazed effect.

[0042] Step S50: inputting the dehazed image into a pre-built fault recognition model, and outputting the fault type of the power transmission line.

[0043] As a further optimization of this embodiment, the method further includes: constructing a fault identification model, including: Acquire power transmission line sample data, pre-process the power transmission line sample data, and obtain pre-processed power transmission line sample data; the power transmission line sample data contains labels with defects, such as: conductor faults such as broken wires, loose strands, and insulator damage, insulator faults such as insulator rupture, insulator flashover, and insulator aging, hardware faults such as loose bolts and poor oxidation contact, tower faults such as tower tilt, tower material corrosion, and tower material breakage, grounding faults such as ground body corrosion and grounding network breakage, line short circuit faults such as short circuit to ground and phase short circuit, and natural state faults such as ice damage, wind damage, and temperature changes; The convolutional neural network is trained based on the preprocessed power transmission line sample data until the convolutional neural network converges to obtain a trained convolutional neural network, and the trained convolutional neural network is used as a fault recognition model.

[0044] The present invention can accurately calculate the atmospheric light value of the target image through the strong light value estimation model, and simultaneously use the dense pyramid network to perform multi-level feature extraction on the target image, so as to obtain an accurate transmittance map of the target image; then the atmospheric light value and the transmittance map are introduced into the atmospheric scattering model, so as to solve the defogging image and obtain a high-quality defogging image; and by performing fault identification on the high-quality defogging image, the accuracy of line fault identification can be improved.

[0045] Embodiment 2 Figure 2 is a block diagram of a power transmission line fault identification system based on deep learning provided by an embodiment of the present invention, such as Figure 2 As shown, this embodiment provides a power transmission line fault identification system based on deep learning, the system is used to implement the power transmission line fault identification method based on deep learning of embodiment 1, and the system includes: A data acquisition module is used to acquire a monitoring image of the power transmission line, pre-process the monitoring image of the power transmission line, and obtain a target image; A light value solving module is used to solve the light value of the target image based on the enhanced light value estimation model to obtain the atmospheric light value of the target image; A feature extraction module is used to perform multi-level feature extraction on the target image based on a dense pyramid network to obtain a transmittance map of the target image; An image defogging module is used to input the atmospheric light value of the target image and the transmittance map of the target image into the atmospheric scattering model to obtain a defogging image; The fault identification module is used to input the dehazed image into a pre-built fault identification model and output the fault type of the power transmission line.

[0046] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the above-mentioned power transmission line fault identification method based on deep learning is implemented.

[0047] The present invention can accurately calculate the atmospheric light value of the target image through the strong light value estimation model, and simultaneously use the dense pyramid network to perform multi-level feature extraction on the target image, so as to obtain an accurate transmittance map of the target image; then the atmospheric light value and the transmittance map are introduced into the atmospheric scattering model, so as to solve the defogging image and obtain a high-quality defogging image; and by performing fault identification on the high-quality defogging image, the accuracy of line fault identification can be improved.

[0048] 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 combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0049] 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.

[0050] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for identifying power transmission line faults based on deep learning, characterized in that: The method comprises: Acquire a monitoring image of the transmission line, pre-process the monitoring image of the transmission line, and obtain a target image; Solve the light value of the target image based on the enhanced light value estimation model to obtain the atmospheric light value of the target image; Perform multi-level feature extraction on the target image based on a dense pyramid network to obtain a transmittance map of the target image; The atmospheric light value of the target image and the transmittance map of the target image are input into the atmospheric scattering model to obtain a defogging image; The dehazed image is input into the pre-built fault identification model, and the fault type of the power transmission line is output.

2. The method for identifying power transmission line faults based on deep learning according to claim 1, characterized in that: The dense pyramid network includes: a first encoder and a first decoder, wherein the first encoder and the first decoder are both internally composed of densely connected dense blocks and multi-layer pyramid pooling modules.

3. The method for identifying power transmission line faults based on deep learning according to claim 1, characterized in that: The enhanced light value estimation model includes: an enhanced U-net network, an attention network and a fusion network; The enhanced U-net network is used to: extract features from the target image to obtain a light value feature matrix; The attention network is used to: perform feature transformation on the target image based on the attention mechanism to obtain an initial feature matrix; extract high-frequency information from the initial feature matrix to obtain a high-frequency feature matrix; fuse the high-frequency feature matrix with the initial feature matrix to obtain an attention feature matrix; The fusion network is used to fuse the attention feature matrix and the light value feature matrix to obtain an atmospheric light value feature map, and use the atmospheric light value feature map as the atmospheric light value of the target image.

4. The method for identifying power transmission line faults based on deep learning according to claim 3, characterized in that: Extract high-frequency information from the initial feature matrix to obtain a high-frequency feature matrix, including: Constructing a plurality of first residual modules connected in sequence; wherein the output of the mth first residual module is used as the input of the m+1th first residual module, where m is a positive integer; In each first residual module, convolving the input matrix of the first residual module to obtain a first matrix; Performing global average pooling on the first matrix to obtain a first low-channel matrix; Deconvolve the first low channel matrix to obtain a first low channel weight matrix; The first low channel weight matrix is ​​multiplied by the first low channel matrix to obtain a second matrix, and the second matrix is ​​used as the output matrix of the first residual module; wherein the initial feature matrix is ​​used as the input matrix of the first first residual module, and the second matrix output by the last first residual module is used as the high-frequency feature matrix.

5. The method for identifying power transmission line faults based on deep learning according to claim 3, characterized in that: The enhanced U-net network includes: a second encoder, a feature recovery module and a second decoder; Perform feature extraction on the target image to obtain the light value feature matrix, including: Perform multi-level downsampling on the target image based on the second encoder to obtain a downsampling feature matrix; Based on the feature recovery module, multi-level feature recovery is performed on the downsampled feature matrix to obtain the recovered features; Based on the second decoder, multi-level upsampling and feature enhancement are performed on the restored features to obtain a light value feature matrix.

6. The method for identifying power transmission line faults based on deep learning according to claim 5, characterized in that: Based on the second decoder, the restored features are multi-level up-sampled and feature enhanced to obtain a light value feature matrix, including: Initialize the input sampling matrix of each level of upsampling, deconvolve the output of the j-th level upsampling as the input of the j-1-th level upsampling, and deconvolve the restored features as the input of the highest level upsampling, where j is a positive integer greater than or equal to 2; Deconvolve the j-th level upsampled input to obtain the j-th level sampling features; The down-sampled feature matrix with the same feature dimension as the j-th level sampling feature and the j-th level sampling feature are fused and enhanced to obtain the j-th level enhanced feature; Convolve the j-th level enhancement feature to obtain the j-th level enhancement feature after convolution, and use the j-th level enhancement feature after convolution as the output of the j-th level upsampling; use the first-level enhancement feature after convolution output of the first-level upsampling as the light value feature matrix.

7. The method for identifying power transmission line faults based on deep learning according to claim 5, characterized in that: The feature recovery module includes a plurality of connected second residual modules, and the output of the nth second residual module is used as the input of the n+1th second residual module, where n is a positive integer; Based on the feature recovery module, the downsampled feature matrix is ​​restored at multiple levels to obtain the restored features, including: In each second residual module, convolving the input matrix of the second residual module to obtain a third matrix; Perform global average pooling on the third matrix to obtain the second low-channel matrix; Deconvolve the second low channel matrix to obtain the second low channel weight matrix; The second low channel weight matrix is ​​multiplied by the second low channel matrix to obtain a fourth matrix, and the fourth matrix is ​​used as the output matrix of the second residual module; wherein the downsampled feature matrix is ​​used as the input matrix of the first second residual module, and the fourth matrix output by the last second residual module is used as the restored feature.

8. The method for identifying power transmission line faults based on deep learning according to claim 1, characterized in that: The method further comprises: Based on the discriminator, the atmospheric light value, transmittance map and dehazed image are optimized by game theory to obtain the optimized dehazed image.

9. The method for identifying power transmission line faults based on deep learning according to claim 1, characterized in that: The method further includes: constructing a fault identification model, including: Acquire power transmission line sample data, preprocess the power transmission line sample data, and obtain preprocessed power transmission line sample data; The convolutional neural network is trained based on the preprocessed power transmission line sample data until the convolutional neural network converges to obtain a trained convolutional neural network, and the trained convolutional neural network is used as a fault recognition model.

10. A power transmission line fault identification system based on deep learning, the system is used to implement the power transmission line fault identification method based on deep learning as described in any one of claims 1 to 9, characterized in that: The system comprises: A data acquisition module is used to acquire a monitoring image of the power transmission line, pre-process the monitoring image of the power transmission line, and obtain a target image; A light value solving module is used to solve the light value of the target image based on the enhanced light value estimation model to obtain the atmospheric light value of the target image; A feature extraction module is used to perform multi-level feature extraction on the target image based on a dense pyramid network to obtain a transmittance map of the target image; An image defogging module is used to input the atmospheric light value of the target image and the transmittance map of the target image into the atmospheric scattering model to obtain a defogging image; The fault identification module is used to input the dehazed image into a pre-built fault identification model and output the fault type of the power transmission line.

Images