Power distribution line ground wire broken strand defect identification method, device, equipment and medium
By improving the binary classification recognition module and feature fusion mechanism in the YOLOv11 target detection network, the false detection problem of UAVs when identifying small target defects in power distribution lines was solved, thereby improving the recognition effect and system reliability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAYAN ZHIKE (HANGZHOU) INFORMATION TECH CO LTD
- Filing Date
- 2026-04-28
- Publication Date
- 2026-05-29
AI Technical Summary
When using drone images to identify small target defects in power distribution lines, such as broken conductors, existing algorithms struggle to achieve effective identification and are prone to false detections, increasing the cost of manual review.
A binary classification recognition module and an adaptive feature enhancement mechanism are introduced, and combined with feature fusion, a multi-scale feature map is processed through an improved YOLOv11 object detection network, including a backbone network, a neck network, and a head network, to perform binary classification prediction, adaptive enhancement, and feature fusion of the feature maps.
It significantly improves the feature representation capability and positioning accuracy of small-sized defects, effectively suppresses false detections caused by background interference, and enhances the reliability and practicality of the automated inspection system for power distribution lines.
Smart Images
Figure CN122115448A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of target detection technology, and in particular to a method, apparatus, equipment and medium for identifying broken strand defects in power distribution line conductors and ground wires. Background Technology
[0002] In power line inspections, drones can assist inspectors in quickly collecting power line data, and then use optimized target detection algorithms to locate and characterize line defects, bringing convenience to subsequent maintenance.
[0003] However, drone images are usually high-resolution, which can quickly locate and identify large-scale defects in power distribution lines. But for small target defects in power distribution lines, such as broken conductors, existing algorithms usually cannot achieve good defect identification and will produce a large number of false detections, increasing the manual review costs for staff. Summary of the Invention
[0004] In view of this, the purpose of this invention is to overcome the shortcomings of the prior art and provide a method, device, equipment and medium for identifying conductor and ground wire breakage defects in power distribution lines. This invention introduces a binary classification identification module and an adaptive feature enhancement mechanism, and combines feature fusion to significantly improve the feature expression capability and positioning accuracy of small-sized defects, effectively suppress false detection caused by background interference, and improve the reliability and practicality of the automated inspection system for power distribution lines.
[0005] This invention provides the following technical solution: In a first aspect, the present invention proposes a method for identifying broken strand defects in the conductors and ground wires of power distribution lines, comprising: The image of the power distribution line collected by the UAV inspection is acquired, and the image of the power distribution line is input into the improved YOLOv11 target detection network. The improved YOLOv11 target detection network includes a backbone network, a neck network and a head network. The neck network includes a binary classification recognition module and a feature fusion module. Multi-scale feature maps are extracted from the power distribution line images using the backbone network; The binary classification recognition module performs binary classification prediction on the multi-scale feature map to obtain the features. Figure 2 Classification results; According to the features Figure 2 The classification results are adaptively enhanced to obtain a reset feature map; The reset feature map is downsampled by the feature fusion module, and feature fusion is performed on the downsampled reset feature map and the multi-scale feature map to obtain the target fused feature map; The target fusion feature map is identified by the head network to obtain the target detection and identification result of broken strands in the conductor and ground wire of the power distribution line.
[0006] In one embodiment, the multi-scale feature map includes a first feature map, a second feature map, and a third feature map. The binary classification recognition module performs binary classification prediction on the multi-scale feature map to obtain the feature map. Figure 2 The classification results include: The binary classification recognition module upsamples the second feature map and the third feature map respectively to obtain the upsampled second feature map and the upsampled third feature map. The first feature map, the upsampled second feature map, and the upsampled third feature map are concatenated to obtain a concatenated feature map. The stitched feature map is cropped into multiple sub-images, and binary classification prediction is performed on each sub-image to obtain the binary classification prediction result of each sub-image, which is used as the feature. Figure 2 Classification results.
[0007] In one embodiment, the step according to the feature Figure 2 The classification results are adaptively enhanced to obtain a reset feature map, including: The feature mean of each subgraph is determined based on the binary classification prediction results of each subgraph; A new feature map for each subgraph is obtained based on the average feature value of each subgraph, and is used as the reset feature map.
[0008] In one embodiment, the downsampled reset feature map includes multiple downsampled feature maps. The step of downsampling the reset feature map using the feature fusion module and performing feature fusion based on the downsampled reset feature map and the multi-scale feature map to obtain a target fused feature map includes: The reset feature map is downsampled based on multiple preset channel numbers to obtain downsampled feature maps corresponding to each preset channel number. Calculate the sum of pixels in each of the downsampled feature maps to obtain the sum of pixels in the feature map corresponding to each downsampled feature map; Calculate the sum of all the pixel sums in the feature map to obtain the pixel sum of the downsampled reset feature map; Calculate the quotient of each feature map pixel sum with the sum of the pixel sum and the sum of the values to obtain the first feature map weights corresponding to each downsampled feature map; Based on the weights of each of the first feature maps, feature fusion is performed on the downsampled reset feature map and the multi-scale feature map to obtain the target fused feature map.
[0009] In one embodiment, the target fusion feature map includes multiple fusion feature maps, and the step of fusing the downsampled reset feature map and the multi-scale feature map according to the weights of each of the first feature maps to obtain the target fusion feature map includes: For each of the first feature map weights, calculate the second feature map weights corresponding to the first feature map weights based on the first feature map weights. The downsampled reset feature map and the multi-scale feature map are weighted and summed according to the first feature map weight and its corresponding second feature map weight to obtain the fused feature map corresponding to the first feature map weight.
[0010] In one embodiment, the formula for calculating the new feature map is: In the formula, For the first New feature maps of each subgraph For the first The feature mean of each subplot.
[0011] In one embodiment, the formula for calculating the weight of the second feature map is: In the formula, For the first The weights of the second feature map corresponding to the weights of the first feature map. For the first The weights of the first feature map.
[0012] Secondly, the present invention proposes a device for identifying broken strand defects in the conductors and ground wires of power distribution lines, comprising: The input module is used to acquire images of power distribution lines collected by UAV inspections and input the images of power distribution lines into an improved YOLOv11 target detection network. The improved YOLOv11 target detection network includes a backbone network, a neck network, and a head network. The neck network includes a binary classification recognition module and a feature fusion module. The extraction module is used to extract multi-scale feature maps from the power distribution line image through the backbone network; The prediction module is used to perform binary classification prediction on the multi-scale feature map through the binary classification recognition module to obtain features. Figure 2 Classification results; Enhancement module, used to enhance based on the features Figure 2 The classification results are adaptively enhanced to obtain a reset feature map; The fusion module is used to downsample the reset feature map through the feature fusion module, and perform feature fusion based on the downsampled reset feature map and the multi-scale feature map to obtain a target fused feature map; The identification module is used to identify the target fusion feature map through the head network to obtain the target detection and identification result of the broken strand of the conductor and ground wire of the power distribution line.
[0013] Thirdly, the present invention proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the method for identifying broken strands in power distribution lines as described in the first aspect.
[0014] Fourthly, the present invention proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for identifying broken strand defects in power distribution lines as described in the first aspect.
[0015] This invention discloses a method, apparatus, equipment, and medium for identifying conductor and ground wire breakage defects in power distribution lines. The method involves acquiring images of power distribution lines collected by unmanned aerial vehicle (UAV) inspections, and inputting these images into an improved YOLOv11 target detection network. The improved YOLOv11 target detection network includes a backbone network, a neck network, and a head network. The neck network includes a binary classification module and a feature fusion module. Multi-scale feature maps are extracted from the power distribution line images through the backbone network. The binary classification module performs binary classification prediction on the multi-scale feature maps to obtain the feature maps. Figure 2 The classification results are used to adaptively enhance the binary classification results, resulting in a reset feature map. The reset feature map is then downsampled using a feature fusion module, and feature fusion is performed between the downsampled reset feature map and the multi-scale feature map to obtain a target fusion feature map. The head network then identifies the target fusion feature map to obtain the target detection and recognition result for broken conductor / ground wire strands in the power distribution line. Thus, based on the standard target detection network structure, a binary classification module, adaptive feature enhancement, and dynamic fusion mechanism are introduced into the neck network to perform pixel-level foreground-background binary classification on the multi-scale feature map. Furthermore, a reset feature map is generated based on the binary classification results, and feature fusion is achieved, effectively improving the identification effect of broken conductor / ground wire strand defects while significantly reducing false alarms and improving the overall recognition performance of the model network. Attached Figure Description
[0016] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope of protection of the present invention. In the various drawings, similar components are numbered similarly.
[0017] Figure 1 This illustration shows the first flowchart of the method for identifying broken strand defects in power distribution line conductors and ground wires proposed in this embodiment. Figure 2This illustration shows the second flowchart of the method for identifying broken strand defects in power distribution line conductors and ground wires proposed in this embodiment; Figure 3 The diagram illustrates the third step of the method for identifying broken strands in power distribution line conductors and ground wires proposed in this embodiment. Figure 4 The fourth flowchart of the method for identifying broken strand defects in power distribution line conductors and ground wires proposed in this embodiment is shown. Figure 5 The fifth flowchart of the method for identifying broken strand defects in power distribution line conductors and ground wires proposed in this embodiment is shown. Figure 6 The sixth flowchart of the method for identifying broken strand defects in power distribution line conductors and ground wires proposed in this embodiment is shown. Figure 7 A schematic diagram of the structure of the power distribution line conductor and ground wire broken strand defect identification device proposed in this embodiment is shown.
[0018] Explanation of reference numerals in the attached diagram: 700-Distribution line conductor / ground wire broken strand defect identification device; 701-Input module; 702-Extraction module; 703-Prediction module; 704-Enhancement module; 705-Fusion module; 706-Identification module. Detailed Implementation
[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0020] The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0021] In the following, the terms “comprising,” “having,” and their cognates, which may be used in various embodiments of the invention, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as excluding, firstly, the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more features, numbers, steps, operations, elements, components, or combinations thereof.
[0022] Furthermore, the terms "first," "second," and "third" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.
[0023] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of the invention pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be interpreted as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of the invention.
[0024] Example 1 This disclosure provides a method for identifying conductor and ground wire breakage defects in power distribution lines. It introduces a binary classification identification module and an adaptive feature enhancement mechanism, and combines feature fusion to significantly improve the feature representation capability and positioning accuracy of small-sized defects. It effectively suppresses false detection caused by background interference and improves the reliability and practicality of the automated inspection system for power distribution lines.
[0025] Please see Figure 1 The method for identifying broken strand defects in the conductor and ground wire of the power distribution line includes steps S101 to S106, and each step is described in detail below.
[0026] Step S101: Obtain images of power distribution lines collected by the UAV during inspection, and input the power distribution line images into the improved YOLOv11 target detection network.
[0027] In this embodiment, images of power distribution lines collected by UAV inspections are acquired and input into an improved YOLOv11 target detection network for identifying conductor / ground wire breakage defects. The improved YOLOv11 target detection network includes a backbone network, a neck network, and a head network. The neck network includes a Two Classification Recognition Module (TCRM) and a feature fusion module. The TCRM performs pixel-level foreground-background binary classification based on the feature map output by the backbone network. The feature fusion module dynamically enhances target region features and suppresses background interference.
[0028] Step S102: Extract multi-scale feature maps from the power distribution line image through the backbone network.
[0029] In this embodiment, multi-level spatial semantic information is extracted from the power distribution line image through the backbone network to obtain a multi-scale feature map. The multi-scale feature map enhances the receptive field of the target detection network, acquires information about the surroundings of the target, and can focus on small targets, which can more effectively identify small target defects such as broken strands in the conductor and ground wire of the power distribution line.
[0030] Step S103: The multi-scale feature map is subjected to binary classification prediction by the binary classification recognition module to obtain the features. Figure 2 Classification results.
[0031] In this embodiment, a binary classification recognition module performs binary classification prediction on the multi-scale feature map to obtain the features. Figure 2 The classification results distinguish between target and background regions in the original image, significantly reducing the possibility of false positives caused by network prediction. Furthermore, the binary classification network fully utilizes multi-scale feature maps for classification, ensuring feature map consistency and reducing the need for further feature extraction.
[0032] Please see Figure 2 In one specific embodiment, the multi-scale feature map includes a first feature map. Feature 1 Second feature map Feature 2 and the third feature map Feature 3 Step S103 includes steps S1031 to S1033, and each step is described in detail below.
[0033] Step S1031: The second feature map and the third feature map are upsampled by the binary classification recognition module to obtain the upsampled second feature map and the upsampled third feature map.
[0034] In this embodiment, please refer to Figure 3 The second feature map is classified using a binary classification module. Feature 2 and the third feature map Feature 3 Perform upsampling ( ), to obtain the second feature map after upsampling. Feature2_up and the third feature map after upsampling Feature3_up The specific expression is as follows: , .
[0035] Step S1032: The first feature map, the upsampled second feature map, and the upsampled third feature map are concatenated to obtain a concatenated feature map.
[0036] In this embodiment, the first feature map Feature 1 The second feature map after upsampling Feature2_up and the third feature map after upsampling Feature3_up splicing ( ), to obtain the spliced feature map Feature_123 Its specific expression is: .
[0037] Understandably, the second and third feature maps of the higher layers are upsampled and then concatenated with the first feature map of the lower layer to achieve cross-layer semantic complementarity and enhance the consistency between local details and global context.
[0038] Step S1033: The stitched feature map is cropped into multiple sub-images, and binary classification prediction is performed on each sub-image to obtain the binary classification prediction result of each sub-image, which is used as the feature. Figure 2 Classification results.
[0039] In this embodiment, the stitched feature map is cropped into multiple sub-maps, such as nine sub-maps, to simulate a local region attention mechanism, enabling the model to focus on potential defect areas; further, binary classification prediction is performed on each sub-map. Two Detect The binary classification prediction results of each subgraph are obtained and used as features. Figure 2 Classification results. Please see again. Figure 3 The red area represents the area where the target appears after binary classification prediction, and the blank area represents the area where the background appears after binary classification prediction.
[0040] Understandably, when predicted as a background region, its predicted output value is small, and the region receives a lower weight during adaptive feature fusion. This leads to an appropriate reduction in the fusion weight of the region's features, thus reducing interference from background features. Conversely, when predicted as a target region, its predicted output value is large, and the region receives a higher weight during adaptive feature fusion. This leads to an appropriate increase in the fusion weight of the region's features, enhancing the display of target region features and helping the YOLOv11 model better distinguish between foreground and background features.
[0041] Step S104, based on the features Figure 2 The classification results are adaptively enhanced to obtain a reset feature map.
[0042] In this embodiment, adaptive enhancement is performed based on the binary classification results to obtain a reset feature map, which can adapt to the differences in features between the target and background regions under different scenarios, highlight the response signal of the defect region of the small target, and suppress the response signal of the background region.
[0043] Please see Figure 4 In one specific embodiment, step S104 includes steps S1041 to S1042, and each step is described in detail below.
[0044] Step S1041: Determine the feature mean of each sub-graph based on the binary classification prediction results of each sub-graph.
[0045] In this embodiment, the feature mean of each sub-image is determined based on the binary classification prediction results of each sub-image. Specifically, for each sub-image, the feature mean is the sum of the results of each pixel in the binary classification prediction results of the sub-image, divided by the average number of pixels in the sub-image.
[0046] Step S1042: Obtain a new feature map for each sub-graph based on the average feature value of each sub-graph, and use it as the reset feature map.
[0047] In this embodiment, a new feature map for each sub-graph is obtained based on the average feature value of each sub-graph, and all new feature maps are used as reset feature maps. The calculation formula for the new feature map is as follows: In the formula, For the first New feature maps of each subgraph For the first The feature mean of each subplot.
[0048] Understandably, reconstructing a new feature map using the mean of features achieves spatial redistribution and energy concentration of the original features. Compared to directly using the original features, the reconstructed features are more discriminative and help alleviate the problem of small targets being easily lost in deep networks.
[0049] Step S105: The reset feature map is downsampled by the feature fusion module, and feature fusion is performed based on the downsampled reset feature map and the multi-scale feature map to obtain the target fused feature map.
[0050] In this embodiment, the reset feature map is downsampled by the feature fusion module to obtain more abstract semantic information. Based on the high-level semantic features in the downsampled reset feature map and the spatial detail features retained in the multi-scale feature map, feature fusion is performed to obtain a target fused feature map that contains both rich semantic information and accurate spatial information, which greatly enhances the expressive power of a single feature map.
[0051] Please see Figure 5 In one specific embodiment, the downsampled reset feature map includes multiple downsampled feature maps, and step S105 includes steps S1051 to S1055. Each step is described in detail below.
[0052] Step S1051: The reset feature map is downsampled based on multiple preset channel numbers to obtain downsampled feature maps corresponding to each preset channel number.
[0053] In this embodiment, the reset feature map is downsampled based on multiple different preset channel numbers to obtain downsampled feature maps corresponding to each preset channel number, thereby preserving its expressive power at different levels of abstraction. For example, downsampling can be performed based on three different preset channel numbers to obtain a first downsampled feature map, a second downsampled feature map, and a third downsampled feature map.
[0054] Step S1052: Calculate the sum of pixels in each of the downsampled feature maps to obtain the sum of pixels in the feature map corresponding to each downsampled feature map.
[0055] In this embodiment, the sum of pixels in each downsampled feature map is calculated as an indicator of its overall activation level, reflecting the potential contribution of that branch to the final decision.
[0056] Step S1053: Calculate the sum of all the feature map pixels to obtain the sum of the pixels corresponding to the downsampled reset feature map.
[0057] In this embodiment, the sum of all feature map pixels is calculated and used as the sum of pixels corresponding to the downsampled reset feature map.
[0058] Step S1054: Calculate the quotient of each feature map pixel sum with the sum of the pixel sum values to obtain the first feature map weights corresponding to each downsampled feature map.
[0059] In this embodiment, the sum of each feature map pixel is divided by the sum of the pixel values to normalize the sum of each feature map pixel, thereby obtaining the first feature map weight corresponding to each branch and realizing an automatic weighting mechanism based on energy distribution.
[0060] Step S1055: Perform feature fusion on the downsampled reset feature map and the multi-scale feature map according to the weights of each of the first feature maps to obtain the target fused feature map.
[0061] In this embodiment, the downsampled reset feature map and the multi-scale feature map are fused according to the weights of each first feature map to obtain the target fused feature map, thereby dynamically responding to changes in the content of the input image and truly realizing data-driven feature fusion.
[0062] Please see Figure 6 In one specific embodiment, the target fusion feature map includes multiple fusion feature maps, and step S1055 includes steps S601 to S602. Each step is described in detail below.
[0063] Step S601: For each of the first feature map weights, calculate the second feature map weight corresponding to the first feature map weight based on the first feature map weight.
[0064] In this embodiment, for each first feature map weight, the corresponding second feature map weight is calculated based on the first feature map weight. The formula for calculating the second feature map weight is as follows: In the formula, For the first The weights of the second feature map corresponding to the weights of the first feature map. For the first The weights of the first feature map.
[0065] Step S602: The downsampled reset feature map and the multi-scale feature map are weighted and summed according to the first feature map weight and its corresponding second feature map weight to obtain the fused feature map corresponding to the first feature map weight.
[0066] In this embodiment, based on the weight of the first feature map and its corresponding second feature map weights Reset feature map after downsampling and multi-scale feature maps Perform a weighted summation to obtain the weights of the first feature map. Corresponding fusion feature map Its expression is: .
[0067] Understandably, this is achieved by utilizing the weights of the first feature map. and its corresponding second feature map weights A dual-level control mechanism is implemented to improve the flexibility and robustness of feature fusion. The final target fusion feature map integrates the original multi-scale features and the screened and enhanced reconfigured features, combining breadth and depth, and significantly improving the performance of small target recognition.
[0068] Step S106: The target fusion feature map is identified through the head network to obtain the target detection and identification result of the broken strand of the power distribution line conductor and ground wire.
[0069] In this embodiment, the head network identifies each fused feature map in the target fusion feature map to obtain the target detection and identification result of broken strands in the conductor and ground wire of the power distribution line. This greatly enhances the ability to identify small, blurry, and low-contrast broken strand defects, reduces false detections and missed detections, and can maintain stable output even in complex backgrounds (such as tree branches blocking the light or uneven lighting). At the same time, it outputs standardized target boxes and category labels, which are convenient for subsequent maintenance decision-making systems to call, and promotes the construction of a closed loop for intelligent operation and maintenance of power distribution lines.
[0070] The proposed method for identifying conductor / ground wire breakage defects in power distribution lines involves acquiring images of power distribution lines collected by drone inspections and inputting these images into an improved YOLOv11 target detection network. This improved network comprises a backbone network, a neck network, and a head network. The neck network includes a binary classification module and a feature fusion module. Multi-scale feature maps are extracted from the power distribution line images via the backbone network. The binary classification module then performs binary classification prediction on these multi-scale feature maps to obtain the feature maps. Figure 2 The classification results are used to adaptively enhance the binary classification results, resulting in a reset feature map. The reset feature map is then downsampled using a feature fusion module, and feature fusion is performed between the downsampled reset feature map and the multi-scale feature map to obtain a target fusion feature map. The head network then identifies the target fusion feature map to obtain the target detection and recognition result for broken conductor / ground wire strands in the power distribution line. Thus, based on the standard target detection network structure, a binary classification module, adaptive feature enhancement, and dynamic fusion mechanism are introduced into the neck network to perform pixel-level foreground-background binary classification on the multi-scale feature map. Furthermore, a reset feature map is generated based on the binary classification results, and feature fusion is achieved, effectively improving the identification effect of broken conductor / ground wire strand defects while significantly reducing false alarms and improving the overall recognition performance of the model network.
[0071] Example 2 Furthermore, this disclosure provides a device 700 for identifying broken strands in power distribution line conductors and ground wires; please refer to [link to relevant documentation]. Figure 7 The device includes: The input module 701 is used to acquire images of power distribution lines collected by the UAV during inspection, and input the images of the power distribution lines into the improved YOLOv11 target detection network; the improved YOLOv11 target detection network includes a backbone network, a neck network and a head network, and the neck network includes a binary classification recognition module and a feature fusion module. Extraction module 702 is used to extract multi-scale feature maps from the power distribution line image through the backbone network; Prediction module 703 is used to perform binary classification prediction on the multi-scale feature map through the binary classification recognition module to obtain features. Figure 2 Classification results; Enhancement module 704, configured to, based on the features Figure 2 The classification results are adaptively enhanced to obtain a reset feature map; The fusion module 705 is used to downsample the reset feature map through the feature fusion module, and perform feature fusion based on the downsampled reset feature map and the multi-scale feature map to obtain a target fused feature map; The identification module 706 is used to identify the target fusion feature map through the head network to obtain the target detection and identification result of the broken strand of the conductor and ground wire of the power distribution line.
[0072] In one embodiment, the multi-scale feature map includes a first feature map, a second feature map, and a third feature map. The prediction module 703 is further configured to upsample the second feature map and the third feature map respectively through the binary classification recognition module to obtain an upsampled second feature map and an upsampled third feature map; to concatenate the first feature map, the upsampled second feature map, and the upsampled third feature map to obtain a concatenated feature map; to crop the concatenated feature map into multiple sub-images; and to perform binary classification prediction on each sub-image to obtain the binary classification prediction result of each sub-image, which is used as the feature map. Figure 2 Classification results.
[0073] In one embodiment, the enhancement module 704 is further configured to determine the feature mean of each sub-graph based on the binary classification prediction result of each sub-graph; and to obtain a new feature map of each sub-graph based on the feature mean of each sub-graph, as the reset feature map.
[0074] In one embodiment, the downsampled reset feature map includes multiple downsampled feature maps. The fusion module 705 is further configured to downsample the reset feature map based on multiple preset channel numbers to obtain downsampled feature maps corresponding to each preset channel number; calculate the pixel sum of each downsampled feature map to obtain the feature map pixel sum corresponding to each downsampled feature map; calculate the sum of all feature map pixel sums to obtain the pixel sum sum of the downsampled reset feature map; calculate the quotient of each feature map pixel sum with the pixel sum sum value to obtain the first feature map weight corresponding to each downsampled feature map; and perform feature fusion on the downsampled reset feature map and the multi-scale feature map according to each first feature map weight to obtain the target fused feature map.
[0075] In one embodiment, the target fusion feature map includes multiple fusion feature maps. The fusion module 705 is further configured to calculate the second feature map weight corresponding to the first feature map weight based on the first feature map weight for each first feature map weight; and to perform a weighted summation of the downsampled reset feature map and the multi-scale feature map based on the first feature map weight and its corresponding second feature map weight to obtain the fusion feature map corresponding to the first feature map weight.
[0076] In one embodiment, the formula for calculating the new feature map is: In the formula, For the first New feature maps of each subgraph For the first The feature mean of each subplot.
[0077] In one embodiment, the formula for calculating the weight of the second feature map is: In the formula, For the first The weights of the second feature map corresponding to the weights of the first feature map. For the first The weights of the first feature map.
[0078] The apparatus provided in this embodiment can perform the steps of the method for identifying broken strand defects in power distribution line conductors and ground wires provided in Embodiment 1. To avoid repetition, these steps will not be repeated.
[0079] The power distribution line conductor / ground wire breakage defect identification device proposed in this embodiment acquires power distribution line images collected by UAV inspections and inputs these images into an improved YOLOv11 target detection network. The improved YOLOv11 target detection network includes a backbone network, a neck network, and a head network. The neck network includes a binary classification module and a feature fusion module. Multi-scale feature maps are extracted from the power distribution line images through the backbone network. The binary classification module performs binary classification prediction on the multi-scale feature maps to obtain the feature maps. Figure 2 The classification results are used to adaptively enhance the binary classification results, resulting in a reset feature map. The reset feature map is then downsampled using a feature fusion module, and feature fusion is performed between the downsampled reset feature map and the multi-scale feature map to obtain a target fusion feature map. The head network then identifies the target fusion feature map to obtain the target detection and recognition result for broken conductor / ground wire strands in the power distribution line. Thus, based on the standard target detection network structure, a binary classification module, adaptive feature enhancement, and dynamic fusion mechanism are introduced into the neck network to perform pixel-level foreground-background binary classification on the multi-scale feature map. Furthermore, a reset feature map is generated based on the binary classification results, and feature fusion is achieved, effectively improving the identification effect of broken conductor / ground wire strand defects while significantly reducing false alarms and improving the overall recognition performance of the model network.
[0080] Example 3 Furthermore, this disclosure provides a computer device including a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the method for identifying broken strands in power distribution line conductors and ground wires as described in Embodiment 1.
[0081] The device provided in this embodiment can perform the steps of the method for identifying broken strand defects in power distribution line conductors and ground wires provided in Embodiment 1. To avoid repetition, these steps will not be repeated.
[0082] Example 4 This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for identifying broken strands in power distribution lines as described in Embodiment 1.
[0083] In this embodiment, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.
[0084] The computer-readable storage medium provided in this embodiment can implement the method for identifying broken strand defects in power distribution line conductors and ground wires provided in Embodiment 1. To avoid repetition, it will not be described again here.
[0085] In all examples shown and described herein, any specific values should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.
[0086] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0087] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.
Claims
1. A method for identifying broken strand defects in the conductors and ground wires of a power distribution line, characterized in that, include: The image of the power distribution line collected by the UAV inspection is acquired, and the image of the power distribution line is input into the improved YOLOv11 target detection network. The improved YOLOv11 target detection network includes a backbone network, a neck network and a head network. The neck network includes a binary classification recognition module and a feature fusion module. Multi-scale feature maps are extracted from the power distribution line images using the backbone network; The binary classification recognition module performs binary classification prediction on the multi-scale feature map to obtain the binary classification result of the feature map. Adaptive enhancement is performed based on the binary classification results of the feature map to obtain a reset feature map; The reset feature map is downsampled by the feature fusion module, and feature fusion is performed on the downsampled reset feature map and the multi-scale feature map to obtain the target fused feature map; The target fusion feature map is identified by the head network to obtain the target detection and identification result of broken strands in the conductor and ground wire of the power distribution line.
2. The method for identifying broken strand defects in power distribution line conductors and ground wires according to claim 1, characterized in that, The multi-scale feature map includes a first feature map, a second feature map, and a third feature map. The step of performing binary classification prediction on the multi-scale feature map using the binary classification recognition module to obtain the feature map binary classification result includes: The binary classification recognition module upsamples the second feature map and the third feature map respectively to obtain the upsampled second feature map and the upsampled third feature map. The first feature map, the upsampled second feature map, and the upsampled third feature map are concatenated to obtain a concatenated feature map. The spliced feature map is cropped into multiple sub-images, and binary classification prediction is performed on each sub-image to obtain the binary classification prediction result of each sub-image, which is used as the binary classification result of the feature map.
3. The method for identifying broken strand defects in power distribution line conductors and ground wires according to claim 2, characterized in that, The adaptive enhancement based on the binary classification result of the feature map to obtain the reset feature map includes: The feature mean of each subgraph is determined based on the binary classification prediction results of each subgraph; A new feature map for each subgraph is obtained based on the average feature value of each subgraph, and is used as the reset feature map.
4. The method for identifying broken strand defects in power distribution line conductors and ground wires according to claim 1, characterized in that, The downsampled reset feature map includes multiple downsampled feature maps. The step of downsampling the reset feature map using the feature fusion module and fusing features based on the downsampled reset feature map and the multi-scale feature map to obtain the target fused feature map includes: The reset feature map is downsampled based on multiple preset channel numbers to obtain downsampled feature maps corresponding to each preset channel number. Calculate the sum of pixels in each of the downsampled feature maps to obtain the sum of pixels in the feature map corresponding to each downsampled feature map; Calculate the sum of all the pixel sums in the feature map to obtain the pixel sum of the downsampled reset feature map; Calculate the quotient of each feature map pixel sum with the sum of the pixel sum and the sum of the values to obtain the first feature map weights corresponding to each downsampled feature map; Based on the weights of each of the first feature maps, feature fusion is performed on the downsampled reset feature map and the multi-scale feature map to obtain the target fused feature map.
5. The method for identifying broken strand defects in power distribution line conductors and ground wires according to claim 4, characterized in that, The target fusion feature map includes multiple fusion feature maps. The step of fusing the downsampled reset feature map and the multi-scale feature map according to the weights of each of the first feature maps to obtain the target fusion feature map includes: For each of the first feature map weights, calculate the second feature map weights corresponding to the first feature map weights based on the first feature map weights. The downsampled reset feature map and the multi-scale feature map are weighted and summed according to the first feature map weight and its corresponding second feature map weight to obtain the fused feature map corresponding to the first feature map weight.
6. The method for identifying broken strand defects in power distribution line conductors and ground wires according to claim 3, characterized in that, The formula for calculating the new feature map is: In the formula, For the first New feature maps of each subgraph For the first The feature mean of each subplot.
7. The method for identifying broken strand defects in power distribution line conductors and ground wires according to claim 5, characterized in that, The formula for calculating the weights of the second feature map is: In the formula, For the first The weights of the second feature map corresponding to the weights of the first feature map. For the first The weights of the first feature map.
8. A device for identifying broken strand defects in power distribution line conductors and ground wires, characterized in that, include: The input module is used to acquire images of power distribution lines collected by UAV inspections and input the images of power distribution lines into an improved YOLOv11 target detection network. The improved YOLOv11 target detection network includes a backbone network, a neck network, and a head network. The neck network includes a binary classification recognition module and a feature fusion module. The extraction module is used to extract multi-scale feature maps from the power distribution line image through the backbone network; The prediction module is used to perform binary classification prediction on the multi-scale feature map through the binary classification recognition module to obtain the binary classification result of the feature map; The enhancement module is used to perform adaptive enhancement based on the binary classification result of the feature map to obtain a reset feature map; The fusion module is used to downsample the reset feature map through the feature fusion module, and perform feature fusion based on the downsampled reset feature map and the multi-scale feature map to obtain a target fused feature map; The identification module is used to identify the target fusion feature map through the head network to obtain the target detection and identification result of the broken strand of the conductor and ground wire of the power distribution line.
9. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method for identifying broken strands in power distribution lines as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method for identifying broken strands in power distribution lines as described in any one of claims 1 to 7.