Power transmission line external damage prevention detection method, device, equipment and medium

Abstract

The invention discloses a power transmission line external damage prevention detection method, device and equipment and a medium, and the method comprises the steps: constructing a power transmission line external damage prevention detection model which is used for recognizing and positioning a hidden danger target corresponding to a power transmission line; obtaining a monitored image and point cloud data of the power transmission line; inputting the monitored image into the power transmission line external damage prevention detection model to identify and position a hidden danger target, and outputting two-dimensional pixel coordinate data of the hidden danger target in an image plane coordinate system; determining a spatial distance between the power transmission line and the hidden danger target based on the monitored image, the two-dimensional pixel coordinate data and the point cloud data; and based on the spatial distance, evaluating the external damage risk of the power transmission line. Automatic monitoring and early warning of the power transmission line are achieved, the response speed and processing efficiency of external damage prevention of the power transmission line are greatly improved, and a powerful guarantee is provided for safe and stable operation of the power transmission line.

Description

Technical Field

[0001] This invention relates to the field of power transmission line safety monitoring technology, and in particular to a method, device, equipment and medium for detecting external damage to power transmission lines. Background Technology

[0002] With the increasing complexity of the external environment of power transmission lines, the risk of damage from external forces has significantly increased, making the monitoring of potential external damage to power transmission lines increasingly important. Currently, the monitoring of potential external damage to power transmission lines mainly relies on manual foot patrols and helicopter inspections. These methods generally suffer from many drawbacks, such as low inspection efficiency, insufficient real-time performance, and high labor costs. Although the rapid development of drone technology and intelligent sensing technology in recent years has promoted the application of new inspection models and optimized the detection process to some extent, the existing regular inspection mechanism is still insufficient to meet the urgent needs of the power system for real-time monitoring of potential hazards. Summary of the Invention

[0003] This invention provides a method, device, electronic equipment, and medium for detecting external damage to power transmission lines, in order to solve the technical problems of traditional manual foot patrols and helicopter inspections, which have prominent drawbacks such as low inspection efficiency, insufficient real-time performance, and high labor costs.

[0004] Firstly, a method for detecting external damage to transmission lines is provided, including:

[0005] A detection model for external damage prevention of transmission lines is constructed. This model is used to identify and locate potential hazards associated with transmission lines.

[0006] Acquire surveillance images and point cloud data of power transmission lines;

[0007] The monitoring images are input into the transmission line external damage prevention and detection model to identify and locate potential hazards, and the two-dimensional pixel coordinate data of the potential hazards in the image plane coordinate system are output.

[0008] Based on surveillance images, two-dimensional pixel coordinate data, and point cloud data, the spatial distance between the power transmission line and the potential hazard is determined.

[0009] Based on spatial distance, assess the risk of external damage to transmission lines.

[0010] Secondly, a power transmission line external damage detection device is provided, comprising:

[0011] The module is used to build a detection model for external damage to transmission lines. This model is used to identify and locate potential hazards associated with transmission lines.

[0012] The acquisition module is used to acquire surveillance images and point cloud data of transmission lines;

[0013] The generation module is used to input the surveillance images into the transmission line external damage prevention detection model to identify and locate potential hazards, and output the two-dimensional pixel coordinate data of the potential hazards in the image plane coordinate system;

[0014] The determination module is used to determine the spatial distance between the transmission line and the potential hazard target based on the surveillance images, two-dimensional pixel coordinate data and point cloud data;

[0015] The risk assessment module is used to assess the risk of external damage to transmission lines based on spatial distance.

[0016] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for detecting external damage to transmission lines.

[0017] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described method for detecting external damage to transmission lines.

[0018] The aforementioned method, device, electronic equipment, and storage medium for detecting external damage to transmission lines first utilize deep learning to achieve real-time detection and localization of potential hazards. Then, high-precision laser point cloud data is fused with two-dimensional images to obtain the three-dimensional spatial distance from the potential hazard to the power line. Finally, based on this three-dimensional spatial distance, the existence of a potential external force damage risk is determined. This achieves automatic monitoring and early warning of transmission lines, significantly improving the response speed and processing efficiency for preventing external damage to transmission lines, and providing a strong guarantee for the safe and stable operation of transmission lines. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a method for detecting external damage to transmission lines in one embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the result of Zero-DCE++ in one embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of the C3Gnet module in one embodiment of the present invention;

[0023] Figure 4 This is a schematic diagram of the C3 structure of YOLOv5 in one embodiment of the present invention;

[0024] Figure 5 This is a schematic diagram of the backbone network structure of a transmission line external damage detection model in one embodiment of the present invention;

[0025] Figure 6 This is a schematic diagram of the experimental results of CARAFE parameter comparison in one embodiment of the present invention;

[0026] Figure 7 This is a schematic diagram of comparative experimental results of the backbone network in one embodiment of the present invention;

[0027] Figure 8 This is a schematic diagram of ablation experiment results with the detection module integrated into different structures in one embodiment of the present invention;

[0028] Figure 9 This is a schematic diagram of the structure of a transmission line external damage detection device in one embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in the present invention are only for illustrative and descriptive purposes and are not intended to limit the scope of protection of the present invention.

[0030] Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or performed simultaneously. Moreover, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0031] Furthermore, the embodiments described herein are merely some, not all, of the embodiments of the invention. 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.

[0032] It should be noted that the term "comprising" will be used in the embodiments of the present invention to indicate the presence of a feature subsequently declared, but does not exclude the addition of other features. It should also be noted that similar reference numerals 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.

[0033] The following is a detailed description of this case, in conjunction with the relevant accompanying drawings in the instruction manual.

[0034] In the embodiments of this specification, traditional power transmission line hazard monitoring mainly relies on manual foot patrols and helicopter inspections, which have significant limitations such as low inspection efficiency, insufficient real-time performance, and high labor costs. Although the development of drone technology and intelligent sensing technology has promoted the application of new inspection modes and optimized the detection process to some extent, the current periodic inspection mechanism is still difficult to meet the urgent need for real-time hazard monitoring. In addition, some studies use binocular vision algorithms to measure the distance between hazard targets and power transmission lines. Although this method can achieve three-dimensional monitoring, it has extremely high requirements for camera calibration accuracy, requires maintaining a long baseline distance, and even a small shift in the relative position of the camera will lead to a significant decrease in ranging accuracy. While drone-based LiDAR (Light Detection and Ranging) point cloud processing technology can accurately assess the threat level of hazards based on actual distance, it suffers from practical problems such as large data processing latency, high equipment costs, and difficulty in large-scale deployment.

[0035] Furthermore, monocular cameras have been widely used in the field of intelligent monitoring of power transmission lines due to their advantages such as low cost and convenient deployment. However, the related technologies for monitoring potential hazards such as construction machinery on power transmission lines based on monocular vision still have significant shortcomings: First, traditional monocular vision detection methods cannot acquire three-dimensional distance information, making it difficult to accurately assess the risk level based on power safety distance standards, leading to frequent false alarms; second, current mainstream intelligent detection models often adopt complex network structures to improve accuracy, which not only makes it difficult to adapt to resource-constrained edge computing devices, but also results in poor detection of potential hazards that occupy a small proportion in power transmission line images; finally, in nighttime construction scenarios, the visual features of potential hazards are significantly weakened due to lighting conditions, causing a sharp decline in the detection performance of existing algorithms. It is evident that current technologies for monitoring potential hazards on power transmission lines still face key bottlenecks, and there is an urgent need for an intelligent detection and early warning method that combines high precision, lightweight design, and all-weather adaptability.

[0036] To address the aforementioned issues, this application proposes a method for detecting external damage to power transmission lines. First, it utilizes deep learning to achieve real-time detection and localization of potential hazards. Then, it fuses high-precision laser point cloud data with two-dimensional images to obtain the three-dimensional spatial distance from the hazard to the power line. Finally, based on this three-dimensional spatial distance, it determines whether there is a risk of external force damage. This method enables automatic monitoring and early warning of power transmission lines, significantly improving the response speed and processing efficiency for preventing external damage, and providing a strong guarantee for the safe and stable operation of power transmission lines.

[0037] Please see Figure 1 This specification provides a method for detecting external damage to transmission lines, which specifically includes the following steps:

[0038] S10: Construct a detection model for external damage to transmission lines. This model is used to identify and locate potential hazards associated with transmission lines.

[0039] It is understood that the executing entity of this invention can be a transmission line external damage detection device, a terminal, or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0040] In this step, a detection model for external damage to transmission lines is constructed. This model has the ability to identify and locate various hidden potential hazards around the transmission lines and determine their specific locations.

[0041] In one embodiment of this application, a specific scheme for constructing a transmission line external damage prevention detection model is provided. In S10, that is, constructing the transmission line external damage prevention detection model, the following steps S11-S13 are specifically included:

[0042] S11: Construct a sample dataset of power transmission line protection against external damage.

[0043] In this step, a sample dataset for external force damage to transmission lines is constructed. This dataset systematically collects a massive number of representative images of potential hazards to transmission lines. Each sample case fully records various typical scenarios of external force damage, including construction machinery intrusion, illegal buildings, and tree obstructions, comprehensively covering all types of risk conditions that may endanger the safe operation of transmission lines. The large amount of sample data makes the subsequent model training process more stable. The diversity and richness of the data can reduce the model's over-reliance on individual samples and lower the risk of overfitting. Overfitting can cause a model to perform well on training data but its performance will drop sharply on new data. A rich dataset allows the model to learn more general features and patterns, thereby improving its reliability in practical applications.

[0044] In one embodiment of this application, a specific sample dataset construction scheme is provided. In S11, that is, constructing a sample dataset for protection against external damage to transmission lines, the following steps S111-S114 are specifically included:

[0045] S111: Acquire multiple hazard images corresponding to the power transmission lines in the target area.

[0046] In this step, a multi-source acquisition method is used to construct an image library of potential hazards in the target area's power transmission lines. These hazard images can be acquired through various means, such as monitoring cameras installed on power poles, data on power transmission lines and their surrounding environment captured by drone inspections, and can also be combined with previously collected power scene images from different sources and time periods. These power scene images cover a variety of typical external force damage hazard scenarios, such as construction machinery intrusion, illegal buildings, and tree obstructions, providing rich and diverse visual features for the model. The collected two-dimensional hazard images completely cover various hazard targets existing in the target area's power transmission lines and surrounding environment, including typical external force damage hazard targets such as construction machinery (e.g., cranes, excavators), foreign objects on the lines (kite strings, plastic film), and illegal buildings, providing comprehensive data support for subsequent intelligent detection algorithm training.

[0047] In practical applications, high-resolution cameras are often used for long-distance monitoring of power transmission lines. The resolution of the monitored scene images used in this application is mainly 2560×1440, with some at 3840×2160 and 1980×1080. Furthermore, LiDAR point clouds are used as ranging verification labels to quantitatively evaluate the ranging accuracy of the proposed method.

[0048] S112: Preprocess each hazard image to remove invalid images, and label each preprocessed hazard image. The label includes the hazard type and location information of the hazard target in the hazard image.

[0049] In this step, the acquired raw images may contain a large amount of useless or poor-quality data. Image preprocessing is performed to filter out invalid information. Specifically, blurry, overexposed, or severely obstructed images are excluded, as these are difficult to accurately identify potential hazards from. Furthermore, images unrelated to potential hazards along the power transmission line are discarded; for example, images that only show the sky and ground but lack power lines and surrounding scenery are removed.

[0050] Furthermore, potential hazards in the image are labeled. The labels include hazard type, specifying the exact category of the hazard, such as construction machinery or trees. Additionally, the labels include location information, typically indicated by bounding boxes (rectangles or other polygons) showing the hazard's position in the image. This is defined by providing the coordinates of the top-left and bottom-right corners of the bounding box, or the center point coordinates, width, and height, enabling the model to learn the precise location of the hazard.

[0051] S113: Expand each hazard image after annotation.

[0052] In this step, to enhance the diversity and quantity of sample data and improve the model's generalization ability, the labeled images are augmented through image transformation and noise addition. Specifically, image transformation refers to generating new images through operations such as rotation, flipping, scaling, translation, and cropping. For example, rotating the original image 90 degrees clockwise or flipping it horizontally changes the appearance of the image, but the essential characteristics of the potential target remain unchanged. Adding noise involves adding an appropriate amount of random noise to the image to simulate interference situations that may be encountered in practical applications, such as Gaussian noise.

[0053] S114: Based on the expanded images of multiple potential hazards, construct a sample dataset for preventing external damage to power transmission lines.

[0054] In this step, a large number of two-dimensional hazard images that have been screened, labeled and expanded are integrated to build a database specifically for the identification and analysis of hazard targets in power transmission channels. These hazard images are placed in the database to form a sample dataset for preventing external damage to power transmission lines.

[0055] By using the above methods, a sample dataset for preventing external damage to transmission lines is constructed, providing high-quality, large-scale training data for the subsequent development of transmission line external damage detection models, supporting model training and optimization, and thus more accurately identifying and locating potential hazards in transmission channels.

[0056] S12: Construct an optimized YOLOv7-tiny detector.

[0057] In this step, to achieve more accurate and efficient monitoring of external damage to transmission lines, a YOLOv7-tiny detector was constructed and optimized. Based on the original YOLOv7-tiny architecture, and considering the unique requirements of transmission line external damage prevention scenarios, comprehensive optimizations were made across multiple dimensions, including algorithm improvement, parameter tuning, and feature extraction optimization. This resulted in a better balance between detection accuracy, speed, and stability. The optimized YOLOv7-tiny detector served as the neural network for detecting external damage to transmission lines. This neural network fully leverages the advantages of the optimized detector, deeply integrates relevant data features of the transmission lines, and constructs a highly adaptable and reliable intelligent detection system. This system can quickly and accurately identify various external damaging factors that may endanger the safety of transmission lines, thus safeguarding the stable operation of the transmission lines.

[0058] In one embodiment of this application, a specific neural network construction scheme is provided. In S12, namely, constructing an optimized YOLOv7-tiny detector, the following steps S121-S122 are specifically included:

[0059] S121: Construct the original architecture of the YOLOv7-tiny detector.

[0060] S122: Integrates the LLIE module, C3GNet module, upsampling operator CARAFE, and Involution module to optimize the YOLOv7-tiny detector, generating an optimized YOLOv7-tiny detector.

[0061] For steps S121-S122, in order to build a high-performance object detection framework, the original architecture of the YOLOv7-tiny detector is first constructed, which includes core components such as convolutional layers, pooling layers, and activation functions. This ensures that the original architecture has efficient feature extraction and object detection capabilities, laying a solid foundation for subsequent optimization work.

[0062] Furthermore, after completing the original architecture, to further improve the detection accuracy, speed, and generalization ability of the YOLOv7-tiny detector in complex scenes, the LLIE (Low-Light Image Enhancement) module was integrated into the architecture. This module effectively improves the lighting conditions of images, enhances the feature representation of targets in images, and enables the detector to accurately capture target information under different lighting conditions. By adding the LLIE module, the problem of the visual algorithm's detection performance deviation in dark scenes is compensated for.

[0063] In practical applications, the LLIE module is optimized and improved based on the Zero-DCE++ model. Zero-DCE++ designs a learnable illumination enhancement curve function and constructs a lightweight deep neural network to dynamically predict image curve parameter mappings. The network architecture employs an efficient structure consisting of 7 layers of depthwise separable convolutions, capable of generating adaptive parameter prediction results for each of the RGB channels. In terms of training strategy, the model comprehensively utilizes four constraints: spatial consistency loss, exposure control loss, color constancy loss, and illumination smoothness loss, to jointly optimize the network parameters. The mathematical expression for the illumination enhancement curve is:

[0064] E t (x)=E t-1 (x)+a t (x)E t-1 (x)(1-E t-1 (x));

[0065] Where x represents the pixel coordinate position; t is the iteration number; E t Represents the input pixel value; E t-1 The enhanced output value; a(x) represents the curve amplitude adjustment parameter at position x (its value range is limited to [-1,1]). The illumination enhancement curve defines a high-order nonlinear mapping relationship based on pixel-level estimation. Through this pixel-by-pixel prediction curve parameter mechanism, the model can achieve accurate mapping transformation from low-light images to enhanced images. Zero-DCE++, with its ultra-lightweight network structure (containing only about 0.1M parameters) and unsupervised learning paradigm (no paired training data required), demonstrates significant technical advantages in industrial-grade low-light image enhancement applications.

[0066] Optionally, in practical applications, the transmission line monitoring images processed by Zero-DCE++ suffer from overexposure. This may be due to the larger proportion of sky and higher image resolution in the monitoring images. Therefore, this application adds a Gamma correction operator (GC) to the inference part of Zero-DCE++. The new Zero-GC can improve overexposed images. Within the grayscale values ​​[0, 255], GC is defined as follows:

[0067]

[0068] Where γ is the correction parameter. When γ > 1, it stretches low grayscale areas and compresses high grayscale areas in the image, which can be used to fine-tune low-light images; when γ < 1, high grayscale areas in the image are stretched while low grayscale areas are compressed, which can be used to fine-tune overexposed images. Figure 2As shown, the image optimized by the Zero-DCE++ correction operator (γ>1) reveals more details of clutter in the magnified area, and the contrast of the black car is enhanced. The Gamma correction operator reduces the overexposure effect of the Zero-DCE++ prediction results, improving image clarity. Furthermore, this application incorporates a low-light discrimination module in the model to prevent LLIE processing of images under normal lighting conditions. The grayscale histogram of the nighttime dataset of power transmission lines is analyzed, and a brightness threshold of 39% is set based on the average brightness of the images. The introduction of the Zero-GC module improves the overall accuracy of identifying potential hazards and provides better visualization for nighttime monitoring of power transmission lines. Further, to assess the optimized performance in dark scenes, comparative experiments are conducted on the proposed detector model using the low-light enhancement module, trained and evaluated on a constructed dataset of nighttime monitoring scenes of power transmission lines. The proposed low-light enhancement method, Zero-GC, is trained for 100 epochs. The trained low-light enhancement model has a computational cost of less than 0.12 GFLOPs, only 0.01M parameters, and a memory footprint of only 0.05MB. Therefore, compared to the detector module, the addition of Zero-GC has a negligible impact on the lightweight performance of the proposed method. Compared to other mainstream LLIE methods, Zero-DCE++ achieves the highest detection accuracy at all three thresholds, providing the detector with optimal performance in nighttime hazard detection and enabling efficient 24 / 7 monitoring.

[0069] Furthermore, a novel lightweight feature extraction module, C3GNet, is introduced. Its unique network structure can improve the efficiency and quality of feature extraction, enabling the model to better learn the deep features of the target.

[0070] In practical application scenarios, such as Figure 3The diagram shows the structure of the C3Gnet module. C3GNet uses CBS convolutional blocks and Ghostbottle as its basic components. By combining these units appropriately, a lightweight and efficient feature extraction network is constructed, which can reduce the complexity and computational cost of the model while maintaining a certain feature extraction capability. Specifically, the CBS convolutional block is a classic convolutional module, composed of three key parts arranged in an orderly manner. First is the basic convolutional layer (Conv), which, as the core component of feature extraction, performs convolution operations on the input data to extract feature information at different scales and levels. Following this is the batch normalization layer (BN), whose main function is to normalize the output of the convolutional layer, making the data distribution more stable, effectively accelerating the model training process and improving the model's generalization ability. Finally, there is the SiLU (Sigmoid-Weighted Linear Unit) activation function, which introduces non-linear characteristics to the convolutional block, enabling it to better fit complex data distributions and enhance the model's expressive power. Ghostbottle is a unique bottleneck structure built from Ghost modules. The Ghost module is designed to generate rich feature maps with low computational cost. Through a series of linear transformations and feature fusion operations, it maintains good feature representation capabilities while reducing the number of parameters and computational cost. Ghostbottle leverages these advantages of the Ghost module to construct an efficient bottleneck structure, effectively improving model performance and efficiency in tasks such as object detection. The idea for improving the C3GNet structure originates from the C3 structure of YOLOv5, such as... Figure 4 The diagram shows the C3 structure of YOLOv5, which borrows the ideas of residual and splitting extraction to achieve cross-layer information transfer. This application draws on the experience of DenseNet, adding densely connected branches while removing the convolutional layers of the original branches. This increases the gradient flow information of the network, thereby improving the convergence speed and performance of deep networks, and reducing unnecessary convolutional computations and parameter counts. Furthermore, this application incorporates lightweight Ghost layers in both bottleneck structures. The constructed Ghostbottle possesses the powerful learning capabilities of residual structures, and the multi-channel grouping feature extraction method of Ghost has fewer parameters and computational cost. Therefore, the proposed C3GNet module has rich gradient flow information and a lighter connection structure, and it serves as a basic component of the feature extraction network. Figure 5The diagram shows the backbone network structure of the transmission line external damage detection model. It uses basic convolutional layers to build a four-stage C3GNet, consisting of three core modules: Low-Light Enhancement (LLIE), Detector, and Range. The Low-Light Enhancement (LLIE) module improves image quality under low-light conditions such as nighttime and foggy weather, ensuring the accuracy of subsequent detection. Specifically, it includes three core components: ZeroDCE++ (GC) (a lightweight enhancement model based on illumination curve learning), which uses Gamma Correction to optimize illumination distribution; a Discriminator (adversarially trained to improve the realism of the enhancement effect); and Multi-scene input (supporting diverse inputs for complex scenes such as nighttime and foggy weather). Input is a low-light image (e.g., a nighttime transmission line monitoring image). ZeroDCE++ dynamically predicts the illumination enhancement curve, adjusting image brightness and contrast. The Discriminator is used for adversarial training to optimize the enhancement effect. High-quality images are output for use by the Detector module. The target detection module is used to identify potential hazardous targets in images (such as cranes, drones, floating objects, etc.). Specifically, it includes: D-LDRNet: a lightweight detection network; Backbone: a feature extraction backbone network (C3GNet); Head: a detection head that outputs the target category (e.g., Circle and Invasion, i.e., circular equipment (insulators) or intruding objects (construction machinery)). Specifically, this module takes an enhanced image as input, the Backbone extracts multi-scale features, the Head predicts the target bounding box and category, and outputs the two-dimensional pixel coordinates of the hazardous target (e.g., the crane's position). The risk ranging module (Range) is used to calculate the three-dimensional spatial distance from the hazardous target to the transmission line and assess the risk level. Specifically, it includes: Gamma correction for optical compensation of depth information to improve ranging accuracy; Cubic: three-dimensional spatial modeling; and Tension combined with transmission line mechanical parameters to determine risk. Specifically, the detected two-dimensional pixel coordinates are aligned with the LiDAR point cloud and transformed to a three-dimensional spatial coordinate system using camera calibration parameters. The shortest distance from the target to the power line is calculated using the three-dimensional Euclidean distance formula, and graded warnings are triggered based on a safe distance threshold.

[0071] Furthermore, to optimize the feature map upsampling process, the efficient and lightweight upsampling operator CARAFE (Content-Aware ReAssembly of Features) is employed, and its parameters are optimized to better suit the proposed model and power transmission monitoring scenarios. Compared with traditional upsampling methods, CARAFE can adaptively generate upsampling kernels, thereby more accurately recovering the detailed information of the feature map and avoiding information loss.

[0072] In practical applications, CARAFE achieves feature map upsampling by dynamically predicting and intelligently recombining the recombination kernel. Its specific calculation process can be formally represented as follows:

[0073] D i* =R(N(F) i K conv ));

[0074] Given an input feature map F, with the target position i as the center and a size of K... conv The local neighborhood is denoted as N(F) i K conv Through a specially designed recombinant kernel prediction module R, for each output target location I... * Generate position-adaptive recombination kernel D i* Meanwhile, the perceptual reassembly module T introduced in this architecture utilizes a size of K. re The recombination kernel is used to ultimately output the recombination feature map F. * i* The calculation process is shown in the following formula:

[0075]

[0076] This upsampling mechanism has three significant advantages: First, it achieves dynamic feature recombination through a position-adaptive recombination kernel, effectively improving feature representation capabilities; second, it expands the receptive field to fully integrate contextual information, significantly enhancing the ability to capture features of small targets; and most importantly, this scheme achieves a substantial improvement in detection performance without increasing model computational complexity (FLOPs) and the number of parameters, perfectly meeting the design requirements of lightweight models.

[0077] Optionally, comparative experiments can be conducted on the constructed dataset to explore the optimal parameter combinations that achieve the best performance on both the model and the dataset. Different K values ​​can be set. conv and K re Ten sets of comparative experiments were conducted, such as Figure 6 The image shows a schematic diagram of the CARAFE parameter comparison experiment results. The quantitative evaluation metrics for the detector evaluation mainly include floating-point operations (GFLOPs), number of parameters (Params), memory usage (Mem), and average precision (mAP, mAP) at different thresholds. 75 mAP 50 It can be seen that, with K... conv and K reWith the increase of K, the lightweight indicators GFLOPs, Params, and Mem generally show an increasing trend. However, the lightweight nature of CARAFE also makes their changes very small. Therefore, the detection accuracy without a specific pattern of change is the focus of this application. conv =5, K re When the value is 3, the model achieves the best accuracy performance, and the lightweight performance is not significantly different from the original model. CARAFE compensates for some of the accuracy loss caused by the introduction of C3GNet. Therefore, these two parameter values ​​are the size of the convolution kernel and the recombination kernel of the CARAFE operator in the proposed model.

[0078] In addition, an Involution module is incorporated into the last layer of the two detection heads. This module enhances the model's adaptability to target rotation, scale changes, etc. by dynamically generating convolution kernels, so as to achieve the best improvement effect and improve the robustness of detection.

[0079] In practical applications, Involution employs a cross-channel weight sharing mechanism, effectively reducing the parameter redundancy of traditional convolutional kernels. Its core operation involves dynamically generating a spatially adaptive kernel and performing feature transformations, specifically comprising two key computational stages: First, a kernel generation function φ is defined, which dynamically predicts a dedicated kernel I for each spatial location (i, j) based on the input feature map. i，j Its mathematical expression is shown in the following formula:

[0080] I i，j =(X i，j ) = L1(L0X i，j );

[0081] Among them, X i，j σ represents the pixel feature at position (i, j) of the feature map; L0 and L1 are two learnable linear transformation matrices; σ represents the combination operation of the batch normalization (BN) layer and the nonlinear activation function.

[0082] Secondly, the generated Involution kernel is used to perform multiplication and addition operations with the input features to obtain the output feature map Y. i，j，u The calculation process is shown in the following formula:

[0083]

[0084] Where G represents the number of feature groups sharing the same Involution kernel across the channel dimension; K is the kernel size; C is the number of input channels; and (u, v) is the relative coordinate offset within the kernel receptive field ΔK.

[0085] Involution can perform different transformations on the feature information of each small block, thereby capturing the local features of the image more flexibly. While significantly reducing the number of model parameters and computational cost, it can still maintain excellent feature representation ability, providing a solution for lightweight model design.

[0086] In this embodiment of the application, to evaluate the performance of the optimized detector, a comparative experiment is conducted using the proposed backbone network C3GNet and an improved baseline model of a mainstream lightweight backbone network, such as... Figure 7 The diagram shows the comparative experimental results of the backbone network. The results indicate that while some lightweight backbone networks can achieve a comprehensive reduction in computational cost, parameter count, and memory usage, they all lead to a decrease in accuracy. Moreover, except for the proposed C3GNet, other networks all incurred significant accuracy losses (many exceeding 4%). However, excessive accuracy loss due to over-lightweighting is not our goal in improving the model. The proposed C3GNet not only significantly reduces accuracy across the three lightweighting metrics compared to the baseline model (and outperforms other mainstream backbone networks), but also achieves an accuracy loss of only about 1% for each metric. Therefore, C3GNet is the best choice for improving the lightweight performance of a model.

[0087] In this embodiment of the application, ablation experiments were conducted on three improved modules based on the proposed new backbone network. For example... Figure 8 The diagram shows the ablation experimental results of the detection module integrated into different structures. Carafe(5,3) represents the result from... Figure 6 The selected Carafe module is shown. It can be seen that Carafe, while ensuring model lightweightness, also compensates for the accuracy loss caused by C3GNet. The introduction of Involutional further improves the model's lightweight performance, while also offering an accuracy performance advantage compared to the architecture that only includes the C3GNet module. The final optimized proposed model not only improves accuracy compared to the original model, but also reduces computational cost, parameter count, and memory usage by 5.3 GFLOPs, 3.22M, and 6.4MB, respectively. In summary, the experimental results indicate that C3GNet and Involutional mainly improve the model's lightweight performance, while Carafe (5, 3) compensates for the accuracy loss caused by the lightweight module. The optimized model architecture is a more lightweight detector model that balances accuracy metrics.

[0088] S13: Using the sample dataset of transmission line protection against external force damage as training data, the optimized YOLOv7-tiny detector is trained to obtain the transmission line protection against external force damage detection model.

[0089] In this step, a sample dataset of transmission line external force damage samples, constructed and selected, is used as the core training data. This data contains rich and representative information on external force damage scenarios for transmission lines, covering various typical hidden dangers such as construction machinery intrusion, illegal building construction, and tree obstruction threats. This data is input into a transmission line external force damage detection neural network, and advanced deep learning algorithms and optimization strategies are used to iteratively adjust the network parameters. During training, the neural network continuously learns the features and patterns in the data, gradually improving its ability to identify and detect various external force damage hazards. After multiple rounds of meticulous and efficient training, the neural network continuously optimizes its structure and performance, ultimately successfully obtaining a transmission line external force damage detection model with accurate detection capabilities, high generalization, and strong adaptability. This model can quickly and accurately identify and precisely locate potential external force damage factors threatening the safety of transmission lines in complex and ever-changing real-world scenarios, providing reliable protection for the stable operation of transmission lines.

[0090] By employing the above methods, an optimized YOLOv7-tiny single-stage detector is constructed, enabling it to accurately identify and locate various potential hazards in transmission lines, thus providing strong technical support for the safe and stable operation of transmission lines.

[0091] S20: Acquire surveillance image data and point cloud data of the transmission line.

[0092] In this step, a monitoring system consisting of LiDAR and surveillance cameras is deployed on the power pole. The LiDAR performs periodic 3D scans of the transmission line and its surrounding environment to obtain high-precision point cloud data. Simultaneously, the surveillance cameras automatically collect visual monitoring images of the transmission line (referred to as surveillance images) at preset time intervals (e.g., every 5 minutes), enabling timed and visual recording of the line's status.

[0093] Through the above methods, the fusion acquisition of three-dimensional point cloud data and two-dimensional images is achieved by combining periodic laser point cloud scanning with timed visual monitoring. This provides multimodal data support for the detection of external force damage to power transmission lines, effectively improving the accuracy and reliability of hazard identification.

[0094] S30: Input the monitored image into the transmission line external damage prevention detection model to identify and locate the hidden danger target, and output the two-dimensional pixel coordinate data of the hidden danger target in the image coordinate system.

[0095] In this step, the two-dimensional surveillance image is input into the transmission line external damage prevention detection model. The detection algorithm driven by deep learning is used to accurately identify and locate the hidden danger targets in the surveillance image, and finally outputs the accurate two-dimensional pixel coordinate data of the hidden danger targets in the image plane coordinate system.

[0096] S40: Based on surveillance images, two-dimensional pixel coordinate data, and point cloud data, determine the spatial distance between the power transmission line and the potential hazard.

[0097] In this step, based on the spatial pose transformation relationship between the surveillance images and laser point cloud data, the two-dimensional pixel coordinate data of the detected potential hazard target are mapped to a three-dimensional point cloud space to solve for the three-dimensional geographic coordinates and actual elevation of the potential hazard target. Based on this, a three-dimensional Euclidean distance calculation method is used to solve for the minimum spatial distance between the potential hazard target and the transmission line.

[0098] In one embodiment of this application, a specific spatial distance calculation scheme is provided. In S40, based on the surveillance image, two-dimensional pixel coordinate data, and point cloud data, the spatial distance between the transmission line and the potential hazard target is determined, specifically including the following steps S41-S44:

[0099] S41: Obtain the intrinsic parameter matrix and distortion coefficients of a monocular camera through camera calibration technology.

[0100] S42: Obtain the spatial pose relationship between the captured image and the laser point cloud through a pose estimation algorithm.

[0101] S43: Based on the intrinsic parameter matrix, distortion coefficients, and spatial pose relationship, establish the coordinate transformation relationship from two-dimensional pixel coordinates to three-dimensional spatial coordinates.

[0102] For steps S41-S43, the intrinsic parameter matrix M and distortion coefficients D of the monocular camera are obtained through camera calibration technology, denoted as I(M, D). The EPnP pose estimation algorithm is used to accurately calculate the spatial pose relationship E(R, T) between the captured image and the LiDAR point cloud, where R represents the rotation matrix and T is the translation vector. Subsequently, a mapping relationship from two-dimensional pixel coordinates to three-dimensional world coordinates is established based on the above parameters. Specifically, from the three-dimensional spatial coordinates W... m ([X W Y W Z W ] T ) to two-dimensional pixel coordinates P m ([u, v]) T The projection transformation of ) can be expressed as:

[0103]

[0104] Among them, P h and W h P m and W m homogeneous coordinate form; Z c This represents the depth value in the camera coordinate system.

[0105] Subsequently, by introducing ground height ZW As a priori constraint on LiDAR point clouds, the inverse coordinate transformation relationship from pixel coordinates to 3D coordinates can be derived:

[0106]

[0107] Where M1 = R -1 T, M2 = R -1 M -1 [uv 1] T The rotation matrix R and translation vector T are both obtained by optimization using the EPNP algorithm.

[0108] The above method effectively solves the problem of accurate coordinate transformation between visual images and LiDAR point clouds, providing a reliable data foundation for subsequent spatial distance calculations.

[0109] S44: Based on the two-dimensional coordinate data and coordinate transformation relationship of the potential hazard target, generate the spatial distance from the potential hazard target to the transmission line.

[0110] In this step, after coordinate transformation, the spatial distance from the potential hazard to the transmission line is accurately calculated by combining the two-dimensional coordinate information of the transmission line. This distance data will intuitively reflect the spatial relationship between the potential hazard and the transmission line, providing an important basis for decision-making in the safety assessment and hazard mitigation of the transmission line.

[0111] In one embodiment of this application, a specific spatial distance calculation scheme is provided. In S44, based on the two-dimensional coordinate data of the potential hazard target and the coordinate transformation relationship, the spatial distance from the potential hazard target to the transmission line is generated, specifically including the following steps S441-S442:

[0112] S441: Calculate the three-dimensional spatial coordinate data of the potential hazard target based on two-dimensional coordinate data and coordinate transformation relationships.

[0113] S442: Based on three-dimensional spatial coordinate data, the spatial distance from the potential hazard target to the transmission line is calculated using a three-dimensional Euclidean distance algorithm.

[0114] For steps S441-S442, by establishing coordinate transformation relationships, the two-dimensional pixel coordinates (u, v) of the detected hazard target are converted into three-dimensional spatial coordinates (X, Y, Z) in a three-dimensional spatial coordinate system. Specifically, based on the assumption that the bottom of the hazard target is in contact with the ground (Z=0), the overdetermined equations are solved to obtain the base point coordinates of the hazard target. Subsequently, combined with the height pixel ratio of the hazard target in the image, the actual treatment height value H is calculated. Finally, the transformation from two-dimensional image coordinates to a three-dimensional spatial coordinate system is completed through coordinate transformation relationships.

[0115] Subsequently, based on the obtained three-dimensional coordinate data, an improved three-dimensional Euclidean distance algorithm was used to calculate the minimum spatial distance between the potential hazard target and the transmission line:

[0116]

[0117] Among them, (X) W0 Y W0 Z W0 (X) represents the three-dimensional spatial coordinates of the potential hazard target. W1 Y W1 Z W1 () represents the three-dimensional point cloud coordinates of the transmission line.

[0118] By using the above methods, the precise spatial distance from the potential hazard to the transmission line can be obtained, providing a scientific and reliable basis for the safety assessment, hazard warning, and maintenance decision-making of the transmission line, thereby effectively ensuring the stable operation of the transmission line and reducing potential risks.

[0119] S50: Assess the risk of external damage to transmission lines based on spatial distance.

[0120] In this step, spatial distance encompasses the relative positional relationship between potential hazards (such as construction machinery, trees, and foreign objects) and power transmission lines in three-dimensional space. Hazardous targets closer to the power transmission lines have a relatively higher probability of causing external damage, while those farther away have a relatively lower risk. By assessing spatial distance, the risk of external damage to power transmission lines can be accurately determined, and targeted preventative measures can be developed based on the assessment results.

[0121] The spatial distance-based assessment method described above can effectively improve the ability to prevent and control the risk of external damage to transmission lines, providing a solid guarantee for the stable operation of the power system.

[0122] In one embodiment of this application, a specific external force damage risk assessment scheme is provided. In S50, that is, based on spatial distance, the external force damage risk of transmission lines is assessed, specifically including the following steps S51-S53:

[0123] S51: Compare the spatial distance with a preset safe distance threshold;

[0124] S52: If the spatial distance is greater than or equal to the preset safe distance threshold, it is determined that the transmission line is at risk of external damage.

[0125] S53: If the spatial distance is less than the preset safe distance threshold, it is determined that there is no risk of external damage to the transmission line.

[0126] For steps S51-S53, a scientifically reasonable safety distance threshold is pre-set based on various factors such as the voltage level of the transmission line, the installation environment, and operational requirements. This threshold is a crucial critical value for ensuring the safe operation of the transmission line, comprehensively considering various factors that may lead to external damage and safety margins. After obtaining the spatial distance, it is compared with the preset safety distance threshold. If the spatial distance is greater than or equal to the preset safety distance threshold, it indicates that the distance between the potential hazard and the transmission line is within a relatively close danger range, and there is a possibility that the potential hazard may touch or affect the transmission line due to various unexpected situations (such as mechanical operation errors, fallen trees, drifting foreign objects, etc.). In this case, it is determined that there is a risk of external damage to the transmission line. In response to this situation, it is necessary to quickly activate the corresponding early warning mechanism and preventive measures, such as notifying the relevant responsible parties to take corrective actions and strengthening the monitoring of the potential hazard.

[0127] Conversely, if the spatial distance is less than the preset safe distance threshold, it indicates that the potential hazard maintains a sufficient safe distance from the transmission line. Under the current conditions, the possibility of the potential hazard causing external damage to the transmission line is extremely low, therefore it is determined that there is no risk of external damage to the transmission line. However, even in a safe state, vigilance cannot be relaxed, and the location and status of the potential hazard must still be continuously and dynamically monitored to respond to any possible changes.

[0128] In practical applications, the calculated three-dimensional distance helps inspection departments scientifically assess the risk level of potential hazards. Based on preset safety distance thresholds and risk assessment models, potential hazards are categorized into different risk levels, such as low, medium, and high risk. For hazards of different risk levels, inspection departments can take corresponding measures, such as strengthening monitoring, timely rectification, or emergency response.

[0129] The above methods can help identify the risk of external damage to power transmission lines in a timely and accurate manner, providing a strong guarantee for the safe and stable operation of power transmission lines.

[0130] As can be seen, the above scheme first achieves real-time detection and localization of potential hazards using deep learning methods. Then, it fuses high-precision laser point cloud data with two-dimensional images to obtain the three-dimensional spatial distance from the potential hazard to the power line. Finally, based on this three-dimensional spatial distance, it determines whether there is a risk of external force damage. This achieves automatic monitoring and early warning of transmission lines, greatly improving the response speed and processing efficiency for preventing external damage to transmission lines, and providing a strong guarantee for the safe and stable operation of transmission lines.

[0131] In one embodiment, a transmission line external damage detection device is provided, which corresponds one-to-one with the transmission line external damage detection method described in the above embodiments. For example... Figure 9As shown, the transmission line external damage detection device 100 includes: a construction module 101, an acquisition module 102, a generation module 103, a determination module 104, and a risk assessment module 105. Detailed descriptions of each functional module are as follows:

[0132] Module 101 is used to build a detection model for external damage to transmission lines. The detection model for external damage to transmission lines is used to identify and locate potential hazards corresponding to transmission lines.

[0133] The acquisition module 102 is used to acquire surveillance images and point cloud data of the transmission line;

[0134] The generation module 103 is used to input the surveillance image into the transmission line external damage prevention detection model to identify and locate the hidden danger target, and output the two-dimensional pixel coordinate data of the hidden danger target in the image plane coordinate system;

[0135] The determination module 104 is used to determine the spatial distance between the power transmission line and the potential hazard target based on the surveillance image, two-dimensional pixel coordinate data and point cloud data.

[0136] Risk assessment module 105 is used to assess the risk of external force damage to transmission lines based on spatial distance.

[0137] In one embodiment, the construction module 101 is specifically used for:

[0138] Construct a sample dataset of power transmission line protection against external damage;

[0139] Build an optimized YOLOv7-tiny detector;

[0140] Using the sample dataset of transmission line protection against external force damage as training data, the optimized YOLOv7-tiny detector was trained to obtain the transmission line protection against external force damage detection model.

[0141] In one embodiment, the construction module 101 is further configured to:

[0142] Acquire multiple hazard images corresponding to power transmission lines in the target area;

[0143] Each hazard image is preprocessed to remove invalid images, and each preprocessed hazard image is labeled. The label includes the hazard type and location information of the hazard target in the hazard image.

[0144] Expand each labeled hazard image;

[0145] Based on the expanded images of multiple potential hazards, a sample dataset for preventing external damage to power transmission lines is constructed.

[0146] In one embodiment, the construction module 101 is further configured to:

[0147] The original architecture for building the YOLOv7-tiny detector;

[0148] The YOLOv7-tiny detector is optimized by integrating the LLIE module, C3GNet module, upsampling operator CARAFE, and Involution module, resulting in an optimized YOLOv7-tiny detector.

[0149] In one embodiment, the determining module 104 is specifically used for:

[0150] The intrinsic parameter matrix and distortion coefficients of a monocular camera are obtained through camera calibration technology.

[0151] The spatial pose relationship between the captured image and the laser point cloud is obtained through a pose estimation algorithm.

[0152] Based on the intrinsic parameter matrix, distortion coefficients, and spatial pose relationship, a coordinate transformation relationship from two-dimensional pixel coordinates to three-dimensional spatial coordinates is established.

[0153] Based on the two-dimensional coordinate data and coordinate transformation relationship of the potential hazard target, the spatial distance from the potential hazard target to the transmission line is generated.

[0154] In one embodiment, the determining module 104 is further configured to:

[0155] Based on two-dimensional coordinate data and coordinate transformation relationships, calculate the three-dimensional spatial coordinate data of the potential hazard target;

[0156] Based on three-dimensional spatial coordinate data, the spatial distance from the potential hazard target to the transmission line is calculated using a three-dimensional Euclidean distance algorithm.

[0157] In one embodiment, the risk assessment module 105 is specifically used for:

[0158] Compare the spatial distance with a preset safe distance threshold;

[0159] If the spatial distance is greater than or equal to the preset safe distance threshold, it is determined that the transmission line is at risk of external damage.

[0160] If the spatial distance is less than the preset safe distance threshold, it is determined that there is no risk of external damage to the transmission line.

[0161] This invention provides a device for detecting external damage to power transmission lines. First, it uses deep learning to achieve real-time detection and localization of potential hazards. Then, it fuses high-precision laser point cloud data with two-dimensional images to obtain the three-dimensional spatial distance from the hazard to the power line. Finally, based on this three-dimensional spatial distance, it determines whether there is a risk of external force damage. This device enables automatic monitoring and early warning of power transmission lines, significantly improving the response speed and processing efficiency for preventing external damage, and providing a strong guarantee for the safe and stable operation of power transmission lines.

[0162] Specific limitations regarding the external damage prevention detection device for transmission lines can be found in the limitations of the external damage prevention detection method for transmission lines mentioned above, and will not be repeated here. Each module in the aforementioned external damage prevention detection device for transmission lines can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device in hardware form, or stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0163] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0164] A detection model for external damage prevention of transmission lines is constructed. This model is used to identify and locate potential hazards associated with transmission lines.

[0165] Acquire surveillance images and point cloud data of power transmission lines;

[0166] The monitoring images are input into the transmission line external damage prevention and detection model to identify and locate potential hazards, and the two-dimensional pixel coordinate data of the potential hazards in the image plane coordinate system are output.

[0167] Based on surveillance images, two-dimensional pixel coordinate data, and point cloud data, the spatial distance between the power transmission line and the potential hazard is determined.

[0168] Based on spatial distance, assess the risk of external damage to transmission lines.

[0169] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0170] A detection model for external damage prevention of transmission lines is constructed. This model is used to identify and locate potential hazards associated with transmission lines.

[0171] Acquire surveillance images and point cloud data of power transmission lines;

[0172] The monitoring images are input into the transmission line external damage prevention and detection model to identify and locate potential hazards, and the two-dimensional pixel coordinate data of the potential hazards in the image plane coordinate system are output.

[0173] Based on surveillance images, two-dimensional pixel coordinate data, and point cloud data, the spatial distance between the power transmission line and the potential hazard is determined.

[0174] Based on spatial distance, assess the risk of external damage to transmission lines.

[0175] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or electronic device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0176] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0177] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0178] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for detecting external damage to transmission lines, characterized in that, include: An external damage prevention detection model for transmission lines is constructed, which is used to identify and locate potential hazards corresponding to transmission lines. Acquire surveillance images and point cloud data of power transmission lines; The monitored image is input into the transmission line external damage prevention detection model to identify and locate potential hazards, and the two-dimensional pixel coordinate data of the potential hazard in the image plane coordinate system is output. Based on the monitored image, the two-dimensional pixel coordinate data, and the point cloud data, the spatial distance between the power transmission line and the potential hazard target is determined. Based on the aforementioned spatial distance, the risk of external force damage to the transmission line is assessed.

2. The method according to claim 1, characterized in that, The steps for constructing the external damage prevention detection model for transmission lines specifically include: Construct a sample dataset of power transmission line protection against external damage; Build an optimized YOLOv7-tiny detector; The sample dataset of external force damage to the transmission line is used as training data to train the optimized YOLOv7-tiny detector, thereby obtaining the external force damage detection model for the transmission line.

3. The method according to claim 2, characterized in that, The steps for constructing the sample dataset of power transmission line protection against external damage specifically include: Acquire multiple hazard images corresponding to power transmission lines in the target area; Each hazard image is preprocessed to remove invalid images, and each preprocessed hazard image is labeled. The label includes the hazard type and location information of the hazard target in the hazard image. Expand each labeled hazard image; Based on the expanded images of multiple potential hazards, a sample dataset for preventing external damage to the transmission line is constructed.

4. The method according to claim 2, characterized in that, The steps for constructing the optimized YOLOv7-tiny detector specifically include: The original architecture for building the YOLOv7-tiny detector; The YOLOv7-tiny detector is optimized by integrating the LLIE module, C3GNet module, upsampling operator CARAFE, and Involution module to generate the optimized YOLOv7-tiny detector.

5. The method according to claim 1, characterized in that, The step of determining the spatial distance between the transmission line and the potential hazard target based on the monitored image, the two-dimensional pixel coordinate data, and the point cloud data specifically includes: The intrinsic parameter matrix and distortion coefficients of a monocular camera are obtained through camera calibration technology. The spatial pose relationship between the captured image and the laser point cloud is obtained through a pose estimation algorithm. Based on the intrinsic parameter matrix, the distortion coefficients, and the spatial pose relationship, a coordinate transformation relationship from two-dimensional pixel coordinates to three-dimensional spatial coordinates is established; Based on the two-dimensional coordinate data of the potential hazard target and the coordinate transformation relationship, the spatial distance from the potential hazard target to the power transmission line is generated.

6. The method according to claim 5, characterized in that, The step of generating the spatial distance from the potential hazard to the transmission line based on the two-dimensional coordinate data of the potential hazard and the coordinate transformation relationship specifically includes: Based on the two-dimensional coordinate data and the coordinate transformation relationship, the three-dimensional spatial coordinate data of the potential hazard target are calculated; Based on the three-dimensional spatial coordinate data, the spatial distance from the potential hazard target to the transmission line is calculated using a three-dimensional Euclidean distance algorithm.

7. The method according to claim 1, characterized in that, The step of assessing the risk of external force damage to the transmission line based on the spatial distance specifically includes: The spatial distance is compared with a preset safe distance threshold. If the spatial distance is greater than or equal to the preset safety distance threshold, it is determined that the power transmission line is at risk of external damage. If the spatial distance is less than the preset safe distance threshold, it is determined that the transmission line is not at risk of external damage.

8. A device for detecting external damage to power transmission lines, characterized in that, include: The module is used to build a detection model for external damage to transmission lines, which is used to identify and locate potential hazards on transmission lines. The acquisition module is used to acquire surveillance images and point cloud data of transmission lines; The generation module is used to input the monitored image into the transmission line external damage prevention detection model to identify and locate the hidden danger target, and output the two-dimensional pixel coordinate data of the hidden danger target in the image plane coordinate system; The determination module is used to determine the spatial distance between the power transmission line and the potential hazard target based on the monitored image, the two-dimensional pixel coordinate data, and the point cloud data. The risk assessment module is used to assess the risk of external force damage to the transmission line based on the spatial distance.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the transmission line external damage detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the transmission line external damage detection method as described in any one of claims 1 to 7.

Images