Power transmission line inspection method and system

Abstract

The application relates to a power field power transmission line inspection method and system, and relates to the technical field of power transmission lines, and comprises the following steps: acquiring a power distribution line video of a power transmission line, and performing boundary box and device category labeling on the power distribution line video to obtain a power transmission line labeled image; the power transmission line labeled image is input into a preset power distribution line inspection model to judge whether the power transmission line has a fault defect; if yes, the power transmission line is continuously tracked based on a preset DeepSORT model based on the fault defect to obtain a motion track of the fault defect in the power distribution line video; the power transmission line is evaluated in a defect state based on the motion track to generate an inspection report of the power transmission line, and the technical problems that the prior art cannot continuously track and state analyze dynamic fault defects appearing in a video and cannot accurately judge the development trend and severity of defects are solved.

Description

Technical Field

[0001] This invention relates to the field of power transmission line technology, and in particular to a method and system for inspecting power transmission lines in power scenarios. Background Technology

[0002] To improve the efficiency and safety of power transmission line inspections, some regions have begun to introduce technologies such as drone inspections and video surveillance to achieve remote, visual inspections by acquiring video data of power transmission lines. However, in practical applications, these technologies still have significant shortcomings: On the one hand, the amount of video data collected by drones and monitoring equipment is enormous, requiring manual review and defect identification, which not only consumes a lot of manpower but is also prone to missed or misjudged faults due to factors such as human fatigue and lack of experience; on the other hand, existing video-based inspection technologies are mostly limited to defect identification based on static images, and cannot continuously track and analyze the status of dynamic faults and defects appearing in the video (such as abnormal conductor swaying, foreign objects drifting in the wind, etc.), making it difficult to accurately judge the development trend and severity of defects, which brings many inconveniences to the maintenance decisions of power transmission lines. Summary of the Invention

[0003] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0004] To achieve the above objectives, the present invention provides a method for inspecting power transmission lines in a power scenario, comprising the following steps:

[0005] Acquire video of the distribution line of the transmission line, and annotate the distribution line video with bounding boxes and equipment categories to obtain an annotated image of the transmission line;

[0006] The labeled image of the transmission line is input into a preset power distribution line inspection model to determine whether there are any faults or defects in the transmission line.

[0007] If it exists, the fault defect is input into the preset DeepSORT model, which includes a feature extraction module, a state prediction module, and a trajectory association module.

[0008] The fault defect is input into the feature extraction module for feature vector calculation to obtain the defect feature description vector;

[0009] The defect feature description vector is input into the state prediction module to estimate the motion state and obtain the defect motion parameters.

[0010] The defect motion parameters are input into the trajectory association module for time-series association analysis to obtain the motion trajectory of the fault defect in the power distribution line video.

[0011] Based on the motion trajectory, the defect status of the transmission line is assessed, and an inspection report of the transmission line is generated.

[0012] Furthermore, acquiring the distribution line video of the transmission line includes:

[0013] Video acquisition equipment was deployed at key detection points on the power transmission line to obtain a monitoring and acquisition scheme.

[0014] Based on the aforementioned monitoring and acquisition scheme, multi-angle video acquisition is performed on the transmission lines to obtain the original video of the distribution lines.

[0015] The original video of the power distribution line is processed with image stabilization to obtain the power distribution line video.

[0016] Furthermore, the process of acquiring original video of the power distribution line from multiple angles based on the monitoring and acquisition scheme includes:

[0017] Based on the monitoring and acquisition scheme, the parameters of each preset video acquisition device are configured, the acquisition frame rate, resolution and acquisition angle range are set to obtain the configured acquisition device group, and the preliminary video acquisition of the transmission line at different angles is carried out through the configured acquisition device group to obtain multiple sets of preliminary angle videos.

[0018] The video quality of the multiple sets of preliminary angle videos is checked, and videos with clear images, no obstructions, and angles that meet the preset requirements are selected to obtain qualified angle video groups. Then, the time synchronization processing of each video in the qualified angle video group is performed to make the time start point of each video consistent, so as to obtain the original video of the power distribution line.

[0019] Furthermore, the step of annotating the power distribution line video with bounding boxes and equipment categories to obtain an annotated transmission line image includes:

[0020] The video of the power distribution line is extracted frame by frame to obtain a single frame of power distribution image, and the equipment edges in the single frame of power distribution image are contour detected to obtain the equipment contour image.

[0021] Based on the equipment outline image, the equipment in the transmission line is initially classified to obtain a preliminary classified equipment image. Boundary boxes are drawn for the equipment in the preliminary classified equipment image. A rectangular boundary box is drawn by extending a certain pixel distance outward from the outermost edge of the equipment outline to obtain a bounding box equipment image.

[0022] Based on a preset standard image library of power equipment, the bounding box equipment images are classified and matched to obtain labeled images of transmission lines.

[0023] Furthermore, the step of performing contour detection on the equipment edges in the single-frame power distribution image to obtain the equipment contour image includes:

[0024] The single-frame power distribution image is converted to grayscale to obtain a grayscale power distribution image. Pixel gradients are calculated on the grayscale power distribution image to obtain the gradient value of each pixel. Edge pixels are then selected based on the gradient values.

[0025] Connectivity analysis is performed on the edge pixels, and adjacent pixels with similar gradient values ​​are connected into line segments to obtain edge line segment groups. Based on the edge line segment groups, closed contours are constructed for the equipment edges in a single frame of the power distribution image to obtain the equipment contour image.

[0026] Furthermore, the marked images of the transmission lines are input into a preset distribution line inspection model to determine whether the transmission lines have faults or defects, including:

[0027] The labeled image of the transmission line is input into a preset distribution line inspection model. The preset distribution line inspection model includes the Backbone, Neck, and Head parts of the original YOLOv8s model, a C2f_Faster lightweight feature extraction module, a SimSPPF simplified spatial pyramid pooling module, and a NAM normalized attention module. The C2f_Faster lightweight feature extraction module replaces the C2f modules in the Backbone and Neck parts with C2f_Faster modules. The SimSPPF simplified spatial pyramid pooling module replaces the SPPF module in the Backbone part with a SimSPPF module and adds a NAM attention mechanism between the SimSPPF module and the Neck part.

[0028] The labeled image of the transmission line is input into the C2f_Faster lightweight feature extraction module for partial channel convolution and feature stitching to obtain the feature map of the lightweight equipment.

[0029] The feature map of the lightweight device is input into the SimSPPF simplified spatial pyramid pooling module for multi-scale spatial pooling processing to obtain a multi-scale pooling feature map. The multi-scale pooling feature map is then subjected to cross-scale feature stitching processing to obtain a spatial pyramid fused feature map.

[0030] The spatial pyramid fusion feature map is input into the NAM normalized attention module for channel attention weighting processing to obtain a channel weighted feature map, and the channel weighted feature map is then subjected to spatial attention weighting processing to obtain an attention-enhanced feature map.

[0031] The attention-enhanced feature map is input into the Head section for bounding box regression and fault category prediction to determine whether the transmission line has any faults or defects.

[0032] Furthermore, based on the motion trajectory, a defect status assessment of the transmission line is performed, generating an inspection report for the transmission line, including:

[0033] The motion trajectory of the fault defect in the power distribution line video is extracted to obtain a set of trajectory points, and each trajectory point in the set of trajectory points is timestamped to obtain a set of timestamped trajectory points.

[0034] Based on the set of timestamped trajectory points, the movement speed of the transmission line is calculated to obtain the movement speed sequence of the fault in each time period. Based on the movement speed sequence, the speed change of the transmission line is analyzed to obtain the speed change characteristic curve.

[0035] The defect status of the transmission line is evaluated based on the speed change characteristic curve to obtain the defect status evaluation result, and an inspection report of the transmission line is generated based on the defect status evaluation result.

[0036] Furthermore, the defect status of the transmission line is evaluated based on the speed change characteristic curve to obtain the defect status evaluation result, including:

[0037] Peak points are extracted from the velocity change characteristic curve to obtain a velocity peak sequence, and the slope of the velocity peak sequence is calculated to obtain a velocity change slope map including the positive and negative values ​​of the slope and the distribution of inflection points.

[0038] Based on the velocity change slope diagram, the fault is dynamically quantified to obtain a defect development score. Based on the defect development score, the transmission line is classified into states to obtain a defect state assessment result.

[0039] This invention also provides a power transmission line inspection system, comprising:

[0040] The acquisition module is used to acquire video of the distribution line of the transmission line, and to annotate the distribution line video with bounding boxes and equipment categories to obtain an annotated image of the transmission line;

[0041] The judgment module is used to input the marked image of the transmission line into a preset distribution line inspection model to determine whether the transmission line has any faults or defects.

[0042] The tracking module is used to input the fault defect into a preset DeepSORT model if it exists. The preset DeepSORT model includes a feature extraction module, a state prediction module, and a trajectory association module.

[0043] The fault defect is input into the feature extraction module for feature vector calculation to obtain the defect feature description vector;

[0044] The defect feature description vector is input into the state prediction module to estimate the motion state and obtain the defect motion parameters.

[0045] The defect motion parameters are input into the trajectory association module for time-series association analysis to obtain the motion trajectory of the fault defect in the power distribution line video.

[0046] The evaluation module is used to evaluate the defect status of the transmission line based on the motion trajectory and generate an inspection report of the transmission line.

[0047] This invention provides a method for inspecting power transmission lines in a power scenario, comprising the following steps: acquiring a video of the distribution line of the transmission line, and annotating the video with bounding boxes and equipment categories to obtain an annotated image of the transmission line; inputting the annotated image of the transmission line into a preset distribution line inspection model to determine whether there are faults or defects in the transmission line; if so, continuously tracking the transmission line based on the faults or defects using a preset DeepSORT model to obtain the motion trajectory of the faults or defects in the distribution line video; and evaluating the defect status of the transmission line based on the motion trajectory to generate an inspection report of the transmission line. This method solves the technical problem that existing technologies cannot continuously track and analyze the status of dynamic faults and defects appearing in videos, making it difficult to accurately determine the development trend and severity of defects. It significantly reduces the number of manual intervention steps, not only reducing the labor intensity and subjective error of manual analysis, but also making the inspection results more systematic and traceable. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the steps of a power transmission line inspection method in an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the steps of a power transmission line inspection system in an embodiment of the present invention;

[0051] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0052] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0053] The following describes in detail, with reference to the accompanying drawings, a method for inspecting power transmission lines in a power scenario according to an embodiment of the present invention.

[0054] Figure 1 This invention provides a method for inspecting power transmission lines in a power scenario, comprising the following steps:

[0055] Step S1: Obtain the video of the power distribution line of the transmission line, and annotate the bounding box and equipment category of the power distribution line video to obtain the labeled image of the transmission line.

[0056] Specifically, using drones equipped with high-definition cameras or fixed monitoring equipment along the transmission line, video of the distribution lines covering the entire area of ​​the transmission line is collected. It's important to note that video acquisition must consider different lighting and weather conditions; for example, images of the line under cloudy or windy conditions must be included to avoid missing data from special scenarios. Next, the collected distribution line video is broken down frame by frame, and bounding boxes and equipment category annotations are performed on each frame. Bounding box annotation involves precisely selecting core equipment such as transmission conductors, insulators, and fittings in the image using rectangular boxes. For example, when annotating insulators, the selection area must completely cover the insulator body and the connection points at both ends, without omitting edge details. Equipment category annotation assigns a clear category label to each target corresponding to the bounding box; for example, cylindrical insulating components are labeled "insulators," and metal connection components are labeled "fittings." After the individual frames are annotated, all annotated frames are integrated to form a complete annotated dataset. These processed images are the annotated images of the transmission line.

[0057] Step S2: Input the marked image of the transmission line into the preset distribution line inspection model to determine whether there are any faults or defects in the transmission line.

[0058] Specifically, after preparing the labeled images of transmission lines, the next step is to import these images into a pre-set distribution line inspection model. It's important to note that before importing, the labeled images need simple standardization preprocessing, such as adjusting the image size to the model's input specifications to avoid affecting model recognition due to size differences. The pre-set distribution line inspection model will call its built-in feature extraction module to extract equipment features from the labeled images—for example, for insulator labeled images, it will focus on extracting surface cracks, damage, and other defect features; for transmission conductor labeled images, it will focus on features such as broken strands and corrosion. The model compares the extracted features with a pre-set normal equipment feature library. If the feature matching degree is below a threshold, the transmission line is determined to have a fault defect; if the matching degree meets the threshold, it is determined to be defect-free.

[0059] Step S3: If the fault exists, the fault defect is input into the preset DeepSORT model, which includes a feature extraction module, a state prediction module, and a trajectory association module.

[0060] The fault defect is input into the feature extraction module for feature vector calculation to obtain the defect feature description vector;

[0061] The defect feature description vector is input into the state prediction module to estimate the motion state and obtain the defect motion parameters.

[0062] The defect motion parameters are input into the trajectory association module for time-series association analysis to obtain the motion trajectory of the fault defect in the power distribution line video.

[0063] The specific steps are as follows: In the actual application scenario, the transmission line labeled image frames corresponding to the fault defects are subjected to timestamp alignment processing. Based on the UTC timestamp, the fault defect areas of each frame are bound to the frame sequence of the distribution line video. The timestamp alignment accuracy is set to 1ms. The inter-frame time offset is corrected by linear interpolation to obtain the time-aligned fault defect frame sequence. It should be noted that the time-aligned fault defect frame sequence includes the frame index, fault defect bounding box coordinates, equipment category label and corresponding frame timestamp. The coordinate format is (x1, y1, x2, y2), the pixel accuracy is 1px, and the equipment category label is consistent with the equipment category of the transmission line labeled image.

[0064] The time-aligned fault / defect frame sequence is input into the feature extraction module of a preset DeepSORT model. Features of the fault / defect regions in each frame are extracted using a ConvNeXt-Tiny convolutional network. The convolutional layer parameters are set as follows: the first layer uses 3×3 convolutional kernels (64 kernels), stride = 2, padding = 1, and the GELU activation function is used with batch normalization (BN layer momentum is 0.9, weight decay is 1e-4). The second to fourth layers use... A 3×3 depthwise separable convolution is used, with the number of convolution kernels being 128, 256, and 512 respectively, and the stride and padding being 1 for all kernels. The extracted feature maps are compressed into 512-dimensional feature vectors through a global average pooling layer to obtain the initial feature vectors of fault defects in each frame. In addition, the dimensions of the initial feature vectors are [1, 512], and the data format is float32. Each feature dimension corresponds to the texture, contour, and grayscale feature parameters of the fault defect, which can accurately characterize the unique features of transmission line faults (such as broken conductor strands and damaged insulators).

[0065] Based on the initial feature vectors of faults in each frame, the feature vectors are optimized using the Triplet Loss metric learning algorithm. The sampling ratio of anchor features, positive examples, and negative examples is set to 1:3:3, and the margin parameter is 0.5. The feature space distribution is iteratively optimized using stochastic gradient descent (SGD) with a learning rate of 1e-5 and 100 iterations. This ensures that the Euclidean distance between feature vectors of the same type of fault is less than 0.3, and the Euclidean distance between feature vectors of different types of faults is greater than 0.8, resulting in the optimized defect feature description vector. The optimized defect feature description vector still has a dimension of [1, 512], a data format of float32, and contains unique feature identifiers of faults. This effectively distinguishes between different types of faults in transmission lines and faults of the same type in different locations, providing support for accurate correlation of subsequent motion states.

[0066] The optimized defect feature description vector is input into the state prediction module of the preset DeepSORT model. The extended Kalman filter (EKF) algorithm combined with a bidirectional LSTM network is used to estimate the motion state of the fault defect. The EKF state vector is set as [position x, position y, velocity vx, velocity vy, acceleration ax, acceleration ay], the state transition matrix is ​​a 6×6 identity matrix, the process noise covariance matrix Q is a diagonal matrix with all diagonal elements of 0.01, and the observation noise covariance matrix R is a diagonal matrix with all diagonal elements of 0.1. The bidirectional LSTM network contains two hidden layers, each containing a hidden unit. The network has 128 inputs, uses tanh as the activation function, and has an input sequence length of 10 frames. It performs temporal modeling on the feature vectors of consecutive frames and outputs the predicted motion state values ​​of faults in each frame to obtain the fault motion parameters. It should be noted that the fault motion parameters include the position coordinates, velocity, acceleration, and direction angle of the fault in each frame. The position coordinate accuracy is 1px, the velocity unit is px / frame, the acceleration unit is px / frame², and the direction angle range is 0°~360° with an accuracy of 0.1°, which can accurately reflect the motion law of faults (such as wire swaying and foreign object floating).

[0067] The defect motion parameters are input into the trajectory association module of the preset DeepSORT model. A Hungarian algorithm combined with cosine similarity measurement is used for temporal correlation analysis of the fault and defect. First, the cosine similarity between the motion parameters of the current frame's fault and defect and the motion parameters of historical trajectories is calculated, with a similarity association threshold set to 0.75. When the similarity is greater than the threshold, it is determined to be a continuation of the trajectory of the same fault and defect. The optimal trajectory matching pair is solved using the Hungarian algorithm, with a penalty coefficient of 0.5 set for the trajectory cost matrix to resolve the problem of trajectory conflict among multiple targets. Simultaneously, the maximum number of frames for trajectory disappearance is set to 30. When a trajectory fails to detect a matching fault and defect for 30 consecutive frames, the trajectory is determined to terminate, resulting in the preliminary motion trajectory of the fault and defect. The preliminary motion trajectory includes a trajectory ID, motion parameters for each frame, and the number of frames the trajectory continues. The trajectory ID is a unique identifier (formatted as "fault_track_XXX", where XXX is an integer from 001 to 999), which can distinguish the motion trajectories of different faults and defects, avoiding trajectory confusion in multi-fault scenarios.

[0068] Based on the preliminary motion trajectory of the fault defect, a cubic Bézier curve fitting algorithm, a conventional technique, is used to smooth the trajectory. The fitting window size is set to 5 frames, and the least squares method is used to solve for the control point parameters of the Bézier curve (4 control points with a coordinate accuracy of 1px), eliminating trajectory deviations caused by inter-frame motion parameter fluctuations. Simultaneously, outliers in the trajectory are removed using the 3σ criterion. The deviation between the motion parameters of each frame and the average motion parameters of the adjacent 5 frames is calculated. When the deviation is greater than 3 times the standard deviation, it is identified as an outlier and replaced using linear interpolation, thus obtaining the motion trajectory of the fault defect in the power distribution line video. It should be noted that the motion trajectory includes the smoothed position coordinates of each frame, motion parameters, and the overall trajectory trend curve. The trajectory error is less than 2px, accurately reflecting the movement state of the fault defect on the power transmission line.

[0069] Step S4: Based on the motion trajectory, perform a defect status assessment on the transmission line and generate an inspection report for the transmission line.

[0070] Specifically, after obtaining the movement trajectory of the fault / defect in the video of the distribution line, a defect status assessment of the transmission line is conducted based on this trajectory. Specifically, key parameters of the trajectory are analyzed first, such as the amplitude of the trajectory's swing, the speed of movement, and the range of equipment covered by the trajectory. For example, if the trajectory is that of an insulator crack, the severity of the defect can be determined by analyzing whether the swing amplitude at the crack exceeds a safety threshold. After the assessment is completed, the system automatically integrates the assessment results, including information such as defect type, severity level, and location, and generates a complete transmission line inspection report according to the standard format for power line inspection. The report clearly marks the assessment basis related to the movement trajectory for convenient reference in subsequent operation and maintenance.

[0071] In a specific embodiment, acquiring the distribution line video of the transmission line includes:

[0072] Video acquisition equipment was deployed at key detection points on the power transmission line to obtain a monitoring and acquisition scheme.

[0073] Based on the aforementioned monitoring and acquisition scheme, multi-angle video acquisition is performed on the transmission lines to obtain the original video of the distribution lines.

[0074] The original video of the power distribution line is processed with image stabilization to obtain the power distribution line video.

[0075] Specifically, the first step is to deploy video acquisition equipment at key monitoring points along the transmission line to form a monitoring and acquisition plan. In practice, the core sections of the transmission line are first identified, such as conductor joints, insulator strings, and tension towers—areas prone to failure—and these are designated as key monitoring points. Then, appropriate equipment is selected based on the environmental characteristics of each monitoring point—for example, small, lightweight cameras are used at high-altitude conductor joints, while fixed cameras are deployed at tower foundations near the ground. The installation angles and shooting ranges of the equipment are planned to ensure no blind spots. These elements combined form the complete monitoring and acquisition plan. Following this plan, multi-angle video acquisition is conducted. For the same key monitoring point, the camera angle is adjusted or the acquisition equipment is moved (e.g., by drone patrols) to obtain images from the front, side, and overhead views, thus obtaining the original video of the power distribution line. Considering the impact of equipment vibration and wind on outdoor data collection, it is also necessary to perform image stabilization processing on the original video of the power distribution line. The common method is to correct the offset of each frame image through algorithms. For example, when the video collected by the drone shows shaking, the algorithm will automatically calibrate the position of the conductor within the frame to eliminate shaking interference and finally obtain a stable and clear video of the power distribution line.

[0076] In a specific embodiment, the step of acquiring original video of the distribution line from multiple angles based on the monitoring and acquisition scheme includes:

[0077] Based on the monitoring and acquisition scheme, the parameters of each preset video acquisition device are configured, the acquisition frame rate, resolution and acquisition angle range are set to obtain the configured acquisition device group, and the preliminary video acquisition of the transmission line at different angles is carried out through the configured acquisition device group to obtain multiple sets of preliminary angle videos.

[0078] The video quality of the multiple sets of preliminary angle videos is checked, and videos with clear images, no obstructions, and angles that meet the preset requirements are selected to obtain qualified angle video groups. Then, the time synchronization processing of each video in the qualified angle video group is performed to make the time start point of each video consistent, so as to obtain the original video of the power distribution line.

[0079] Specifically, the first step is to configure the parameters of each pre-set video acquisition device according to the monitoring and acquisition plan. The key is to determine the three core parameters: frame rate, resolution, and acquisition angle range. For example, for acquisition devices targeting dynamically vulnerable parts such as high-altitude conductor joints, a frame rate of 30fps is usually required to capture the details of slight conductor swaying, a resolution of 1080P is adjusted to ensure clear defect textures, and the acquisition angle range is aimed at the joint and the surrounding 1.5-meter area. For static parts such as tower foundations, a frame rate of 15fps is sufficient, and a resolution of 720P is adequate. After configuring according to needs, a set of acquisition devices adapted to different acquisition scenarios is formed. Next, this set of devices is started to perform preliminary video acquisition of the transmission line from different angles. For example, images of insulator strings are acquired from frontal, side, and overhead angles, and images of conductor sections are acquired from horizontal and elevation angles, ultimately obtaining multiple sets of preliminary angle videos. After the initial data collection is completed, the video quality of these multiple sets of preliminary angle videos must be checked. The focus should be on checking for blurry images, overexposed lighting, or obstructions from foliage. Simultaneously, the acquisition range of each angle should be verified to meet the preset requirements of the monitoring acquisition plan. Videos that meet the standards should be selected to form a qualified angle video group. Because the startup times of different devices may vary, time synchronization processing must be performed on the videos in the qualified angle video group. This is generally done by calibrating the built-in clocks of the devices, aligning the start point of all videos to the same time, for example, starting at 9:00:00 on a certain day. After completing these operations, the original video of the power distribution line that meets the requirements can be obtained.

[0080] In a specific embodiment, the step of annotating the power distribution line video with bounding boxes and equipment categories to obtain an annotated image of the power transmission line includes:

[0081] The video of the power distribution line is extracted frame by frame to obtain a single frame of power distribution image, and the equipment edges in the single frame of power distribution image are contour detected to obtain the equipment contour image.

[0082] Based on the equipment outline image, the equipment in the transmission line is initially classified to obtain a preliminary classified equipment image. Boundary boxes are drawn for the equipment in the preliminary classified equipment image. A rectangular boundary box is drawn by extending a certain pixel distance outward from the outermost edge of the equipment outline to obtain a bounding box equipment image.

[0083] Based on a preset standard image library of power equipment, the bounding box equipment images are classified and matched to obtain labeled images of transmission lines.

[0084] Specifically, the first step is to extract frames from the power distribution line video. Using video analysis tools, the continuous video stream is broken down into independent single-frame power distribution images. It's crucial to preserve the temporal order of the frames to avoid logical inconsistencies in subsequent processing. Next, for each single-frame power distribution image, contour detection of equipment edges is performed. Edge detection algorithms are typically used to capture the contour features of equipment in the image, such as identifying the linear contours of transmission lines and the columnar contours of insulators. After algorithm processing, equipment contour images highlighting the equipment's shape are obtained. Based on these, preliminary classification judgments are made of the equipment in the transmission line according to the differences in the shape of the equipment contours. For example, linear, slender contours are highly likely to be identified as transmission lines, while umbrella-shaped or columnar contours are initially classified as insulators. This forms a preliminary classification of the equipment images. Next, bounding box drawing is performed. The core principle is to use the outermost edge of the equipment outline as a reference and extend it outward by a certain number of pixels to draw a rectangular bounding box. For example, if the coordinates of the outermost edge of the insulator outline are (x1, y1) to (x2, y2), it can be extended outward by 5 pixels, setting the bounding box coordinates to (x1-5, y1-5) to (x2+5, y2+5), ensuring complete coverage of the equipment and surrounding defect areas, ultimately obtaining an image of the equipment with a bounding box. The final step is to perform category matching based on a preset standard library of power equipment. The equipment features in the bounding box image are compared with the standard features of various types of power equipment (such as conductors, insulators, and fittings) in the standard library. If a match is successful, the equipment corresponding to the bounding box is assigned a precise category label. After completing the labeling of all single frames, these images with bounding boxes and category labels are integrated to obtain the labeled image of the transmission line.

[0085] In a specific embodiment, the step of performing contour detection on the equipment edges in the single-frame power distribution image to obtain the equipment contour image includes:

[0086] The single-frame power distribution image is converted to grayscale to obtain a grayscale power distribution image. Pixel gradients are calculated on the grayscale power distribution image to obtain the gradient value of each pixel. Edge pixels are then selected based on the gradient values.

[0087] Connectivity analysis is performed on the edge pixels, and adjacent pixels with similar gradient values ​​are connected into line segments to obtain edge line segment groups. Based on the edge line segment groups, closed contours are constructed for the equipment edges in a single frame of the power distribution image to obtain the equipment contour image.

[0088] Specifically, the first step is to extract the power distribution line video frame by frame. Using video analysis tools, the continuous video stream is broken down into independent single-frame power distribution images. It's crucial to preserve the temporal order of the frames to avoid logical inconsistencies in subsequent processing. Next, contour detection of equipment edges is performed on each single-frame power distribution image. This begins with grayscale conversion of the single-frame image. The RGB channel values ​​of the color image are converted to a single grayscale value according to a specific ratio to obtain a grayscale power distribution image, simplifying the complexity of subsequent pixel analysis. Then, pixel gradient calculations are performed on the grayscale power distribution image. Common algorithms calculate the rate of grayscale change of each pixel in the horizontal and vertical directions, thus obtaining the gradient value of each pixel. A reasonable gradient threshold is then set to filter out edge pixels whose gradient values ​​exceed the threshold—for example, at the boundary between the conductor and the background, where the pixel grayscale change is drastic and the gradient value is significantly higher. These points are then identified as edge pixels. After filtering edge pixels, connectivity analysis is performed. Based on the spatial relationship of pixels, adjacent pixels with similar gradient values ​​are connected into continuous line segments, forming edge segment groups. Then, an algorithm is used to fit and complete these edge segment groups, connecting discrete line segments into complete closed curves. This completes the construction of the closed contours of equipment edges in a single frame of the power distribution image, ultimately obtaining an equipment contour image that highlights the equipment's shape. Based on this, preliminary classification of equipment in the transmission line is made according to the differences in the shape of the equipment contours. Linear, slender contours are likely identified as transmission conductors, while umbrella-shaped or columnar contours are initially classified as insulators. This forms a preliminary classification of the equipment images. Next, bounding box drawing is performed. The core principle is to use the outermost edge of the equipment outline as a reference and extend it outward by a certain number of pixels to draw a rectangular bounding box. For example, if the coordinates of the outermost edge of the insulator outline are (x1, y1) to (x2, y2), it can be extended outward by 5 pixels, setting the bounding box coordinates to (x1-5, y1-5) to (x2+5, y2+5), ensuring complete coverage of the equipment and surrounding defect areas, ultimately obtaining an image of the equipment with a bounding box. The final step is to perform category matching based on a preset standard library of power equipment. The equipment features in the bounding box image are compared with the standard features of various types of power equipment (such as conductors, insulators, and fittings) in the standard library. If the match is successful, the equipment corresponding to the bounding box is assigned a precise category label. After completing the labeling of all single frames, these images with bounding boxes and category labels are integrated to obtain the labeled image of the transmission line. In addition, pixels with similar gradient values ​​are specifically defined as: the absolute value of the difference in gradient values ​​between two adjacent edge pixels does not exceed a preset gradient threshold. The preset gradient threshold can be adaptively adjusted according to the actual imaging quality of the power distribution screen (such as sharpness and noise level), and the value range is generally 5-20 (corresponding to the gradient value quantization range of 0-255 for an 8-bit grayscale image).When the gradient values ​​of two adjacent edge pixels meet this condition, they are determined to have "similar gradient values" and are regarded as continuous pixels on the same device edge, and then connected to form a continuous line segment; if the difference in gradient values ​​exceeds a preset threshold, they are determined to belong to different edge structures and are not connected.

[0089] In a specific embodiment, the marked image of the transmission line is input into a preset distribution line inspection model to determine whether the transmission line has any faults or defects, including:

[0090] The labeled image of the transmission line is input into a preset distribution line inspection model. The preset distribution line inspection model includes the Backbone, Neck, and Head parts of the original YOLOv8s model, a C2f_Faster lightweight feature extraction module, a SimSPPF simplified spatial pyramid pooling module, and a NAM normalized attention module. The C2f_Faster lightweight feature extraction module replaces the C2f modules in the Backbone and Neck parts with C2f_Faster modules. The SimSPPF simplified spatial pyramid pooling module replaces the SPPF module in the Backbone part with a SimSPPF module and adds a NAM attention mechanism between the SimSPPF module and the Neck part.

[0091] The labeled image of the transmission line is input into the C2f_Faster lightweight feature extraction module for partial channel convolution and feature stitching to obtain the feature map of the lightweight equipment.

[0092] The feature map of the lightweight device is input into the SimSPPF simplified spatial pyramid pooling module for multi-scale spatial pooling processing to obtain a multi-scale pooling feature map. The multi-scale pooling feature map is then subjected to cross-scale feature stitching processing to obtain a spatial pyramid fused feature map.

[0093] The spatial pyramid fusion feature map is input into the NAM normalized attention module for channel attention weighting processing to obtain a channel weighted feature map, and the channel weighted feature map is then subjected to spatial attention weighting processing to obtain an attention-enhanced feature map.

[0094] The attention-enhanced feature map is input into the Head section for bounding box regression and fault category prediction to determine whether the transmission line has any faults or defects.

[0095] Specifically, the annotated images of transmission lines are input into a pre-defined distribution line inspection model to determine whether there are faults or defects in the transmission lines. The entire process must strictly follow the connection logic of each module in the model. First, the previously obtained annotated images of transmission lines are directly imported into the pre-defined distribution line inspection model. It should be noted that this model is an optimized version of the original YOLOv8s model, retaining the Backbone, Neck, and Head parts of the original model. At the same time, a new C2f_Faster lightweight feature extraction module, a SimSPPF simplified spatial pyramid pooling module, and a NAM normalized attention module are added. The specific optimization method is to replace the original C2f module in the Backbone and Neck parts with the C2f_Faster module, replace the SPPF module in the Backbone part with the SimSPPF module, and embed a NAM attention mechanism between the SimSPPF module and the Neck part. After the image is input, it first enters the C2f_Faster lightweight feature extraction module. This module performs partial channel convolution processing on the labeled image—for example, only performing convolution operations on the channels corresponding to device features in the image to reduce redundant calculations. Then, the features extracted from different levels are concatenated to finally obtain a lightweight device feature map that balances efficiency and information integrity. Next, the lightweight device feature map is input into the SimSPPF simplified spatial pyramid pooling module. Inside the module, multi-scale spatial pooling is performed on the feature map using pooling kernels of different scales. For example, 1×1, 3×3, and 5×5 pooling kernels are used to extract local details and global features respectively, resulting in multi-scale pooled feature maps. Subsequently, cross-scale feature concatenation is performed on these pooled feature maps of different scales, fusing small-scale detail features with large-scale global features to form a spatial pyramid fused feature map. Next, the spatial pyramid fused feature map is fed into the NAM normalized attention module. First, channel attention weighting is performed, assigning weights based on the importance of different channel features. For example, feature channels corresponding to critical defects like insulator cracks are given higher weights, resulting in a channel-weighted feature map. Then, spatial attention weighting is applied to the channel-weighted feature map, focusing on the spatial region where the equipment defect occurs, strengthening the target region features, and suppressing background interference, ultimately outputting an attention-enhanced feature map. Finally, the attention-enhanced feature map is input into the model's Head section. The Head section performs bounding box regression and fault category prediction. Bounding box regression corrects the position of bounding boxes in the labeled image to ensure accurate location of the equipment region. Fault category prediction compares the enhanced features with a pre-set fault feature library, such as determining whether there are defects like broken conductor strands or damaged insulators. Through these two processes, the presence of faults in the transmission line can be ultimately determined.

[0096] In a specific embodiment, a defect status assessment of the transmission line is performed based on the motion trajectory, generating an inspection report for the transmission line, including:

[0097] The motion trajectory of the fault defect in the power distribution line video is extracted to obtain a set of trajectory points, and each trajectory point in the set of trajectory points is timestamped to obtain a set of timestamped trajectory points.

[0098] Based on the set of timestamped trajectory points, the movement speed of the transmission line is calculated to obtain the movement speed sequence of the fault in each time period. Based on the movement speed sequence, the speed change of the transmission line is analyzed to obtain the speed change characteristic curve.

[0099] The defect status of the transmission line is evaluated based on the speed change characteristic curve to obtain the defect status evaluation result, and an inspection report of the transmission line is generated based on the defect status evaluation result.

[0100] Specifically, the process of assessing the status of transmission line defects and generating inspection reports based on the motion trajectory of faults and defects in distribution line videos requires close integration of trajectory tracking results and assessment analysis logic. First, trajectory points are extracted from the acquired fault and defect motion trajectories. Key coordinate points are then selected from continuous trajectory curves using trajectory parsing tools, forming a set of trajectory points. Next, each trajectory point in the set is labeled with a corresponding timestamp. This timestamp must strictly correspond to the frame sequence of the distribution line video. For example, trajectory point A corresponds to 0.02 seconds of frame 100, and trajectory point B corresponds to 0.03 seconds of frame 150, thus obtaining a set of timestamped trajectory points. Based on this, the movement speed of transmission line faults is calculated using a set of timestamped trajectory points. Specifically, the displacement distance is calculated by the coordinate difference between two adjacent trajectory points, and then divided by the timestamp difference between the two trajectory points to obtain the instantaneous speed for adjacent time periods. Arranging these instantaneous speeds in chronological order forms a sequence of the movement speed of faults in each time period. Then, speed change analysis is performed on the speed sequence, and visualization tools are used to plot the speed values ​​and corresponding timestamps as curves. For example, in windy weather, the speed of a conductor strand breakage defect initially increases and then stabilizes, resulting in a curve that first rises and then flattens out – this is the speed change characteristic curve. Finally, a transmission line defect status assessment is conducted based on the speed change characteristic curve. Speed ​​thresholds are set according to power equipment safety operation standards. If a segment of the characteristic curve exceeds the threshold and lasts for a long time, the defect is considered to be in a severe state, posing a risk of equipment failure; otherwise, it is considered a general defect. This yields the defect status assessment result. The assessment result is then integrated with previous defect type, location, and trajectory information. Following the standardized format of power industry inspection reports, the assessment basis, defect level, and maintenance recommendations are clearly listed, ultimately generating a complete transmission line inspection report.

[0101] In a specific embodiment, the defect status of the transmission line is evaluated based on the speed change characteristic curve to obtain a defect status evaluation result, including:

[0102] Peak points are extracted from the velocity change characteristic curve to obtain a velocity peak sequence, and the slope of the velocity peak sequence is calculated to obtain a velocity change slope map including the positive and negative values ​​of the slope and the distribution of inflection points.

[0103] Based on the velocity change slope diagram, the fault is dynamically quantified to obtain a defect development score. Based on the defect development score, the transmission line is classified into states to obtain a defect state assessment result.

[0104] Specifically, the process of assessing the status of transmission line defects and generating inspection reports based on the motion trajectory of faults and defects in distribution line videos requires close integration of trajectory tracking results and assessment analysis logic. First, trajectory points are extracted from the acquired fault and defect motion trajectories. Key coordinate points are then selected from continuous trajectory curves using trajectory parsing tools, forming a set of trajectory points. Next, each trajectory point in the set is labeled with a corresponding timestamp. This timestamp must strictly correspond to the frame sequence of the distribution line video. For example, trajectory point A corresponds to 0.02 seconds of frame 100, and trajectory point B corresponds to 0.03 seconds of frame 150, thus obtaining a set of timestamped trajectory points. Based on this, the movement speed of transmission line faults is calculated using a set of timestamped trajectory points. Specifically, the displacement distance is calculated by the coordinate difference between two adjacent trajectory points, and then divided by the timestamp difference between the two trajectory points to obtain the instantaneous speed in adjacent time periods. Arranging these instantaneous speeds in chronological order forms the movement speed sequence of the fault in each time period. Then, the speed sequence is analyzed for speed changes, and visualization tools are used to plot the speed values ​​and corresponding timestamps as curves. For example, for conductor strand breakage defects, the speed first increases and then tends to stabilize in windy weather, and the curve will show a shape of first rising and then flattening out. This is the speed change characteristic curve. Finally, based on the velocity change characteristic curve, a defect status assessment of the transmission line is conducted. Specifically, the peak points of the velocity change characteristic curve are first extracted. Using curve analysis tools, peak points with velocity values ​​higher than their neighboring points are selected and arranged chronologically to form a velocity peak sequence. Next, the slope of the velocity peak sequence is calculated by dividing the velocity difference between two adjacent peak points by the time difference. Simultaneously, the positive and negative values ​​of the slope (positive values ​​represent velocity increase, negative values ​​represent velocity decrease) and the distribution of inflection points where the slope changes from positive to negative or vice versa are recorded. Based on this, a velocity change slope graph containing the positive and negative values ​​of the slope and the distribution of inflection points is plotted. On this basis, dynamic quantitative analysis of faults and defects is performed based on the velocity change slope graph, combined with preset scoring rules—for example, a larger positive slope value and a denser number of inflection points indicate more intense defect movement and a faster development trend, corresponding to a higher defect development score; conversely, a lower score is obtained through this quantitative calculation. Finally, the transmission lines are classified according to their defect development scores. For example, scores above 80 are considered serious defects, scores between 40 and 80 are considered general defects, and scores below 40 are considered minor defects. This yields the defect status assessment results. The assessment results are then integrated with previous defect type, location, and trajectory information. Following the standard format of power industry inspection reports, the assessment basis, defect level, and maintenance recommendations are clearly listed, ultimately generating a complete inspection report for the transmission lines.

[0105] The above describes a power transmission line inspection method in an embodiment of the present invention. The following describes a power transmission line inspection system in an embodiment of the present invention. Please refer to [link / reference]. Figure 2One embodiment of the power transmission line inspection system in this invention includes:

[0106] The acquisition module 21 is used to acquire the distribution line video of the transmission line, and to annotate the distribution line video with bounding boxes and equipment categories to obtain the transmission line annotated image;

[0107] The judgment module 22 is used to input the marked image of the transmission line into a preset distribution line inspection model to determine whether there are faults or defects in the transmission line.

[0108] The tracking module 23 is used to input the fault defect into a preset DeepSORT model if it exists. The preset DeepSORT model includes a feature extraction module, a state prediction module, and a trajectory association module.

[0109] The fault defect is input into the feature extraction module for feature vector calculation to obtain the defect feature description vector;

[0110] The defect feature description vector is input into the state prediction module to estimate the motion state and obtain the defect motion parameters.

[0111] The defect motion parameters are input into the trajectory association module for time-series association analysis to obtain the motion trajectory of the fault defect in the power distribution line video.

[0112] The evaluation module 24 is used to evaluate the defect status of the transmission line based on the motion trajectory and generate an inspection report of the transmission line.

[0113] In this embodiment, the specific implementation of each module in the above system embodiment is described in the above method embodiment, and will not be repeated here.

Claims

1. A method for inspecting power transmission lines in a power scenario, characterized in that, Includes the following steps: Acquire video of the distribution line of the transmission line, and annotate the distribution line video with bounding boxes and equipment categories to obtain an annotated image of the transmission line; The labeled image of the transmission line is input into a preset power distribution line inspection model to determine whether there are any faults or defects in the transmission line. If it exists, the fault defect is input into the preset DeepSORT model, which includes a feature extraction module, a state prediction module, and a trajectory association module. The fault defect is input into the feature extraction module for feature vector calculation to obtain the defect feature description vector; The defect feature description vector is input into the state prediction module to estimate the motion state and obtain the defect motion parameters. The defect motion parameters are input into the trajectory association module for time-series association analysis to obtain the motion trajectory of the fault defect in the power distribution line video. Based on the motion trajectory, the defect status of the transmission line is assessed, and an inspection report of the transmission line is generated. The labeled image of the transmission line is input into a preset distribution line inspection model to determine whether the transmission line has any faults or defects, including: The labeled image of the transmission line is input into a preset distribution line inspection model. The preset distribution line inspection model includes the Backbone, Neck, and Head parts of the original YOLOv8s model, a C2f_Faster lightweight feature extraction module, a SimSPPF simplified spatial pyramid pooling module, and a NAM normalized attention module. The C2f_Faster lightweight feature extraction module replaces the C2f module in the Backbone and Neck parts with the C2f_Faster module. The SimSPPF simplified spatial pyramid pooling module replaces the SPPF module in the Backbone part with the SimSPPF module and adds a NAM attention mechanism between the SimSPPF module and the Neck part. The labeled image of the transmission line is input into the C2f_Faster lightweight feature extraction module for partial channel convolution and feature stitching to obtain the feature map of the lightweight equipment. The feature map of the lightweight device is input into the SimSPPF simplified spatial pyramid pooling module for multi-scale spatial pooling processing to obtain a multi-scale pooling feature map. The multi-scale pooling feature map is then subjected to cross-scale feature stitching processing to obtain a spatial pyramid fused feature map. The spatial pyramid fusion feature map is input into the NAM normalized attention module for channel attention weighting processing to obtain a channel weighted feature map, and the channel weighted feature map is then subjected to spatial attention weighting processing to obtain an attention-enhanced feature map. The attention-enhanced feature map is input into the Head section for bounding box regression and fault category prediction to determine whether the transmission line has any faults or defects.

2. The power transmission line inspection method according to claim 1, characterized in that, The acquisition of the distribution line video of the transmission line includes: Video acquisition equipment was deployed at key detection points on the power transmission line to obtain a monitoring and acquisition scheme. Based on the aforementioned monitoring and acquisition scheme, multi-angle video acquisition is performed on the transmission lines to obtain the original video of the distribution lines. The original video of the power distribution line is processed with image stabilization to obtain the power distribution line video.

3. The power transmission line inspection method according to claim 2, characterized in that, Based on the aforementioned monitoring and acquisition scheme, multi-angle video acquisition is performed on the transmission lines to obtain the original video of the distribution lines, including: Based on the monitoring and acquisition scheme, the parameters of each preset video acquisition device are configured, the acquisition frame rate, resolution and acquisition angle range are set to obtain the configured acquisition device group, and the preliminary video acquisition of the transmission line at different angles is carried out through the configured acquisition device group to obtain multiple sets of preliminary angle videos. The video quality of the multiple sets of preliminary angle videos is checked, and videos with clear images, no obstructions, and angles that meet the preset requirements are selected to obtain qualified angle video groups. Then, the time synchronization processing of each video in the qualified angle video group is performed to make the time start point of each video consistent, so as to obtain the original video of the power distribution line.

4. The power transmission line inspection method according to claim 1, characterized in that, The video of the power distribution line is bounding box and equipment category is labeled to obtain a labeled image of the power transmission line, including: The video of the power distribution line is extracted frame by frame to obtain a single frame of power distribution image, and the equipment edges in the single frame of power distribution image are contour detected to obtain the equipment contour image. Based on the equipment outline image, the equipment in the transmission line is initially classified to obtain a preliminary classified equipment image. Boundary boxes are drawn for the equipment in the preliminary classified equipment image. A rectangular boundary box is drawn by extending a certain pixel distance outward from the outermost edge of the equipment outline to obtain a bounding box equipment image. Based on a preset standard image library of power equipment, the bounding box equipment images are classified and matched to obtain labeled images of transmission lines.

5. The power transmission line inspection method according to claim 4, characterized in that, Contour detection is performed on the equipment edges in the single frame of the power distribution image to obtain equipment contour images, including: The single-frame power distribution image is converted to grayscale to obtain a grayscale power distribution image. Pixel gradients are calculated on the grayscale power distribution image to obtain the gradient value of each pixel. Edge pixels are then selected based on the gradient values. Connectivity analysis is performed on the edge pixels, and adjacent pixels with similar gradient values ​​are connected into line segments to obtain edge line segment groups. Based on the edge line segment groups, closed contours are constructed for the equipment edges in a single frame of the power distribution image to obtain the equipment contour image.

6. The power transmission line inspection method according to claim 1, characterized in that, Based on the motion trajectory, a defect status assessment of the transmission line is performed, generating an inspection report for the transmission line, including: The motion trajectory of the fault defect in the power distribution line video is extracted to obtain a set of trajectory points, and each trajectory point in the set of trajectory points is timestamped to obtain a set of timestamped trajectory points. Based on the set of timestamped trajectory points, the movement speed of the transmission line is calculated to obtain the movement speed sequence of the fault in each time period. Based on the movement speed sequence, the speed change of the transmission line is analyzed to obtain the speed change characteristic curve. The defect status of the transmission line is evaluated based on the speed change characteristic curve to obtain the defect status evaluation result, and an inspection report of the transmission line is generated based on the defect status evaluation result.

7. The power transmission line inspection method according to claim 6, characterized in that, The defect status of the transmission line is assessed based on the aforementioned velocity change characteristic curve, resulting in a defect status assessment, including: Peak points are extracted from the velocity change characteristic curve to obtain a velocity peak sequence, and the slope of the velocity peak sequence is calculated to obtain a velocity change slope map including the positive and negative values ​​of the slope and the distribution of inflection points. Based on the speed change slope diagram, the fault is dynamically quantified to obtain a defect development score. Based on the defect development score, the transmission line is classified into states to obtain a defect state assessment result.

8. A power transmission line inspection system for a power scenario, characterized in that, The method for inspecting power transmission lines in a power scenario according to any one of claims 1 to 7 includes: The acquisition module is used to acquire video of the distribution line of the transmission line, and to annotate the distribution line video with bounding boxes and equipment categories to obtain an annotated image of the transmission line; The judgment module is used to input the marked image of the transmission line into a preset distribution line inspection model to determine whether the transmission line has any faults or defects. The tracking module is used to input the fault defect into a preset DeepSORT model if it exists. The preset DeepSORT model includes a feature extraction module, a state prediction module, and a trajectory association module. The fault defect is input into the feature extraction module for feature vector calculation to obtain the defect feature description vector; The defect feature description vector is input into the state prediction module to estimate the motion state and obtain the defect motion parameters. The defect motion parameters are input into the trajectory association module for time-series association analysis to obtain the motion trajectory of the fault defect in the power distribution line video. The evaluation module is used to evaluate the defect status of the transmission line based on the motion trajectory and generate an inspection report of the transmission line. The labeled image of the transmission line is input into a preset distribution line inspection model to determine whether the transmission line has any faults or defects, including: The labeled image of the transmission line is input into a preset distribution line inspection model. The preset distribution line inspection model includes the Backbone, Neck, and Head parts of the original YOLOv8s model, a C2f_Faster lightweight feature extraction module, a SimSPPF simplified spatial pyramid pooling module, and a NAM normalized attention module. The C2f_Faster lightweight feature extraction module replaces the C2f module in the Backbone and Neck parts with the C2f_Faster module. The SimSPPF simplified spatial pyramid pooling module replaces the SPPF module in the Backbone part with the SimSPPF module and adds a NAM attention mechanism between the SimSPPF module and the Neck part. The labeled image of the transmission line is input into the C2f_Faster lightweight feature extraction module for partial channel convolution and feature stitching to obtain the feature map of the lightweight equipment. The feature map of the lightweight device is input into the SimSPPF simplified spatial pyramid pooling module for multi-scale spatial pooling processing to obtain a multi-scale pooling feature map. The multi-scale pooling feature map is then subjected to cross-scale feature stitching processing to obtain a spatial pyramid fused feature map. The spatial pyramid fusion feature map is input into the NAM normalized attention module for channel attention weighting processing to obtain a channel weighted feature map, and the channel weighted feature map is then subjected to spatial attention weighting processing to obtain an attention-enhanced feature map. The attention-enhanced feature map is input into the Head section for bounding box regression and fault category prediction to determine whether the transmission line has any faults or defects.

Images