A method, apparatus and equipment for detecting external damage to power distribution lines

Abstract

This invention discloses a method, apparatus, and device for detecting external damage to power distribution lines. The method includes acquiring an image of the power distribution line to be analyzed via an edge device; inputting the image into an external damage detection model composed of a feature extraction network, a feature fusion network, and a detection network; using the feature extraction network, which consists of a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected sequentially, to extract features from the image, outputting multiple power distribution line feature maps group by group; using the feature fusion network to perform feature fusion on the multiple feature maps to generate a fused feature map; and using the detection network to perform external damage detection on the fused feature map to determine whether the image of the power distribution line to be analyzed has an external damage risk. This invention solves the technical problem that existing models are too large, resulting in low detection efficiency and failing to meet the real-time detection requirements of practical applications.

Description

Technical Field

[0001] This invention relates to the field of external damage detection technology for power distribution lines, and in particular to a method, apparatus and equipment for detecting external damage to power distribution lines. Background Technology

[0002] Power grid distribution lines are numerous, widely distributed, and geographically diverse. Furthermore, the corridors surrounding these lines are constantly subject to unpredictable factors such as human activity, extreme weather, and construction work. Such situations can easily lead to line tripping. Therefore, it is crucial to conduct external damage detection on power grid distribution lines and take timely corresponding measures to maintain their stable operation.

[0003] Currently, the main method for detecting power distribution lines is to build models using deep learning-based target detection methods. However, this method focuses more on improving the accuracy of external damage detection, resulting in a slow detection speed. Due to the large number of power distribution lines to be analyzed, it cannot meet the real-time detection requirements in practical applications. Summary of the Invention

[0004] This invention provides a method, apparatus, and equipment for detecting external damage to power distribution lines. It solves the technical problem that current deep learning-based target detection methods focus more on improving the accuracy of external damage detection when building models for power distribution lines, resulting in slow model detection speed. Furthermore, due to the large number of power distribution lines to be analyzed, these methods cannot meet the real-time detection requirements of practical applications.

[0005] The first aspect of this invention provides a method for detecting external damage to power distribution lines, comprising:

[0006] Acquire an image of the power distribution line to be analyzed, and input the image of the power distribution line to be analyzed into the external damage detection model, wherein the external damage detection model includes a feature extraction network, a feature fusion network and a detection network;

[0007] The feature extraction network is used to perform feature extraction on the power distribution line image to be analyzed, and outputs multiple power distribution line feature maps in groups. The feature extraction network includes a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected in sequence.

[0008] The feature fusion network is used to perform feature fusion operations on multiple power distribution line feature maps to generate a fused feature map.

[0009] The detection network is used to perform external damage detection on the fused feature map to determine whether there is an external damage risk in the power distribution line image to be analyzed.

[0010] Optionally, the step of using the feature extraction network to perform feature extraction on the power distribution line image to be analyzed, and outputting multiple power distribution line feature maps in groups, includes:

[0011] The first scale convolutional group is used to extract features from the power distribution line image to be analyzed, generating a first power distribution line feature map, wherein the first scale convolutional group includes a first extraction branch, a second extraction branch and a feature fusion layer connected in sequence.

[0012] The first power distribution line feature map is used to extract features by a second-scale convolution group to generate a second power distribution line feature map, wherein the second-scale convolution group includes a second extraction branch.

[0013] The second power distribution line feature map is used to extract features by a third-scale convolution group to generate a third power distribution line feature map, wherein the third-scale convolution group includes a third extraction branch and a feature fusion layer connected in sequence.

[0014] The fourth-scale convolutional group is used to extract features from the third power distribution line feature map to generate the fourth power distribution line feature map. The fourth-scale convolutional group includes a 3×3 standard convolutional layer, a DY-ReLU activation function layer and a second extraction branch connected in sequence.

[0015] The fifth power distribution line feature map is generated by extracting features from the fourth power distribution line feature map using a fifth-scale convolutional group, wherein the fifth-scale convolutional group includes a first extraction branch.

[0016] Optionally, the step of extracting features from the power distribution line image to be analyzed using a first-scale convolutional group to generate a first power distribution line feature map includes:

[0017] The first extraction branch is used to extract features from the power distribution line image to be analyzed, and a first extracted feature map is generated. The first extraction branch includes three first extraction modules connected in sequence.

[0018] The first extracted feature map is used to extract features through the second extraction branch to generate the second extracted feature map. The second extraction branch includes two first extraction models and a 2×2 pooling layer connected in sequence.

[0019] A feature fusion layer is used to perform feature fusion operations on the first extracted feature map and the second extracted feature map to generate a first power distribution line feature map.

[0020] The first extraction module includes a 3×3 standard convolutional layer and a DY-ReLU activation function layer connected in sequence.

[0021] Optionally, the step of extracting features from the second power distribution line feature map using a third-scale convolutional group to generate a third power distribution line feature map includes:

[0022] The third extraction branch is used to extract features from the feature map of the second power distribution line to generate the third extracted feature map. The third extraction branch includes a 3×1 standard convolutional layer, a 1×3 standard convolutional layer, a LeakyReLU activation layer and a 3×3 pooling layer connected in sequence.

[0023] The third power distribution line feature map is generated by performing feature fusion operation on the third extracted feature map and the second power distribution line feature map through the feature fusion layer.

[0024] Optionally, the feature fusion network includes a first bounding box module, a second bounding box module, a feature fusion layer, and a region of interest pooling layer. The step of performing feature fusion operations on multiple power distribution line feature maps through the feature fusion network to generate a fused feature map includes:

[0025] The first box selection module is used to extract features from the feature map of the first power distribution line to generate a first box selection feature map.

[0026] The second box selection module extracts features from the feature map of the second power distribution line to generate a second box selection feature map.

[0027] The first bounding box feature map and the second bounding box feature map are fused by a feature fusion layer to generate a feature map of interest.

[0028] The region of interest pooling layer is used to perform feature fusion operations on the feature map of interest, the feature map of the third power distribution line, the feature map of the fourth power distribution line, and the feature map of the fifth power distribution line to generate a fused feature map.

[0029] Optionally, the first box selection module includes a 5×5 standard convolutional layer, a DY-ReLU activation function layer, a 5×5 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer connected in sequence.

[0030] Optionally, the second box selection module includes a 3×3 standard convolutional layer, a DY-ReLU activation function layer, a 3×3 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer connected in sequence.

[0031] Optionally, the step of using the detection network to perform external damage detection on the fused feature map and determining whether the power distribution line image to be analyzed has an external damage risk includes:

[0032] The detection network is used to perform external damage detection on the fused feature map to generate a confidence score;

[0033] Determine whether the confidence score is less than a preset score threshold;

[0034] If the confidence score is greater than or equal to the score threshold, it is determined that the power distribution line image to be analyzed has an external damage risk.

[0035] If the confidence score is less than the score threshold, it is determined that the power distribution line image to be analyzed does not have the risk of external damage.

[0036] A second aspect of the present invention provides a power distribution line external damage detection device, comprising:

[0037] The image acquisition module is used to acquire the image of the power distribution line to be analyzed and input the image of the power distribution line to be analyzed into the external damage detection model, wherein the external damage detection model includes a feature extraction network, a feature fusion network and a detection network;

[0038] The power distribution line feature map acquisition module is used to perform feature extraction operations on the power distribution line image to be analyzed using the feature extraction network, and output multiple power distribution line feature maps in groups. The feature extraction network includes a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected in sequence.

[0039] The fusion feature map acquisition module is used to perform feature fusion operations on multiple power distribution line feature maps through the feature fusion network to generate a fusion feature map.

[0040] The external damage analysis module is used to perform external damage detection on the fused feature map using the detection network, and to determine whether the power distribution line image to be analyzed has external damage risk.

[0041] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the power distribution line external damage detection method as described in any of the first aspects of the present invention.

[0042] As can be seen from the above technical solutions, the present invention has the following advantages:

[0043] Images of the power distribution lines to be analyzed are acquired through edge devices. These images are then input into an external damage detection model composed of a feature extraction network, a feature fusion network, and a detection network. The feature extraction network, consisting of sequentially connected convolutional groups at first, second, third, fourth, and fifth scales, extracts features from the power distribution lines, outputting multiple feature maps for each group. The feature fusion network then fuses these feature maps to generate a fused feature map. Finally, the detection network performs external damage detection on this fused feature map to determine if the power distribution line image poses an external damage risk. This approach addresses the technical problem that current deep learning-based target detection methods for power distribution line detection focus primarily on improving the accuracy of external damage detection, resulting in slower detection speeds. Furthermore, due to the large number of power distribution lines to be analyzed, this approach fails to meet the real-time detection requirements of practical applications. This invention reduces the computational load of feature extraction by setting multiple selection modules of different sizes on the basis of the original model, and replacing or removing the size of the convolutional layer that has no impact. This enables the external damage detection model to quickly and accurately detect images of power distribution lines with external damage risks. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart of the steps of a method for detecting external damage to power distribution lines provided in Embodiment 1 of the present invention;

[0046] Figure 2 This is a flowchart of the steps of a method for detecting external damage to power distribution lines provided in Embodiment 2 of the present invention;

[0047] Figure 3 This is a schematic diagram of the external damage detection model provided in Embodiment 2 of the present invention;

[0048] Figure 4 This is a structural block diagram of a power distribution line external damage detection device provided in Embodiment 3 of the present invention. Detailed Implementation

[0049] This invention provides a method, apparatus, and device for detecting external damage to power distribution lines. It addresses the technical problem that current deep learning-based target detection methods focus more on improving the accuracy of external damage detection when building models for power distribution line detection, resulting in slow model detection speed. Furthermore, due to the large number of power distribution lines to be analyzed, these methods cannot meet the real-time detection requirements of practical applications.

[0050] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0051] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of a method for detecting external damage to power distribution lines according to Embodiment 1 of the present invention.

[0052] This invention provides a method for detecting external damage to power distribution lines, comprising:

[0053] Step 101: Obtain the image of the power distribution line to be analyzed and input the image of the power distribution line to be analyzed into the external damage detection model, which includes a feature extraction network, a feature fusion network and a detection network.

[0054] The images of the power distribution lines to be analyzed refer to images of the power distribution lines to be analyzed collected by drones, cameras installed on the power distribution lines, or other edge devices.

[0055] In this embodiment of the invention, when the image of the power distribution line to be analyzed is received from the acquisition device, the image of the power distribution line to be analyzed is input into the external damage detection model.

[0056] It should be noted that the external damage detection model is a pre-trained model used to analyze input images of power distribution lines and output corresponding confidence scores. It is an improvement on the original CNN network model, specifically by introducing bounding box modules of different scales to extract features from the images of the power distribution lines to be analyzed. The extracted feature maps at different scales are then fused to obtain a fused feature map. The detection network then analyzes the fused feature map, thereby reducing the false negative rate for small targets. Simultaneously, improvements to the convolutional layer structure in the feature extraction and feature fusion networks reduce the computational cost during feature extraction and increase the receptive field, ensuring that the output feature map size is consistent with the input feature map size. This achieves a balance between detection speed and accuracy in the external damage detection model.

[0057] Step 102: Use a feature extraction network to perform feature extraction on the power distribution line image to be analyzed, and output multiple power distribution line feature maps in groups. The feature extraction network includes a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected in sequence.

[0058] In this embodiment of the invention, a feature extraction network is used to perform feature extraction operations on the power distribution line image to be analyzed, thereby determining multiple power distribution line feature maps.

[0059] It should be noted that the feature extraction network includes a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected in sequence. The third-scale convolutional group uses a third extraction branch, which includes a 3×1 standard convolutional layer, a 1×3 standard convolutional layer, a LeakyReLU activation layer, and a 3×3 pooling layer connected in sequence. This reduces the computational cost of feature extraction while ensuring that the output feature map size is consistent with the input feature map size, thereby increasing the receptive field and improving the accuracy of model detection.

[0060] Step 103: Perform feature fusion operation on multiple power distribution line feature maps through a feature fusion network to generate a fused feature map.

[0061] In this embodiment of the invention, a feature fusion network is used, which has multiple box selection modules to perform feature fusion operations on multiple power distribution line feature maps to generate a fused feature map.

[0062] It should be noted that the feature fusion network selects appropriate feature maps from multiple scales for feature fusion, thereby avoiding the false negative rate of external damage features, enabling the model to adapt to different types of external damage risk detection, and increasing the accuracy of external damage risk detection.

[0063] Step 104: Use a detection network to perform external damage detection on the fused feature map to determine whether there is an external damage risk in the power distribution line image to be analyzed.

[0064] In this embodiment of the invention, a detection network is used to perform external damage detection on the fused feature map and output the corresponding external damage risk detection result.

[0065] In this embodiment of the invention, an image of the power distribution line to be analyzed is acquired through an edge device. This image is then input into an external damage detection model composed of a feature extraction network, a feature fusion network, and a detection network. The feature extraction network, consisting of a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected sequentially, performs feature extraction on the image, outputting multiple power distribution line feature maps group by group. The feature fusion network then performs feature fusion on these multiple feature maps to generate a fused feature map. Finally, the detection network performs external damage detection on the fused feature map to determine whether the power distribution line image to be analyzed has an external damage risk. This addresses the technical problem that current deep learning-based target detection methods for power distribution line detection focus primarily on improving the accuracy of external damage detection, resulting in slow detection speeds. Furthermore, due to the large number of power distribution lines to be analyzed, this method cannot meet the real-time detection requirements of practical applications. This invention reduces the computational load of feature extraction by setting multiple selection modules of different sizes on the basis of the original model, and replacing or removing the size of the convolutional layer that has no impact. This enables the external damage detection model to quickly and accurately detect images of power distribution lines with external damage risks.

[0066] Please see Figure 2 , Figure 2 This is a flowchart illustrating the steps of a method for detecting external damage to power distribution lines according to Embodiment 2 of the present invention.

[0067] This invention provides a method for detecting external damage to power distribution lines, comprising:

[0068] Step 201: Obtain the image of the power distribution line to be analyzed and input the image of the power distribution line to be analyzed into the external damage detection model, which includes a feature extraction network, a feature fusion network and a detection network.

[0069] In this embodiment of the invention, an image of the power distribution line to be analyzed is acquired through an edge device, and the image of the power distribution line to be analyzed is input into a pre-trained external damage detection model.

[0070] It should be noted that, for reference Figure 3As shown, CR is a convolutional layer with activation function layer, C is a convolutional layer, P is a pooling layer, LR is a LeakyReLU activation layer, R is an activation function layer, RE is a deformation layer, RPN1 is the first bounding box selection module, RPN2 is the second bounding box selection module, S is a classification function layer, Proposal is a candidate bounding box layer, F is a fully connected layer with activation function layer, and ROIpooling is a region of interest pooling layer. The external damage detection model includes a feature extraction network, a feature fusion network, and a detection network. The feature fusion network selects four feature maps of different sizes for feature fusion to generate a fused feature map. Then, it is input into two detection branches composed of a fully connected layer, an activation function layer, and a classification function layer to perform external damage detection, generating the confidence score and external damage risk feature map corresponding to the image of the power distribution line to be analyzed.

[0071] Step 202: Use a feature extraction network to perform feature extraction on the power distribution line image to be analyzed, and output multiple power distribution line feature maps in groups. The feature extraction network includes a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected in sequence.

[0072] Further, step 202 includes the following sub-steps:

[0073] S11. The first scale convolution group is used to extract features from the power distribution line image to be analyzed, and the first power distribution line feature map is generated. The first scale convolution group includes a first extraction branch, a second extraction branch and a feature fusion layer connected in sequence.

[0074] Furthermore, S11 includes the following sub-steps:

[0075] S111. The first extraction branch is used to extract features from the power distribution line image to be analyzed, and a first extracted feature map is generated. The first extraction branch includes three first extraction modules connected in sequence.

[0076] In this embodiment of the invention, a first extraction branch is formed by connecting three first extraction modules in sequence to extract features from the power distribution line image to be analyzed, thereby generating a first extracted feature map.

[0077] It is worth mentioning that, in order to improve the detection speed of the external damage detection model, the first extraction module can be a 3×3 standard convolutional layer, a 3×3 standard convolutional layer and a DY-ReLU activation function layer connected in sequence.

[0078] S112. The first extracted feature map is extracted through the second extraction branch to generate the second extracted feature map, wherein the second extraction branch includes two first extraction models and a 2×2 pooling layer connected in sequence.

[0079] In this embodiment of the invention, a second extraction feature map is generated by performing feature extraction on the first extraction feature map by forming a second extraction branch consisting of two first extraction models and a 2×2 pooling layer (with a stride of 1).

[0080] S113. Perform feature fusion operation on the first extracted feature map and the second extracted feature map through the feature fusion layer to generate the first power distribution line feature map.

[0081] In this embodiment of the invention, a feature fusion layer is used to fuse the first extracted feature map and the second extracted feature map to generate a first power distribution line feature map.

[0082] The first extraction module includes a 3×3 standard convolutional layer and a DY-ReLU activation function layer connected in sequence.

[0083] S12. Extract features from the first power distribution line feature map using a second-scale convolution group to generate a second power distribution line feature map, wherein the second-scale convolution group includes a second extraction branch.

[0084] S13. The second power distribution line feature map is extracted by the third-scale convolution group to generate the third power distribution line feature map. The third-scale convolution group includes a third extraction branch and a feature fusion layer connected in sequence.

[0085] Furthermore, S13 includes the following sub-steps:

[0086] S131. The third extraction branch is used to extract features from the feature map of the second power distribution line to generate the third extracted feature map. The third extraction branch includes a 3×1 standard convolutional layer, a 1×3 standard convolutional layer, a LeakyReLU activation layer and a 3×3 pooling layer connected in sequence.

[0087] In this embodiment of the invention, a third extraction branch is formed by sequentially connecting a 3×1 standard convolutional layer, a 1×3 standard convolutional layer, a LeakyReLU activation layer, and a 3×3 pooling layer to extract features from the feature map of the second power distribution line, thereby generating a third extracted feature map.

[0088] It is worth mentioning that, depending on the type of external damage risk to the power distribution line, the third extraction branch includes, but is not limited to, 2×1 standard convolutional layers, 1×2 standard convolutional layers, LeakyReLU activation layers and 2×2 pooling layers, 1×1 standard convolutional layers, 1×1 standard convolutional layers, LeakyReLU activation layers and 2×2 pooling layers, 5×1 standard convolutional layers, 1×5 standard convolutional layers, LeakyReLU activation layers and 3×3 pooling layers.

[0089] S132. Perform feature fusion operation on the third extracted feature map and the second power distribution line feature map through the feature fusion layer to generate the third power distribution line feature map.

[0090] In this embodiment of the invention, the third extracted feature map and the second power distribution line feature map are fused by a feature fusion layer to generate the third power distribution line feature map.

[0091] S14. The feature map of the third power distribution line is extracted by the fourth-scale convolution group to generate the feature map of the fourth power distribution line. The fourth-scale convolution group includes a 3×3 standard convolutional layer, a DY-ReLU activation function layer and a second extraction branch connected in sequence.

[0092] In this embodiment of the invention, features are extracted from the feature map of the third power distribution line by forming a fourth-scale convolution group consisting of a 3×3 standard convolutional layer, a DY-ReLU activation function layer, and a second extraction branch.

[0093] S15. Extract features from the feature map of the fourth power distribution line using the fifth-scale convolution group to generate the feature map of the fifth power distribution line, wherein the fifth-scale convolution group includes the first extraction branch.

[0094] In this embodiment of the invention, the feature map of the fourth power distribution line is extracted by the first extraction branch to generate the feature map of the fifth power distribution line.

[0095] Step 203: Perform feature fusion operation on multiple power distribution line feature maps through a feature fusion network to generate a fused feature map.

[0096] Furthermore, the feature fusion network includes a first bounding box selection module, a second bounding box selection module, a feature fusion layer, and a region of interest pooling layer. Step 203 includes the following sub-steps:

[0097] S21. The first box selection module is used to extract features from the feature map of the first power distribution line to generate the first box selection feature map.

[0098] It is worth mentioning that the first box selection module includes a 5×5 standard convolutional layer, a DY-ReLU activation function layer, a 5×5 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer connected in sequence.

[0099] In this embodiment of the invention, a first box selection module is used to extract features from the feature map of the first power distribution line and generate a first box selection feature map. The module consists of a 5×5 standard convolutional layer, a DY-ReLU activation function layer, a 5×5 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer and a candidate box layer connected in sequence.

[0100] S22. The second box selection module extracts features from the feature map of the second power distribution line to generate the second box selection feature map.

[0101] It is worth mentioning that the second box selection module includes a 3×3 standard convolutional layer, a DY-ReLU activation function layer, a 3×3 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer connected in sequence.

[0102] In this embodiment of the invention, a second box selection module is formed by sequentially connecting a 3×3 standard convolutional layer, a DY-ReLU activation function layer, a 3×3 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer to extract features from the feature map of the second power distribution line and generate a second box selection feature map.

[0103] S23. Perform feature fusion operation on the first bounding box feature map and the second bounding box feature map through the feature fusion layer to generate the feature map of interest.

[0104] In this embodiment of the invention, a feature fusion layer is used to fuse the first bounding box feature map and the second bounding box feature map to generate a feature map of interest.

[0105] S24. Perform feature fusion operation on the feature map of interest, the feature map of the third power distribution line, the feature map of the fourth power distribution line, and the feature map of the fifth power distribution line through the region of interest pooling layer to generate a fused feature map.

[0106] In this embodiment of the invention, a region of interest pooling layer is used to perform feature fusion operations on the feature map of interest, the third power distribution line feature map, the fourth power distribution line feature map, and the fifth power distribution line feature map to generate a fused feature map.

[0107] It should be noted that the feature map of interest, the third power distribution line feature map, the fourth power distribution line feature map, and the fifth power distribution line feature map are all feature maps of different scales.

[0108] Step 204: Use a detection network to perform external damage detection on the fused feature map and generate a confidence score.

[0109] In this embodiment of the invention, a detection network is used to perform external damage detection on the fused feature map to generate a confidence score.

[0110] It should be noted that the detection network includes two detection branches, each consisting of a fully connected layer, an activation function layer, and a classification function layer connected in sequence. The fused feature map is input into the two detection branches to generate a confidence score and an external damage risk feature map.

[0111] Step 205: Determine whether the confidence score is less than the preset score threshold.

[0112] The score threshold refers to the critical value at which the power distribution line under analysis is at risk of external damage, and its value is 0.85.

[0113] In this embodiment of the invention, it is determined whether the confidence score is less than 0.85.

[0114] It should be noted that Table 1 shows the detection accuracy under different score thresholds.

[0115] Score threshold accuracy False positive rate False negative rate 0.6 0.965 0.025 0.01 0.7 0.975 0.013 0.012 0.8 0.976 0.007 0.017 0.85 0.979 0.004 0.017 0.9 0.971 0.002 0.027

[0116] Table 1

[0117] During external damage detection, a score threshold is selected, generally not lower than 0.6. If the score threshold is lower than 0.6, some locations that are not external damage may be falsely detected due to the background being too similar to external damage. Therefore, the accuracy, false positive rate, and false negative rate of detection with score thresholds of 0.6, 0.7, 0.8, 0.85, and 0.9 were statistically analyzed to determine the optimal score threshold of 0.85.

[0118] Step 206: If the confidence score is greater than or equal to the score threshold, it is determined that the power distribution line image to be analyzed has an external damage risk.

[0119] In this embodiment of the invention, if the confidence score is greater than or equal to 0.85, it is determined that the power distribution line image to be analyzed has an external damage risk.

[0120] It should be noted that if the confidence score is greater than or equal to 0.85, the detection network will select areas with external damage risk in the output external damage risk feature map. This is helpful for establishing a complete black spot file, formulating operation and maintenance strategies, recording the location, construction scale, and distribution of black spots in detail, and discovering potential external damage hazards in advance to provide data support for the implementation of subsequent control measures.

[0121] Step 207: If the confidence score is less than the score threshold, it is determined that there is no risk of external damage to the power distribution line image to be analyzed.

[0122] In this embodiment of the invention, if the confidence score is less than 0.85, it is determined that the power distribution line image to be analyzed does not have the risk of external damage.

[0123] In this embodiment of the invention, an image of the power distribution line to be analyzed is acquired through an edge device. This image is then input into an external damage detection model composed of a feature extraction network, a feature fusion network, and a detection network. The feature extraction network, consisting of a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected sequentially, performs feature extraction on the image, outputting multiple power distribution line feature maps group by group. The feature fusion network then performs feature fusion on these multiple feature maps to generate a fused feature map. Finally, the detection network performs external damage detection on the fused feature map to determine whether the power distribution line image to be analyzed has an external damage risk. This addresses the technical problem that current deep learning-based target detection methods for power distribution line detection focus primarily on improving the accuracy of external damage detection, resulting in slow detection speeds. Furthermore, due to the large number of power distribution lines to be analyzed, this method cannot meet the real-time detection requirements of practical applications. This invention reduces the computational load of feature extraction by setting multiple selection modules of different sizes on the basis of the original model, and replacing or removing the size of the convolutional layer that has no impact. This enables the external damage detection model to quickly and accurately detect images of power distribution lines with external damage risks.

[0124] Please see Figure 4 , Figure 4 This is a structural block diagram of a power distribution line external damage detection device provided in Embodiment 3 of the present invention.

[0125] A device for detecting external damage to power distribution lines, comprising:

[0126] The image acquisition module 401 is used to acquire the image of the power distribution line to be analyzed and input the image of the power distribution line to be analyzed into the external damage detection model, wherein the external damage detection model includes a feature extraction network, a feature fusion network and a detection network.

[0127] The power distribution line feature map acquisition module 402 is used to perform feature extraction operations on the power distribution line image to be analyzed using a feature extraction network, and output multiple power distribution line feature maps in groups. The feature extraction network includes a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected in sequence.

[0128] The fusion feature map acquisition module 403 is used to perform feature fusion operations on multiple power distribution line feature maps through a feature fusion network to generate a fusion feature map.

[0129] The external damage analysis module 404 is used to perform external damage detection on the fused feature map using a detection network to determine whether there is an external damage risk in the power distribution line image to be analyzed.

[0130] Furthermore, the power distribution line feature map acquisition module 402 includes:

[0131] The first extraction submodule is used to extract features from the power distribution line image to be analyzed using a first-scale convolution group to generate a first power distribution line feature map, wherein the first-scale convolution group includes a first extraction branch, a second extraction branch and a feature fusion layer connected in sequence.

[0132] The second extraction submodule is used to extract features from the first power distribution line feature map through the second scale convolution group to generate the second power distribution line feature map, wherein the second scale convolution group includes the second extraction branch;

[0133] The third extraction submodule is used to extract features from the second power distribution line feature map through the third-scale convolution group to generate the third power distribution line feature map. The third-scale convolution group includes a third extraction branch and a feature fusion layer connected in sequence.

[0134] The fourth extraction submodule is used to extract features from the feature map of the third power distribution line through the fourth-scale convolution group to generate the feature map of the fourth power distribution line. The fourth-scale convolution group includes a 3×3 standard convolutional layer, a DY-ReLU activation function layer and a second extraction branch connected in sequence.

[0135] The fifth extraction submodule is used to extract features from the feature map of the fourth power distribution line through the fifth-scale convolution group to generate the feature map of the fifth power distribution line, wherein the fifth-scale convolution group includes the first extraction branch.

[0136] Furthermore, the first extraction submodule includes:

[0137] The first extraction unit is used to extract features from the power distribution line image to be analyzed using the first extraction branch to generate a first extracted feature map, wherein the first extraction branch includes three first extraction modules connected in sequence.

[0138] The second extraction unit is used to extract features from the first extracted feature map through the second extraction branch to generate the second extracted feature map. The second extraction branch includes two first extraction models and a 2×2 pooling layer connected in sequence.

[0139] The first fusion unit is used to perform feature fusion operation on the first extracted feature map and the second extracted feature map through the feature fusion layer to generate the first power distribution line feature map.

[0140] The first extraction module includes a 3×3 standard convolutional layer and a DY-ReLU activation function layer connected in sequence.

[0141] Furthermore, the third extraction submodule includes:

[0142] The third extraction unit is used to extract features from the feature map of the second power distribution line using the third extraction branch, and generate the third extracted feature map. The third extraction branch includes a 3×1 standard convolutional layer, a 1×3 standard convolutional layer, a LeakyReLU activation layer and a 3×3 pooling layer connected in sequence.

[0143] The second fusion unit is used to perform feature fusion operations on the third extracted feature map and the second power distribution line feature map through the feature fusion layer to generate the third power distribution line feature map.

[0144] Furthermore, the feature fusion network includes a first bounding box selection module, a second bounding box selection module, a feature fusion layer, and a region of interest pooling layer. The fused feature map acquisition module 403 includes:

[0145] The first bounding box feature map acquisition submodule is used to extract features from the first power distribution line feature map using the first bounding box module to generate the first bounding box feature map.

[0146] The second bounding box feature map acquisition submodule is used to extract features from the second power distribution line feature map through the second bounding box module to generate the second bounding box feature map.

[0147] The feature map of interest acquisition submodule is used to perform feature fusion operation on the first bounding box feature map and the second bounding box feature map through the feature fusion layer to generate the feature map of interest.

[0148] The feature map fusion acquisition submodule is used to perform feature fusion operations on the feature map of interest, the third power distribution line feature map, the fourth power distribution line feature map, and the fifth power distribution line feature map through the region of interest pooling layer to generate a fused feature map.

[0149] Furthermore, the first box selection module includes a 5×5 standard convolutional layer, a DY-ReLU activation function layer, a 5×5 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer connected in sequence.

[0150] Furthermore, the second box selection module includes a 3×3 standard convolutional layer, a DY-ReLU activation function layer, a 3×3 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer connected in sequence.

[0151] Furthermore, the external damage analysis module 404 includes:

[0152] The confidence score acquisition submodule is used to perform external damage detection on the fused feature map using a detection network and generate a confidence score.

[0153] The judgment and analysis submodule is used to determine whether the confidence score is less than a preset score threshold;

[0154] If the confidence score is greater than or equal to the score threshold, the power distribution line image to be analyzed is determined to have an external damage risk.

[0155] If the confidence score is less than the score threshold, it is determined that there is no risk of external damage to the power distribution line image to be analyzed.

[0156] Embodiment 4 of the present invention also provides an electronic device, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the power distribution line external damage detection method as described in any of the above embodiments.

[0157] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0158] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0159] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0160] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0161] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting external damage to power distribution lines, characterized in that, include: Acquire an image of the power distribution line to be analyzed, and input the image of the power distribution line to be analyzed into an external damage detection model, wherein the external damage detection model includes a feature extraction network, a feature fusion network and a detection network; The feature extraction network is used to perform feature extraction on the power distribution line image to be analyzed, and outputs multiple power distribution line feature maps in groups. The feature extraction network includes a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected in sequence. The feature fusion network performs feature fusion operations on multiple power distribution line feature maps to generate a fused feature map; the feature fusion network includes a first bounding box module, a second bounding box module, a feature fusion layer, and a region of interest pooling layer. The detection network is used to perform external damage detection on the fused feature map to determine whether there is an external damage risk in the power distribution line image to be analyzed; The step of using the feature extraction network to perform feature extraction on the power distribution line image to be analyzed, and outputting multiple power distribution line feature maps in groups, includes: The first scale convolution group is used to extract features from the power distribution line image to be analyzed, generating a first power distribution line feature map. The first scale convolution group includes a first extraction branch, a second extraction branch and a feature fusion layer connected in sequence. The first extraction branch includes three first extraction modules connected in sequence, and the second extraction branch includes two first extraction modules and a 2×2 pooling layer connected in sequence. The first extraction module includes a 3×3 standard convolutional layer and a DY-ReLU activation function layer connected in sequence; The first power distribution line feature map is used to extract features by a second-scale convolution group to generate a second power distribution line feature map, wherein the second-scale convolution group includes a second extraction branch. The second power distribution line feature map is extracted by the third-scale convolution group to generate the third power distribution line feature map. The third-scale convolution group includes a third extraction branch and a feature fusion layer connected in sequence. The third extraction branch includes a 3×1 standard convolutional layer, a 1×3 standard convolutional layer, a LeakyReLU activation layer and a 3×3 pooling layer connected in sequence. The fourth-scale convolutional group is used to extract features from the third power distribution line feature map to generate the fourth power distribution line feature map. The fourth-scale convolutional group includes a 3×3 standard convolutional layer, a DY-ReLU activation function layer and a second extraction branch connected in sequence. The fifth distribution line feature map is generated by extracting features from the fourth distribution line feature map using a fifth-scale convolution group, wherein the fifth-scale convolution group includes a first extraction branch. The step of performing feature fusion operations on multiple power distribution line feature maps through the feature fusion network to generate a fused feature map includes: The first box selection module is used to extract features from the feature map of the first power distribution line to generate a first box selection feature map. The second box selection module extracts features from the feature map of the second power distribution line to generate a second box selection feature map. The first bounding box feature map and the second bounding box feature map are fused by a feature fusion layer to generate a feature map of interest. The region of interest pooling layer is used to perform feature fusion operations on the feature map of interest, the feature map of the third power distribution line, the feature map of the fourth power distribution line, and the feature map of the fifth power distribution line to generate a fused feature map.

2. The method for detecting external damage to power distribution lines according to claim 1, characterized in that, The step of extracting features from the power distribution line image to be analyzed using a first-scale convolutional group to generate a first power distribution line feature map includes: The first extraction branch is used to extract features from the power distribution line image to be analyzed, generating a first extracted feature map; The second extraction branch extracts features from the first extracted feature map to generate a second extracted feature map; The first extracted feature map and the second extracted feature map are fused by the feature fusion layer to generate the first power distribution line feature map.

3. The method for detecting external damage to power distribution lines according to claim 1, characterized in that, The step of extracting features from the second power distribution line feature map using a third-scale convolutional group to generate a third power distribution line feature map includes: The third extraction branch is used to extract features from the feature map of the second power distribution line to generate the third extracted feature map; The third power distribution line feature map is generated by performing feature fusion operation on the third extracted feature map and the second power distribution line feature map through the feature fusion layer.

4. The method for detecting external damage to power distribution lines according to claim 1, characterized in that, The first box selection module includes a 5×5 standard convolutional layer, a DY-ReLU activation function layer, a 5×5 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer connected in sequence.

5. The method for detecting external damage to power distribution lines according to claim 1, characterized in that, The second box selection module includes a 3×3 standard convolutional layer, a DY-ReLU activation function layer, a 3×3 standard convolutional layer, a deformation layer, a classification function layer, a deformation layer, and a candidate box layer connected in sequence.

6. The method for detecting external damage to power distribution lines according to claim 1, characterized in that, The step of using the detection network to perform external damage detection on the fused feature map and determining whether the power distribution line image to be analyzed has external damage risk includes: The detection network is used to perform external damage detection on the fused feature map to generate a confidence score; Determine whether the confidence score is less than a preset score threshold; If the confidence score is greater than or equal to the score threshold, it is determined that the power distribution line image to be analyzed has an external damage risk. If the confidence score is less than the score threshold, it is determined that the power distribution line image to be analyzed does not have the risk of external damage.

7. A device for detecting external damage to power distribution lines, characterized in that, include: The image acquisition module is used to acquire the image of the power distribution line to be analyzed and input the image of the power distribution line to be analyzed into the external damage detection model, wherein the external damage detection model includes a feature extraction network, a feature fusion network and a detection network; The power distribution line feature map acquisition module is used to perform feature extraction operations on the power distribution line image to be analyzed using the feature extraction network, and output multiple power distribution line feature maps in groups. The feature extraction network includes a first-scale convolutional group, a second-scale convolutional group, a third-scale convolutional group, a fourth-scale convolutional group, and a fifth-scale convolutional group connected in sequence. The feature map acquisition module is used to perform feature fusion operations on multiple power distribution line feature maps through the feature fusion network to generate a fused feature map; the feature fusion network includes a first bounding box module, a second bounding box module, a feature fusion layer, and a region of interest pooling layer. The external damage analysis module is used to perform external damage detection on the fused feature map using the detection network, and to determine whether the power distribution line image to be analyzed has external damage risk. The power distribution line feature map acquisition module includes: The first extraction submodule is used to extract features from the power distribution line image to be analyzed using a first-scale convolution group to generate a first power distribution line feature map. The first-scale convolution group includes a first extraction branch, a second extraction branch and a feature fusion layer connected in sequence. The first extraction branch includes three first extraction modules connected in sequence. The second extraction branch includes two first extraction modules and a 2×2 pooling layer connected in sequence. The first extraction module includes a 3×3 standard convolutional layer and a DY-ReLU activation function layer connected in sequence; The second extraction submodule is used to extract features from the first power distribution line feature map through a second-scale convolution group to generate a second power distribution line feature map, wherein the second-scale convolution group includes a second extraction branch. The third extraction submodule is used to extract features from the second power distribution line feature map through the third scale convolution group to generate the third power distribution line feature map. The third scale convolution group includes a third extraction branch and a feature fusion layer connected in sequence. The third extraction branch includes a 3×1 standard convolution layer, a 1×3 standard convolution layer, a LeakyReLU activation layer and a 3×3 pooling layer connected in sequence. The fourth extraction submodule is used to extract features from the third power distribution line feature map through a fourth-scale convolution group to generate a fourth power distribution line feature map, wherein the fourth-scale convolution group includes a 3×3 standard convolution layer, a DY-ReLU activation function layer and a second extraction branch connected in sequence. The fifth extraction submodule is used to extract features from the fourth power distribution line feature map through a fifth-scale convolution group to generate a fifth power distribution line feature map, wherein the fifth-scale convolution group includes a first extraction branch. The fused feature map acquisition module includes: The first bounding box feature map acquisition submodule is used to extract features from the first power distribution line feature map using the first bounding box module to generate the first bounding box feature map. The second bounding box feature map acquisition submodule is used to extract features from the second power distribution line feature map through the second bounding box module to generate the second bounding box feature map. The feature map of interest acquisition submodule is used to perform feature fusion operation on the first bounding box feature map and the second bounding box feature map through the feature fusion layer to generate the feature map of interest; The feature map fusion acquisition submodule is used to perform feature fusion operations on the feature map of interest, the third power distribution line feature map, the fourth power distribution line feature map, and the fifth power distribution line feature map through the region of interest pooling layer to generate a fused feature map.

8. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the power distribution line external damage detection method as described in any one of claims 1-6.

Images