A flood disaster under power transmission line fault detection method, device, equipment and medium

Abstract

The application discloses a flood disaster under power transmission line fault detection method, device, equipment and medium, relates to the technical field of fault detection. The method comprises the following steps: inputting a target power transmission line image in a flood scene into a flood interference multi-modal quantization model to obtain a flood interference feature of the target power transmission line image; the flood interference multi-modal quantization model is obtained by training a deep feature extraction network based on a positive and negative sample screening mechanism; visible faults in the target power transmission line image are identified based on the flood interference feature and a classification dynamic threshold; based on target point cloud data, a non-contact spatial geometric measurement method is used to detect hidden faults in the target power transmission line image; and the fault position is calibrated based on the flood interference feature, the target point cloud data, a visible fault identification result and a hidden fault identification result to obtain a fault multi-dimensional position calibration result. The application can quickly and accurately realize power transmission line fault detection.

Description

Technical Field

[0001] This application relates to the field of fault detection technology, and in particular to a method, device, equipment and medium for detecting faults in transmission lines under flood disasters. Background Technology

[0002] Against the backdrop of frequent floods, power transmission lines, as the core carriers of power transmission, are vulnerable to complex environmental factors such as flood erosion, floating debris snagging the lines, and insufficient distance between trees and lines, leading to faults such as tower collapses and line breaks, seriously threatening power grid safety and people's livelihoods. Currently, power transmission line inspection and fault detection mainly rely on manual labor, conventional drone inspections, and existing intelligent detection algorithms. These methods all have significant shortcomings in flood disaster scenarios and cannot achieve rapid and accurate fault detection of power transmission lines. Summary of the Invention

[0003] The purpose of this application is to provide a method, device, equipment and medium for detecting transmission line faults under flood disasters, which can quickly and accurately detect transmission line faults.

[0004] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for detecting transmission line faults during floods, including: Acquire images and point cloud data of target power transmission lines in flood scenarios; The target transmission line image is input into the flood interference multimodal quantization model to obtain the flood interference features of the target transmission line image; the flood interference features include: texture fluctuation intensity, occlusion index and illumination statistics; the flood interference multimodal quantization model is obtained by training a deep feature extraction network based on a positive and negative sample selection mechanism; Based on the flood interference features and classification dynamic threshold of the target transmission line image, visible faults in the target transmission line image are identified to obtain visible fault identification results; Based on the target point cloud data, a non-contact spatial geometric measurement method is used to detect hidden faults in the target power transmission line image, and the hidden fault detection results are obtained. The hidden fault identification results include: whether there are hidden dangers in the tree line safety distance, whether there are hidden dangers in the conductor sag, whether there are hidden dangers in the ground safety distance, and whether there are hidden dangers in the water safety distance. Based on the flood interference features of the target transmission line image, the target point cloud data, the visible fault identification results, and the hidden fault identification results, the fault location is calibrated to obtain a multi-dimensional fault location calibration result.

[0005] In one embodiment, after obtaining the multi-dimensional fault location calibration results, the method further includes: Based on the visible fault identification results, the hidden fault identification results, the multi-dimensional fault location calibration results, and the established fault-handling measures knowledge base, structured operation and maintenance instructions are generated to handle the faults in the target transmission line image.

[0006] In one embodiment, before generating structured operation and maintenance instructions, the method further includes: Calculate the confidence score based on the visible fault identification results and the flood interference features of the target transmission line image; Based on the confidence score and the set score value, it is determined whether to trigger the re-flight mechanism. If so, the target power line image and target point cloud data in the flood scene are reacquired.

[0007] In one embodiment, the method for determining the multimodal quantization model of flood disturbance includes: Acquire historical fault images of power transmission lines in flood scenarios; The historical fault images of the power transmission line are identified based on visual markers, and the fault rectangular area is selected. The fault rectangle region is cropped and its features are extracted to obtain a positive sample set; The region outside the fault rectangle region in the historical fault image of the power transmission line is identified as a negative sample candidate region. The negative sample candidate regions are scored and sorted based on the negative sample screening index to obtain a negative sample set. The negative sample screening index is determined based on the mean color feature vector of the positive samples in the positive sample set, the mean color feature vector of the negative sample candidate regions, the feature difference weight, the scene adaptation weight, and the color interference weight coefficient of the flood scene. The number of positive samples in the positive sample set and the number of negative samples in the negative sample set are balanced to obtain the training sample set; The training sample set and the corresponding texture fluctuation intensity are input into the water surface turbulence texture quantization model, the negative sample set and the corresponding occlusion index are input into the floating object occlusion quantization model, and the training sample set and the corresponding illumination statistics are input into the low-light water body reflection quantization model. Collaborative training is performed with the goal of minimizing the loss function value. The trained water surface turbulence texture quantization model, the trained floating object occlusion quantization model, and the trained low-light water body reflection quantization model are determined as the flood disturbance multimodal quantization model. The water surface turbulence texture quantization model, the floating object occlusion quantization model, and the low-light water body reflection quantization model are all deep feature extraction networks.

[0008] In one embodiment, visible faults in the target transmission line image are identified based on flood interference features and a classification dynamic threshold, resulting in visible fault identification results, specifically including: Based on the aforementioned flood disturbance characteristics, the disturbance weighted nonmaximum suppression algorithm is used to dynamically adjust the suppression threshold to obtain the classification dynamic threshold; The flood interference features of the target transmission line image are input into the fault classifier to obtain preliminary identification results of visible faults; The preliminary identification results of the visible faults are filtered based on the classification dynamic threshold to obtain the final visible fault identification results.

[0009] In one embodiment, based on the target point cloud data, a non-contact spatial geometric measurement method is used to detect latent faults in the target transmission line image to obtain latent fault detection results, specifically including: Based on the target point cloud data, a non-contact spatial geometric measurement method is used to extract the coordinates of the power transmission line pixels in the target power transmission line image to obtain the power transmission line pixel coordinates. The weighted least squares method is used to perform curve fitting on the pixel coordinates of the transmission line to obtain the conductor parabola. Calculate the theoretical height of the power transmission line at the projection point of the vegetation point on the power transmission line based on the parabola of the conductor. Based on the theoretical height of the power transmission line, the Euclidean distance from the vegetation point to the corresponding projection point in the cross section is calculated. The minimum Euclidean distance from all vegetation points to their corresponding projection points is determined as the treeline safety distance, and the treeline safety distance and the set treeline safety threshold are used to determine whether there are any potential hazards. Based on the target point cloud data, the conductor sag is calculated according to the height difference of the transmission line suspension points, span, transmission line parameters and operating conditions, and the presence of potential hazards in the conductor sag is determined based on the conductor sag and the set sag safety threshold range. The ground safety distance and water safety distance are calculated based on the parabola of the conductor, and the presence of potential hazards in the ground safety distance and water safety distance are determined based on the ground safety distance, the water safety distance, the set ground safety threshold, and the set water safety distance threshold.

[0010] In one embodiment, the fault location is calibrated based on the flood interference features of the target transmission line image, the target point cloud data, the visible fault identification result, and the latent fault identification result to obtain a multi-dimensional fault location calibration result, specifically including: The GPS positioning information of the drone used to collect images of the target power transmission line was obtained. Establish the mapping relationship between the pixel coordinates of the target transmission line image and the latitude and longitude coordinates of the GPS positioning information, and obtain the coordinate mapping formula; The initial coordinates of the fault candidate boxes corresponding to the visible fault identification results and the latent fault identification results are determined, and the initial coordinates of the fault candidate boxes are corrected using the flood interference features to obtain the corrected coordinates of the fault candidate boxes. Substitute the center point coordinates in the fault candidate box correction coordinates into the coordinate mapping formula to obtain the latitude and longitude coordinates of the fault center point. Based on the latitude and longitude coordinates of the fault center point, the target point cloud data, and the transmission line ledger information, the location calibration result of the visible fault is determined; The location calibration result of the latent fault is determined based on the latitude and longitude coordinates of the fault center point; Based on the location calibration results of visible faults and latent faults, the multi-dimensional location calibration results of faults are determined.

[0011] Secondly, this application provides a fault detection device for transmission lines under flood disasters, comprising: The data acquisition module is used to acquire images of target power transmission lines and target point cloud data in flood scenarios; The quantization model construction module is used to input the target transmission line image into the flood interference multimodal quantization model to obtain the flood interference features of the target transmission line image; the flood interference features include: texture fluctuation intensity, occlusion index and illumination statistics; the flood interference multimodal quantization model is obtained by training a deep feature extraction network based on a positive and negative sample selection mechanism; The visible fault identification module is used to identify visible faults in the target transmission line image based on the flood interference features and classification dynamic threshold, and obtain visible fault identification results. The latent fault identification module is used to detect latent faults in the target power transmission line image based on the target point cloud data using a non-contact spatial geometric measurement method, and obtain latent fault detection results; the latent fault identification results include: whether there are hidden dangers in the tree line safety distance, whether there are hidden dangers in the conductor sag, whether there are hidden dangers in the ground safety distance, and whether there are hidden dangers in the water safety distance; The fault location calibration module is used to calibrate the fault location based on the flood interference features of the target transmission line image, the target point cloud data, the visible fault identification results, and the hidden fault identification results, so as to obtain the multi-dimensional fault location calibration results.

[0012] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the transmission line fault detection method under flood disaster as described above.

[0013] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the transmission line fault detection method under flood disasters described above.

[0014] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, device, equipment, and medium for detecting transmission line faults under flood disasters. A multimodal quantization model of flood interference is obtained by training a deep feature extraction network based on a positive and negative sample screening mechanism. The training dataset more closely reflects the actual interference characteristics of flood disaster scenarios. Simultaneously, through the multimodal quantization modeling mechanism of flood interference, quantification and adversarial training of water surface turbulence texture, floating object occlusion, and low-light reflection are achieved, improving the model's robustness in fault identification under complex flood interference and effectively reducing missed and false detections. A non-contact spatial geometric measurement method is used to detect latent faults in the target transmission line image. This system detects hidden faults, including those related to tree-line safety distances, conductor sag, ground-to-ground safety distances, and water-to-ground safety distances. It overcomes the limitations of traditional visual recognition, which can only detect visible faults, achieving integrated detection of both visible and hidden faults, thus improving the comprehensiveness of transmission line fault detection. Based on flood interference characteristics, target point cloud data, and the results of visible and hidden fault identification, the system calibrates fault locations, achieving multi-dimensional fault location calibration and improving the response efficiency of emergency inspections of transmission lines in flood-prone scenarios. This application provides rapid and accurate fault detection for transmission lines. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments 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.

[0016] Figure 1 A flowchart illustrating a method for detecting transmission line faults during floods, provided in an embodiment of this application; Figure 2 This is a schematic diagram of transmission line sag measurement provided in an embodiment of this application; Figure 3 This is a schematic diagram of the sag fitting of a transmission line provided in an embodiment of this application; Figure 4 This is a flowchart of transmission line fault detection and location calibration provided in an embodiment of this application; Figure 5This is a schematic diagram of the results of power transmission line fault and hidden danger identification provided in an embodiment of this application; Figure 6 This application provides a schematic diagram of the functional modules of a power transmission line fault detection device under flood disasters. Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] Currently, manual inspections are hampered by road closures and hazardous environments during floods, making them unsuitable for rapid emergency response. While conventional drone inspections can replace personnel, they suffer from high false positive and false negative rates due to interference from water reflections and silt backgrounds. Existing intelligent algorithms lack anti-interference modeling for flood scenarios and are mostly limited to detecting visible faults such as broken wires and damaged insulators, failing to provide spatial geometric quantification for hidden hazards like abnormal sag and insufficient tree-line spacing. Furthermore, related technologies generally lack precise positioning capabilities, making it difficult to output accurate location information such as tower and span information. A mechanism linking fault detection with maintenance and handling has not been established, resulting in a disconnect between detection and response, low emergency response efficiency, and high decision-making delays.

[0020] It is evident that current methods cannot meet the needs of power grids for rapid inspection, accurate identification, and efficient handling during floods. Therefore, this application develops a method, device, equipment, and medium for detecting transmission line faults during floods. Based on anti-interference modeling and spatial perception, it achieves fault image and hidden danger detection, accurate location calibration, and maintenance instruction generation for transmission lines. It is applicable to emergency inspection and accurate handling of transmission lines in various flood disasters (basin-wide floods, localized waterlogging, river flooding, etc.). This application has significant practical importance and is urgently needed.

[0021] In one exemplary embodiment, such as Figure 1 As shown, a method for detecting transmission line faults under flood disasters is provided, including: Step 101: Obtain the target power transmission line image and target point cloud data in the flood scene.

[0022] Step 102: Input the target transmission line image into the flood interference multimodal quantization model to obtain the flood interference features of the target transmission line image.

[0023] The flood disturbance features include: texture fluctuation intensity, occlusion index, and illumination statistics; the flood disturbance multimodal quantification model is obtained by training a deep feature extraction network based on a positive and negative sample selection mechanism.

[0024] Step 103: Based on the flood interference features and classification dynamic threshold of the target transmission line image, the visible faults in the target transmission line image are identified to obtain the visible fault identification results.

[0025] Step 104: Based on the target point cloud data, a non-contact spatial geometric measurement method is used to detect hidden faults in the target power transmission line image to obtain hidden fault detection results.

[0026] The results of the hidden fault identification include: whether there are hidden dangers in the tree line safety distance, whether there are hidden dangers in the conductor sag, whether there are hidden dangers in the ground safety distance, and whether there are hidden dangers in the water safety distance; Step 105: Based on the flood interference features of the target transmission line image, the target point cloud data, the visible fault identification results, and the hidden fault identification results, the fault location is calibrated to obtain the multi-dimensional fault location calibration results.

[0027] In another exemplary embodiment of this application, the method for determining the multimodal quantization model of flood disturbance includes: (1) Use drones to obtain historical fault images of power transmission lines in flood scenarios.

[0028] (2) Identify the historical fault images of the power transmission line based on visual signs and select the fault rectangular area; crop the fault rectangular area and extract features from the cropped fault rectangular area to obtain a positive sample set.

[0029] The positive sample fault rectangle region includes at least one of the following: arc flashover, line breakage, insulator damage, conductor strand breakage, tower tilting, loose or missing hardware, and suspended objects. These faults are manually marked by professionals with experience in diagnosing transmission line faults based on visual indicators such as fault morphology, color anomalies, and structural defects. The coordinates, RGB three-channel color information, color mean, morphological contour parameters, and texture co-occurrence matrix features of the marked regions are extracted as sample features to obtain the positive sample set.

[0030] (3) The region in the historical fault image of the transmission line other than the fault rectangle region is determined as the negative sample candidate region.

[0031] (4) The negative sample candidate regions are scored and sorted based on the negative sample screening index to obtain a negative sample set. The negative sample screening index is determined based on the mean vector of color features of positive samples in the positive sample set, the mean vector of color features of negative sample candidate regions, feature difference weight, scene adaptation weight, and color interference weight coefficient of flood scene. The calculation formula of the negative sample screening index is as follows.

[0032] .

[0033] in, This is the mean vector of color features (RGB features) of the negative sample candidate region. This is the mean vector of color features (RGB features) of positive samples of transmission line faults. For feature difference weights, Adapt weights to different scenarios. Color interference weighting coefficient for flood scenes. Weighted by blue-gray feature of water body Weight of yellowish-brown characteristics of mud and sand The weighted fusion yields, i.e. ,in This is the scene proportion coefficient, with a value range of [0,1]. It is dynamically adjusted based on the pixel proportion of water and sediment in the real-time images collected by the drone: when the water proportion is higher than the sediment proportion, When the proportion of sediment is higher than that of water, When the proportions of the two are close, .

[0034] (5) Balance the number of positive samples in the positive sample set and the number of negative samples in the negative sample set to obtain the training sample set.

[0035] (6) Input the training sample set and the corresponding texture fluctuation intensity into the water surface turbulence texture quantization model, input the negative sample set and the corresponding occlusion index into the floating object occlusion quantization model, input the training sample set and the corresponding illumination statistics into the low-light water body reflection quantization model, and perform collaborative training with the goal of minimizing the loss function value. The trained water surface turbulence texture quantization model, the trained floating object occlusion quantization model and the trained low-light water body reflection quantization model are determined as the flood disturbance multimodal quantization model. The water surface turbulence texture quantization model, the floating object occlusion quantization model and the low-light water body reflection quantization model are all deep feature extraction networks.

[0036] The deep feature extraction network can adopt the YOLO series network. In order to further adapt to the flood scene and improve the robustness of the network to flood interference during inference, the water surface turbulence texture quantization model, floating object occlusion quantization model, and low-light water reflection quantization model are combined through a comprehensive total loss function that includes multiple interference quantization constraints. Joint optimization was performed to obtain a multimodal quantification model of flood disturbance.

[0037] Before inputting the training sample set into the model for training, this step involves preprocessing the training sample set. Preprocessing includes image cropping, standardization, and normalization. The standardization formula is as follows: ,in, To standardize the sample pixel values, These are the standardized sample pixel values. The average value of the sample pixels. The standard deviation of the sample pixels. , After processing, the pixel values ​​are normalized to [-1, 1], eliminating the influence of different lighting and imaging conditions on sample features and providing high-quality data for anti-interference modeling. Instead of the original image, in which i This represents the number of the label box. j Three metrics representing an image (length, width, and RGB values).

[0038] This step collaboratively calculates texture fluctuation intensity, occlusion index, and illumination statistics to achieve comprehensive quantification of flood-specific disturbances, providing core support for anti-interference modeling.

[0039] Specifically, based on the calculation of local Laplace response deviation, a water body segmentation mask M is introduced to calculate the texture fluctuation intensity only for the water area. The specific steps are as follows.

[0040] ① Calculate pixel-level local texture fluctuation deviation ,in Image pixel coordinates, For pixels The Laplace operator response at the given location (calculated using a 3×3 Laplace template). This represents the expected texture response value for a undisturbed, calm background image.

[0041] ② Calculate the global texture fluctuation intensity of the candidate region ,in N Let R be the total number of pixels in the candidate region. Water body segmentation mask, Represents pixels Belongs to water area This indicates non-water areas, ensuring that the calculation of wave intensity strictly focuses on areas flooded.

[0042] Specifically, for flood-specific floating debris (broken wood, straw, plastic products, and household waste), the occlusion of floating debris is quantified by fusing material perception with color-texture-contextual features. The specific steps are as follows.

[0043] ① Train a floating object occlusion quantization model using negative sample data to output pixel-level occlusion probabilities. ,in negative sample pixels Deep features (extracted via ResNet50 network). The weight vector to be learned (initial values ​​are random normal distributions with a mean of 0 and a standard deviation of 0.01). For bias terms, It is a sigmoid activation function with an output probability range of [0,1].

[0044] ② Calculate the regional occlusion index ,in The threshold for determining occlusion. , For indicator functions, hour, (Determine that the pixel is occluded), otherwise . This represents the proportion of pixels in region R that are determined to be occluded, for example... This indicates that 30% of the surface area of ​​the insulator string is disturbed by floating objects, providing a quantitative basis for interference suppression in anti-interference modeling.

[0045] Specifically, a lighting description vector is constructed using global brightness, contrast, and noise levels to quantify the reflectivity of water bodies under low light conditions. The specific steps are as follows.

[0046] ①Based on pixel brightness value The input image is divided into regions, and differential illumination equalization processing is performed.

[0047] ② Calculate global illumination statistics, including brightness. Contrast noise level ,in Image width, This represents the image height.

[0048] ③ Construct the lighting description vector As a luminance statistic, it represents the luminance level, and based on this vector, a luminance degradation quantification rule is defined: the trigger condition for severe low luminance is... Significant overexposure is triggered by the percentage of overexposed pixels. High noise interference triggering condition is .

[0049] During training, the loss function used is the mean squared error loss function, specifically: .

[0050] Among them, China , For texture fluctuation intensity, To obscure the index, This is a metric for illumination. For texture fluctuation loss, To cover up the damage, This is due to illumination loss. The number of training samples is... , The quantized output value of the sample. This represents the physical true value of the sample. This can be obtained by calculating the difference between the Laplacian response of the image patch and the historical normal image patch. It is a true illumination statistic calculated directly from the image. It is the ground truth value of the binary occlusion mask generated by the difference image.

[0051] The expression for the total loss function used during model training is as follows.

[0052] .

[0053] This is the main detection task loss of the original YOLO network, which includes CIoU localization loss, confidence BCE loss, and classification cross-entropy loss, ensuring basic detection accuracy. Interference quantization loss. This ensures that the texture fluctuation feature quantization function, occlusion feature quantization function, and low-light feature quantization function can output accurately. , This refers to flood resistance loss, which corrects the model's prediction errors under flood disturbances. In flood scenarios, this loss ensures that the model can still correctly outline and identify faults such as damaged umbrella skirts, arc burns, broken wire strands, and through-cracks in the internal ceramic when faced with water waves and reflections.

[0054] In another exemplary embodiment of this application, step 103 specifically includes: Based on the flood disturbance characteristics, the disturbance weighted nonmaximum suppression algorithm is used to dynamically adjust the suppression threshold to obtain the classification dynamic threshold.

[0055] The formula for calculating the classification dynamic threshold is as follows.

[0056] .

[0057] in, , , These represent the average texture fluctuation intensity, average occlusion ratio, and average brightness within the candidate bounding box, respectively. , and The IoU is the core metric for calculating the loss, used to determine non-maximum suppression (NMS), and its corresponding weight coefficients for the perturbations. In regions with drastic texture fluctuations or severe occlusion, a dynamic classification threshold is used. Automatically reduce the number of detection boxes of different shapes due to interference, thus avoiding excessive merging and removal of the same target and improving recall.

[0058] (2) Input the flood interference features of the target transmission line image into the fault classifier to obtain the preliminary identification results of visible faults.

[0059] (3) Based on the classification dynamic threshold, the preliminary identification results of the visible faults are filtered to obtain the final identification results of the visible faults.

[0060] The visible fault detection in this embodiment can identify visible faults such as arc flashover, wire breakage, and insulator damage based on the results of anti-interference modeling.

[0061] In another exemplary embodiment of this application, step 104 specifically includes: (1) Based on the target point cloud data, the coordinates of the power transmission line pixels in the target power transmission line image are extracted using a non-contact spatial geometric measurement method to obtain the power transmission line pixel coordinates.

[0062] (2) The weighted least squares method is used to fit the pixel coordinates of the transmission line to obtain the conductor parabola.

[0063] (3) Calculate the theoretical height of the transmission line at the projection point of the vegetation point on the transmission line based on the parabola of the conductor.

[0064] (4) Calculate the Euclidean distance from the vegetation point to the corresponding projection point in the cross section based on the theoretical height of the transmission line.

[0065] (5) The minimum Euclidean distance from all vegetation points to the corresponding projection points is determined as the tree line safety distance, and the tree line safety distance is determined to have potential hazards based on the tree line safety distance and the set tree line safety threshold.

[0066] (6) Based on the target point cloud data, calculate the conductor sag according to the height difference of the transmission line suspension point, the span, the transmission line parameters and the operating conditions, and determine whether there is a hidden danger in the conductor sag according to the conductor sag and the set sag safety threshold range.

[0067] (7) Calculate the safe distance to the ground and the safe distance to water based on the parabola of the conductor, and determine whether there are any hidden dangers in the safe distance to the ground and the safe distance to water based on the safe distance to the ground, the safe distance to water, the set safe threshold for the ground and the set safe distance threshold for the water.

[0068] In this embodiment, the fitting formula for the conductor parabola is: ,in The horizontal coordinates along the route are: Altitude , , The fitting coefficients are obtained using the weighted least squares method.

[0069] The tree line safety distance is calculated using the vertical projection distance of the cross section, and the specific steps are as follows.

[0070] ① Locate vegetation points Projection point along the direction of the power transmission line Calculate the theoretical height of the transmission line at the projection point. .

[0071] ② Within the cross-section, calculate the points. Time Euclidean distance .

[0072] ③ Based on the safety threshold set in DL / T 741-2019, when the tree line safety distance d If the distance is less than the set tree line safety threshold, it is determined that there is a potential risk of insufficient tree line safety distance.

[0073] The inversion formula for conductor sag is as follows: ,in The span (m) is determined by the horizontal distance between the suspension points of two adjacent towers extracted from the point cloud. For transmission line specific load (N / (m·mm)) 2 This represents a comprehensive reflection of loads such as self-weight and wind pressure. It is the elevation difference angle. , The height difference between the suspension points The maximum sag is the stress on the transmission line; when the calculated maximum sag exceeds the design threshold, it is judged as a potential hazard of abnormal conductor sag.

[0074] This embodiment of hidden hazard detection can identify hidden hazards such as insufficient tree-line distance, abnormal conductor sag, and insufficient safe distance to ground or water through non-contact spatial geometric measurement. In practical applications, it can achieve simultaneous and integrated detection of visible faults and hidden hazards.

[0075] In another exemplary embodiment of this application, step 105 specifically includes: (1) Obtain the GPS positioning information of the UAV that collects images of the target power transmission line.

[0076] (2) Establish the mapping relationship between the pixel coordinates of the target transmission line image and the latitude and longitude coordinates of the GPS positioning information, and obtain the coordinate mapping formula.

[0077] (3) Determine the initial coordinates of the fault candidate boxes corresponding to the visible fault identification results and the hidden fault identification results, and use the flood interference features to correct the initial coordinates of the fault candidate boxes to obtain the corrected coordinates of the fault candidate boxes.

[0078] The formula for calculating the corrected coordinates of the fault candidate box is: , ,in , These are the initial coordinates of the fault candidate boxes. This is the coordinate correction amount. The step size of the feature map is increased to improve the accuracy of the position calibration.

[0079] (4) Substitute the center point coordinates in the fault candidate box correction coordinates into the coordinate mapping formula to obtain the latitude and longitude coordinates of the fault center point.

[0080] (5) Based on the latitude and longitude coordinates of the fault center point, the target point cloud data and the transmission line ledger information, determine the location calibration result of the visible fault; based on the latitude and longitude coordinates of the fault center point, determine the location calibration result of the hidden fault; based on the location calibration result of the visible fault and the location calibration result of the hidden fault, determine the multi-dimensional location calibration result of the fault.

[0081] In this embodiment, accurate location calibration is achieved by mapping the candidate box coordinates with the UAV's GPS positioning information: the coordinates of the center point of the fault candidate box are converted into latitude and longitude coordinates, and the specific tower number, distance within the span, and conductor phase are marked to ensure that the fault location can be accurately traced.

[0082] In another exemplary embodiment of this application, such as Figure 1 As shown, after step 105, the following steps are also included: Step 106: Based on the visible fault identification results, the hidden fault identification results, the multi-dimensional fault location calibration results, and the established fault-handling measures knowledge base, generate structured operation and maintenance instructions to handle the faults in the target transmission line image.

[0083] Among them, the structured operation and maintenance instructions are automatically matched with the set fault-handling measures knowledge base based on fault type, hidden danger level and location information. The handling priority is divided into four levels: emergency, high, medium and low. The instructions embed latitude and longitude coordinates, specific fault descriptions and standardized operating steps. Operation and maintenance personnel can directly carry out handling work according to the instructions to realize the closed loop of detection-calibration-handling.

[0084] In another exemplary embodiment of this application, after step 105 and before step 106, the method further includes: calculating a confidence score based on the visible fault identification result and the flood interference characteristics of the target transmission line image; determining whether to trigger a re-flight mechanism based on the confidence score and a set score value; if so, re-acquiring the target transmission line image and target point cloud data in the flood scene.

[0085] In this embodiment, the formula for calculating the confidence score of the fault candidate box is: ,in The confidence level of the regional features (output by the model, ranging from 0 to 1). The weighting coefficients are all 0.25, and the confidence scores range from 0 to 1.

[0086] When the confidence score of the fault candidate box is less than 0.7, the system automatically prompts "partial occlusion, go-around recommended" and triggers the UAV go-around command to re-acquire images of the area for detection, ensuring the accuracy of integrated hazard detection.

[0087] In another exemplary embodiment of this application, when acquiring images of power transmission lines, a drone equipped with a visible light camera is used to acquire images of power transmission lines with a span of less than 500m. The resolution of the acquired images is not less than 1920×1080, and the flight altitude is controlled between 50-100m to ensure that the image clarity meets the requirements of anti-interference modeling, integrated detection, and accurate calibration.

[0088] The following section provides a more detailed introduction to the four core components of this embodiment: anti-interference modeling, integrated fault and hidden danger detection, precise location calibration, and operation and maintenance command generation.

[0089] Step 1: Constructing the anti-interference training dataset.

[0090] 1. Positive sample collection.

[0091] (1) Based on the historical images collected by UAV visible light cameras of the determined categories, a sample set for training the model is constructed. First, cropping and extraction of the three elements are performed.

[0092] Specifically, professionals with experience in diagnosing transmission line faults manually interpret the images. Based on visually significant features such as fault morphology (e.g., broken wire marks, cracks in insulators), abnormal colors (e.g., bright white arcs, stains on damaged insulators), and structural defects (e.g., missing fittings, loose conductor strands), they identify and select typical fault rectangular areas, and extract the coordinates of the center point, upper right corner, and upper left corner. Record the data. Then, using the marked box as a reference, crop out the image area containing only the faulty part (cropping size is uniformly 224×224 pixels) to avoid introducing interference from irrelevant scenes.

[0093] After cropping, the RGB three-channel color information is extracted from the cropped image region. The color feature mean (R mean, G mean, B mean), shape contour parameters (contour area, perimeter, roundness), and texture co-occurrence matrix features (energy, entropy, contrast) are used as the three elements for quantitative representation to form a feature representation that can be used for model input.

[0094] (2) Transform the typical fault image set of transmission lines into an equivalent data sample set to obtain the positive sample set. , as shown below.

[0095] .

[0096] in Represents an image of a known transmission line fault. and These represent the image's length and width, in pixels, respectively. It represents an image of the red, green, and blue channels that contains only the faulty part of the transmission line. Each pixel contains three values ​​(0-255).

[0097] 2. Negative sample collection and screening.

[0098] (1) Coarse screening of negative sample candidate regions.

[0099] In the original images of power transmission line fault inspections, after excluding the labeled positive sample areas, areas that may contain interfering targets (such as water bodies, silt areas, and areas with floating debris accumulation) are preliminarily screened as negative sample candidate areas. A 0-1 variable is introduced. , This indicates that the region does not contain positive sample extraction areas. This indicates the region containing positive samples; it specifies the region containing negative samples. Within, all points must satisfy This yields the candidate region for the negative sample set. .

[0100] (2) Negative sample screening.

[0101] Based on the anti-interference training sample screening index, candidate regions are scored and ranked to eliminate irrelevant interference and avoid excessive negative samples that could lead to bias in the anti-interference modeling. The formula for calculating the screening index is as follows.

[0102] .

[0103] in, This is the mean vector of color features of the negative sample candidate region. This is the mean vector of color features for positive samples of transmission line faults (e.g., the mean pixel value of the metallic luster area of ​​a broken conductor strand, or the mean pixel value of the dirt area of ​​a damaged insulator). Feature difference weights. The value is 0.6, representing the scene adaptation weight. The value is 0.4. The color interference weighting coefficient for flood scenes is... The blue-gray feature weight of water body The weight of the yellowish-brown characteristic of silt Scene proportion coefficient The algorithm dynamically adjusts based on the pixel ratio of water and sediment in real-time images captured by the drone: when the water ratio is higher than 50%. When the sediment content is higher than 50%, When the difference between the two percentages is less than 10%, Candidate negative samples are ranked based on this metric, and those with higher metric values ​​(i.e., those most easily confused with positive samples) are selected as the negative sample set. .

[0104] (3) Sample size control: total negative sample set Total number of positive and negative samples 10%, and must meet the following conditions: For example, if the number of positive samples is 900, then the number of negative samples is 100, for a total of 1000. This ensures a balance between positive and negative samples during training and guarantees the sensitivity of the anti-interference model to the identification of faulty targets.

[0105] 3. Sample preprocessing.

[0106] The cropped image is then standardized and normalized. The standardization formula will not be elaborated here.

[0107] Step 2: Quantitative modeling of multimodal flood disturbances and optimization to prevent missed detections.

[0108] 1. Quantification of flood-specific disturbances.

[0109] This step uses the anti-interference training dataset constructed in step one to perform quantitative modeling of three types of interference unique to flood scenarios: water surface turbulence texture, floating object occlusion, and low-light water reflection. The interference is transformed into calculable feature coefficients, providing a basis for subsequent interference suppression and constructing a complete multimodal quantitative modeling system for flood interference.

[0110] (1) Quantification of water surface turbulence texture.

[0111] To address the unique high-energy turbulent visual patterns on flooded water surfaces, such as large-scale breaking waves, turbulent stripes, and uniform turbulence, a flood texture fluctuation coefficient is constructed by comprehensively employing indicators such as local turbulence intensity estimation based on gradient second derivative and dominant turbulence scale extraction based on frequency domain analysis. The specific steps are as follows.

[0112] Step 1: Pixel-level local texture fluctuation deviation calculation. In flood scenes, the crests, troughs, and edges of broken waves on the water surface will produce a strong Laplacian response. Therefore, this step aims to build a perception module that can quantify such hydrodynamic textures. The calculation formula for pixel-level local texture fluctuation deviation is as follows.

[0113] .

[0114] in, Image pixel coordinates, For pixels The Laplace operator response at the given location (calculated using a 3×3 Laplace template). This is the expected value of the texture response of an image with a calm, undisturbed background, obtained by acquiring 100 images of a calm water surface and calculating the mean value of the Laplacian response. .

[0115] Step 2: Calculate the global texture fluctuation intensity of the candidate region. A water body segmentation mask is specifically introduced. The coefficient was obtained by training the U-Net model with a training dataset of 1000 images of water bodies in flood scenes, achieving an accuracy of 98.5%. The formula for calculating the global texture fluctuation intensity will not be elaborated here.

[0116] (2) Quantification of floating object obstruction.

[0117] For unique floating debris carried by floods, such as broken logs, straw, plastic products, and household waste, a material perception-based occlusion feature quantification function is trained to quantify the degree of occlusion and improve the anti-interference modeling system. The specific steps are as follows.

[0118] Step 1: Calculate pixel-level occlusion probability. Train the occlusion feature quantization function using negative sample data. The training objective is to optimize the function's output (pixel-level occlusion probability). It can accurately characterize the probability that a pixel belongs to an occlusion. The calculation formula for pixel-level occlusion probability will not be repeated here.

[0119] During training, the Adam optimizer is used, with 100 iterations until the loss function converges, ensuring the accuracy of occlusion quantization and providing support for anti-interference modeling.

[0120] Step 2: Calculate the occlusion area percentage at the regional level. After obtaining the occlusion probability map at the pixel level across the entire image, calculate the regional occlusion index for any candidate region R. The formula for calculating the shading index will not be repeated here.

[0121] (3) Quantification of water reflection under low light conditions.

[0122] To address the complex imaging environment during floods, characterized by turbid water, wave reflections, and low light, this study divides regions based on pixel brightness values, implements differentiated processing, constructs illumination description vectors, quantifies the degree of illumination degradation, and supplements the anti-interference modeling system. The specific steps are as follows.

[0123] Step 1: Image Region Segmentation. First, the input image is divided into low-light regions (… ), normal lighting area ( Overexposed areas ).

[0124] The standard formula for converting RGB to grayscale uses histogram equalization for low-light areas and exposure compensation for overexposed areas, outputting an image with equalized illumination for subsequent anti-interference modeling and detection.

[0125] Step 2: Global Illumination Statistics Calculation. The image quality of power line inspections can be affected by deceptive water source interference: the high intensity of specular and diffuse reflection from floodwater surfaces. Therefore, the following calculation measures the brightness, contrast, and noise level of the global image.

[0126] .

[0127] .

[0128] .

[0129] For brightness, For contrast, To measure noise levels, all three factors are considered together in relation to lighting conditions. In this context, the value represents the brightness of all pixels. The mean of the values. Then, to distinguish between overall darkness due to insufficient lighting and localized overexposure due to water reflection, three core optical statistics are extracted from the image to form a lighting condition description vector: .

[0130] This value comprehensively characterizes the illumination degradation of the image, providing a quantitative basis for illumination interference in anti-interference modeling.

[0131] Step 3: Light Degradation Quantification Rules. For severely low light conditions, define the following rules: Time-triggered; for cases of significant overexposure, when the percentage of overexposed pixels is greater than [missing information]. Time-triggered; for high noise interference, when The threshold is triggered at specific times (this threshold is derived from an 8-bit image after a 5×5 Gaussian filter, and the accuracy rate reaches 97.8% after verification with 1000 flood images under different lighting conditions), ensuring the reliability of lighting interference quantification.

[0132] 2. Optimize training for quantifying losses caused by flood disturbances.

[0133] The training objective is to construct corresponding loss functions to ensure that each quantized output approximates its physical true value. The specific training objectives and methods are as follows.

[0134] Minimize texture fluctuation loss This enables it to accurately predict the intensity of texture fluctuations in the region. W ( R Minimize occlusion loss This enables it to accurately predict the shading index. O ( R Minimize illumination loss This enables it to accurately predict light levels. L ( R loss function The expression will not be elaborated here.

[0135] 3. Learning the fault characteristics of flood interference countermeasures.

[0136] After completing the quantification of flood disturbances Building upon this foundation, the core task of this stage is to utilize these quantitative indicators to improve the model's robustness in complex hydrological environments through adversarial training.

[0137] (1) Non-maximum suppression with interference weighting. To reduce false rejections caused by flood interference, this embodiment uses interference weighting. This mechanism dynamically adjusts the merging threshold and appropriately relaxes the suppression conditions in areas with strong interference. The calculation formula for the classification dynamic threshold will not be repeated here.

[0138] (2) Overall training objective. The optimization objective of the model is determined by the multi-task flood adversarial loss function (total loss function). With joint guidance, the expression for the total loss function will not be elaborated here.

[0139] Step 3: Integrated hazard detection.

[0140] The core of this step is to overcome the two-dimensional limitations of visual inspection. Based on the anti-interference modeling results completed in step two, three-dimensional quantitative calculations of tree line safety distance and conductor sag are achieved through image recognition and geometric modeling. This enables accurate determination of hidden hazards and integrated detection of visible and hidden faults in transmission lines, thus overcoming the limitation of existing technologies that can only detect visible faults.

[0141] 1. Visible fault detection.

[0142] Based on the anti-interference modeling completed in step two, the preprocessed UAV-collected images are input into a trained deep feature extraction network to extract anti-interference visual features. The classifier identifies visible faults such as arc flashover, wire breakage, insulator damage, conductor strand breakage, tower tilting, loose or missing hardware, and suspended objects. Combined with the interference-weighted NMS algorithm, redundant detection boxes are removed, and prediction results with an IoU greater than 0.5 and a confidence level greater than 0.7 are retained to achieve accurate detection of visible faults, ensuring that the accuracy of visible fault detection is not less than 95% under flood interference conditions.

[0143] 2. Detection of hidden dangers.

[0144] (1) Pixel coordinate extraction. Based on the model trained in step two, fault identification is performed on the images collected by the UAV. At the same time, the coordinates of all pixels of the power transmission line are filtered out by combining the OpenCV algorithm (using the Canny edge detection algorithm with a threshold of 100-200). and vegetation point set .

[0145] (2) Fit all transmission line pixel coordinates into a curve .in The horizontal coordinates along the route are: Altitude The fitting coefficients; the weight matrix of the weighted least squares method. This is a diagonal matrix, where the diagonal elements represent the confidence scores of the transmission line pixels (output by the model, ranging from 0 to 1). Higher confidence scores correspond to higher weights and thus higher fitting accuracy. This fitting method achieves a fitting error of less than 5% when the sag span is small (span < 500m) and the sag variation is minimal. The accuracy meets engineering requirements, and the method is computationally efficient.

[0146] (3) Calculation of tree line safety distance and hazard assessment. This is simplified to calculating the vertical projection distance from the vegetation point to the power transmission line (within a cross section perpendicular to the line direction). The specific steps are as follows.

[0147] For vegetation points Its fitting transmission line spatial distance It needs to be calculated in three-dimensional space. It can be simplified to calculating the vertical projection distance from the point to the transmission line (in a cross-section perpendicular to the line's direction).

[0148] ① Locate vegetation points Projection point along the direction of the power transmission line Based on the parabolic fitting formula for conductors, the theoretical height of the transmission line at the projection point is calculated: .

[0149] ② Within the cross-section, calculate the points. Time The Euclidean distance is given by the formula: .in This represents the horizontal coordinates of the transmission line on the cross-section.

[0150] ③ Safety threshold setting: Based on "DL / T 741-2019" "Operation Regulations for Overhead Transmission Lines" and combined with the characteristics of flood disaster scenarios, safety thresholds for tree lines at different voltage levels are set, as shown in Table 1.

[0151] Table 1 Treeline Safety Thresholds

[0152] ④ Hazard assessment: Calculate the minimum distance from the vegetation point to the power transmission line using the following formula.

[0153] .

[0154] when When the corresponding voltage level's safety threshold is reached, it is determined to be a potential hazard due to insufficient tree-line spacing; when When the safety threshold is reached, it is determined that there is no hidden danger.

[0155] (4) Quantitative calculation of conductor sag and identification of potential hazards.

[0156] ① Sag Inversion Formula: Quantitative Calculation Model for Transmission Line Sag. The goal of this step is to accurately calculate the current sag under known conditions, including the height difference of the suspension points, span, transmission line parameters, and operating conditions. The calculation formula is as follows.

[0157] .

[0158] in, The span (m) is determined by extracting the horizontal distance between the suspension points of two adjacent towers from the point cloud data collected by the lidar on the UAV, with a collection accuracy error not exceeding ±0.1m; The specific load of a transmission line is a comprehensive reflection of loads such as its own weight and wind pressure (the icing coefficient is taken as 0.3 in flood disaster scenarios), for example, the JL / G1A-120 / 20 type transmission line. N / (m·mm 2 ); For the elevation difference angle, tan , h The height difference between the suspension points is calculated by extracting the altitude of the suspension points from the point cloud data. The stress on the transmission line is determined based on the material and operating conditions of the transmission line, and its range under normal operating conditions is 120-150 N / mm. 2 During floods, due to changes in environmental humidity, the stress correction factor is taken as 0.95. The value range is 114-142.5 N / mm. 2 .

[0159] ② Sag Calculation Steps: First, point cloud data of two adjacent towers is collected using UAV lidar. RANSAC algorithm is used to remove point cloud noise (noise removal threshold is 0.05m), and the three-dimensional coordinates of the suspension point at the top of the tower are extracted. and Secondly, calculate the distance between the gears and the height difference d.

[0160] .

[0161] Then, calculate the elevation difference angle. And determine the specific load according to the type of transmission line. and the corrected stress Substitute into the sag inversion formula to calculate the current sag. Finally, the maximum sag within the span is calculated. When there are terrain undulations within the span, the maximum sag value of each segment is extracted by fitting a parabola to the conductor segment, and the maximum value is taken as the maximum sag within the span. Transmission line sag measurement is as follows: Figure 2 As shown, the sag fitting of the transmission line is as follows: Figure 3 As shown.

[0162] ③ Judgment of abnormal sag hazards: Based on the "Design Code for Overhead Transmission Lines" (GB 50545-2010) and combined with the characteristics of flood disaster scenarios, sag safety thresholds for different voltage levels and different spans are used. When the calculated maximum sag within the span exceeds the corresponding safety threshold, it is judged as an abnormal sag hazard; when the maximum sag is within the safety threshold range, it is judged as no abnormal sag hazard.

[0163] ④ Supplementary determination of safe distances to ground and water: In flood disaster scenarios, it is necessary to additionally determine the safe distances of the conductors to ground and water to avoid conductors being submerged in water and leaking electricity or discharging to ground.

[0164] According to DL / T 741-2019, the minimum safe distance between low-voltage (220 / 380V) transmission lines and ground is 2.5m, and the minimum safe distance between them and water is 3.0m; the minimum safe distance between 10kV and below transmission lines and ground is 3.0m, and the minimum safe distance between them and water is 3.5m; the minimum safe distance between 35kV transmission lines and ground is 4.0m, and the minimum safe distance between them and water is 4.5m. The elevation of the conductor at various points within the span is calculated using the conductor parabolic fitting formula. Combined with ground / water surface elevation data collected by drones Calculate the safe distance: .

[0165] when When the corresponding safety threshold is reached, it is determined to be a potential hazard due to insufficient safe distance to the ground / water, and is included in the results of hidden hazard detection.

[0166] Step 4: Accurately pinpoint the fault location and generate maintenance instructions.

[0167] The core of this step is to integrate anti-interference modeling features with spatial geometric parameters to accurately calibrate the fault location, establish a correlation mechanism between the fault and the handling measures, automatically generate structured operation and maintenance instructions, and realize a closed-loop system of "detection-calibration-handling". This solves the defects of inaccurate positioning and disconnected handling in existing technologies and improves emergency response efficiency.

[0168] 1. Accurate location of the fault.

[0169] By combining the UAV's GPS positioning information, image pixel coordinates, and three-dimensional geometric parameters, multi-dimensional and accurate calibration of the fault location can be achieved, ensuring that maintenance personnel can quickly locate the fault site. The specific steps are as follows.

[0170] (1) Establishment of coordinate mapping relationship. During the flight of the UAV, its own GPS latitude and longitude coordinates are recorded in real time. Flight altitude Heading angle Pitch angle Combined with camera intrinsic parameters (focal length) Pixel size Establish image pixel coordinates ground latitude and longitude coordinates The mapping relationship is shown in the following formula.

[0171] .

[0172] .

[0173] in, The coordinates of the camera's image center pixel (determined by camera parameters; for a 1920×1080 resolution camera). , ); 111319.9 is the conversion factor between latitude and longitude and distance (unit: m / degree); The focal length of the camera is 12mm. The pixel size is set to 1.5μm to ensure that the coordinate mapping accuracy error does not exceed ±0.3m.

[0174] (2) Fault candidate box coordinate correction. Based on the flood interference features (texture fluctuation intensity) extracted in step two. Occlusion Index Illuminance The initial coordinates of the fault candidate boxes are corrected to reduce coordinate deviations caused by interference. The formula for calculating the corrected candidate box coordinates is as follows: , .

[0175] in The coordinates of the center point of the initial candidate box. This is the coordinate correction amount (calculated from the quantized value of the interference features). , ), The step size of the feature map is 16, and the corrected candidate box coordinate error does not exceed ±0.2m.

[0176] (3) Multi-dimensional position calibration. The coordinates of the center point of the corrected fault candidate box... Substituting into the coordinate mapping formula, the latitude and longitude coordinates of the fault center point are calculated. Simultaneously, by combining UAV point cloud data with transmission line ledger information, the fault location is identified by the tower number, span range (e.g., the span between towers 123 and 124), distance within the span (e.g., 50m from tower 123), and conductor phase (A phase, B phase, C phase). For hidden hazards, the specific location of the hazard is additionally marked (e.g., A phase conductor within span 123-124, 3.2m above the ground, insufficient distance from the tree line), forming a multi-dimensional location calibration result with a positioning accuracy error of no more than ±0.5m, ensuring that the fault location can be accurately traced.

[0177] (4) Verification of calibration results. The system automatically compares the fault calibration location with the tower coordinates and span parameters in the transmission line ledger, calculates the deviation value, and determines the calibration to be valid when the deviation value is ≤0.5m; when the deviation value is >0.5m, it automatically triggers the UAV re-fly command to re-collect the image and point cloud data of the area and re-calibrate until the calibration is valid, ensuring the reliability of the location calibration.

[0178] 2. Fault confidence assessment and recovery mechanism.

[0179] To ensure the accuracy of integrated hazard detection, based on flood interference characteristics and fault characteristics, the confidence score of the fault candidate box is calculated, and an adaptive re-flight mechanism is established, as follows: (1) Calculate the confidence score. The calculation formula will not be repeated here.

[0180] (2) Re-flight mechanism trigger. When the confidence score is ≥0.7, the detection result is deemed valid, and the calibration result and detection result are retained; when the confidence score is <0.7, the system automatically prompts "partial obstruction / strong interference, it is recommended to re-flight", and triggers the UAV re-flight command to adjust the flight altitude (±10m) and flight angle (±5°), re-collect the image and point cloud data of the area, and perform anti-interference modeling, integrated detection and position calibration again until the confidence score is ≥0.7; if the confidence score is still <0.7 after 3 consecutive re-flights, the system automatically marks the area as a suspected hidden danger area and generates a manual inspection prompt, notifying the operation and maintenance personnel to bring special equipment to the site for investigation to ensure that no hidden dangers are missed.

[0181] 3. Generation of structured operation and maintenance instructions.

[0182] Based on fault type, hazard level, and location information, and combined with a pre-set fault-handling measures knowledge base, structured operation and maintenance instructions with priorities are automatically generated to ensure that operation and maintenance personnel can directly carry out handling work according to the instructions, realizing a closed loop of "detection-calibration-handling". The specific steps are as follows.

[0183] (1) Fault-Handling Measures Knowledge Base Construction. The knowledge base covers all typical visible faults and hidden hazards of transmission lines, including fault types, hazard level classification standards, standardized handling steps, required tools and equipment, safety precautions, etc. It adopts structured storage and can be dynamically updated according to actual operation and maintenance needs. Among them, the hazard level is divided into four levels: emergency, high, medium and low. The classification standards are as follows: ① Emergency level: faults such as broken wires, tower collapse, arc flashover, and water leakage of conductors that may immediately cause large-scale power outages or casualties, which require on-site handling within 1 hour; ② High level: hazards such as broken insulators, loose or missing hardware, and severe sag exceeding the limit (more than 1.5 times the threshold) that may cause secondary faults, which require on-site handling within 2-4 hours; ③ Medium level: hazards such as insufficient tree-line distance, slight sag exceeding the limit (0.5-1.5 times the threshold), and slight tilting of towers, which require on-site handling within 6-12 hours; ④ Low level: hazards such as slight accumulation of dirt on conductors and slight hanging of floating objects on the line (not affecting line operation), which require on-site handling within 24 hours.

[0184] (2) Operation and maintenance instruction matching. The system automatically extracts the fault type (such as "line break"), the hazard level (such as "emergency level"), and the location information (such as "123 pole, latitude and longitude Lon: 118.XXXXX, Lat: 32.XXXXX"), and matches them with the fault-handling measures knowledge base to select the corresponding standardized handling steps, required tools and equipment (such as carrying wire cutters, wire connectors, insulating gloves, etc. for line break handling), and safety precautions (such as wearing insulating boots and avoiding approaching waterlogged areas in flood scenarios).

[0185] (3) Structured instruction generation. The operation and maintenance instructions adopt a fixed format, which includes nine core modules: instruction number, fault / hazard description, location information (latitude and longitude, tower number, distance within the span), hazard level, handling priority, standardized handling steps, required tools, completion time limit, and safety tips. An example is shown below.

[0186]

Instruction Number

[0187] [Fault / Hazard Description] The A-phase conductor of tower 123 is broken and submerged in water, posing a risk of leakage.

[0188]

Location Information

[0189] [Hazard Level] Emergency.

[0190] [Priority of Handling] Highest.

[0191] [Standardized Handling Procedures] 1. Immediately disconnect the power supply to towers 1, 2, and 3, and hang a "Do Not Close, Work in Progress" sign; 2. Maintenance personnel should wear insulated boots and gloves, carry wire cutters, insulated ropes, wire connectors, and other tools, and proceed to the site by emergency boat; 3. Clear floating debris around the broken wire to prevent secondary entanglement; 4. Replace the damaged wire, connect the wire connectors, and ensure the connections are secure and well-insulated; 5. Test the insulation performance of the wires, and restore power supply after confirming there is no leakage; 6. Record the handling process and update the line maintenance log.

[0192]

Required tools

[0193] [Completion Time Limit] Within 1 hour.

[0194]

Safety Tips

[0195] (4) Command output and feedback. After the structured operation and maintenance command is generated, it is sent synchronously to the mobile terminal of the operation and maintenance personnel and the emergency command center platform through the wireless communication module to ensure that the operation and maintenance personnel receive the command in real time. After the operation and maintenance personnel complete the handling, they provide feedback on the handling result on the terminal (such as "Handling completed, line returned to normal" or "No fault found, suspected false alarm"). The system automatically records the handling result, updates the hidden danger investigation ledger, and forms a complete closed loop of "detection-calibration-handling-feedback".

[0196] The process for fault detection and location calibration of transmission lines is as follows: Figure 4 As shown.

[0197] The overall collaborative workflow of the intelligent method in this embodiment achieves seamless integration of anti-interference modeling, integrated detection, accurate calibration, and operation and maintenance instruction generation. The specific process is as follows.

[0198] 1. Preliminary Preparations. Establish a drone inspection system (equipped with a visible light camera, LiDAR, and GPS positioning module) to collect historical fault images and point cloud data of power transmission lines during floods, construct an anti-interference training dataset, and complete model training and parameter debugging; build a fault-handling measures knowledge base and input standardized handling procedures and tool information.

[0199] 2. On-site inspection. The drone flies along a preset inspection route, collecting real-time images of the power transmission lines, point cloud data, and its own GPS positioning information. The flight altitude is controlled between 50-100m, the image resolution is no less than 1920×1080, and the point cloud acquisition density is no less than 100 points / m². 2 This ensures that the data quality meets the requirements of subsequent processing.

[0200] 3. Anti-interference modeling. The acquired images are preprocessed, and based on the flood interference multimodal quantification modeling system, interferences such as water surface turbulence texture, floating object occlusion, and low-light water reflection are quantified. The interference-weighted NMS algorithm is used to optimize the model detection effect, thus completing the anti-interference modeling.

[0201] 4. Integrated Hazard Detection. Based on anti-interference modeling results, visible faults are identified; through non-contact spatial geometric measurement, tree line safety distances and conductor sag are calculated to determine hidden hazards, and the detection results and hazard levels are output.

[0202] 5. Precise location calibration. By fusing interference features and spatial geometric parameters, the coordinates of the fault candidate boxes are corrected, a mapping between pixel coordinates and latitude and longitude coordinates is established, and multi-dimensional fault location calibration is completed. The calibration effectiveness is verified, and a re-flight mechanism is triggered when necessary.

[0203] 6. Maintenance Instruction Generation and Handling. Based on the detection results, hazard level, and location information, the system matches the fault-handling measures knowledge base to automatically generate structured maintenance instructions, which are then sent to the maintenance personnel's terminals. Maintenance personnel carry out handling work according to the instructions, provide feedback on the handling results, and the system updates the ledger, completing the closed loop.

[0204] 7. Post-incident review. The system automatically calculates indicators such as detection accuracy, missed detection rate, and handling efficiency, analyzes the impact of flood interference on detection results, optimizes anti-interference modeling parameters and fault-handling measures knowledge base, improves the robustness and practicality of the method, and is applicable to emergency inspection and handling of transmission lines in various flood disaster scenarios.

[0205] The overall implementation process of the above embodiments is as follows.

[0206] S1: Construct an anti-interference training dataset, adopt a positive and negative sample selection mechanism based on the optical features of flood hydrology, optimize the sample selection index by combining the color features of water and sediment in flood scenarios, control the balance between the number of negative samples and the number of positive samples, complete sample preprocessing and feature extraction, and provide data support for anti-interference modeling.

[0207] S2: Construct a multimodal quantitative modeling system for flood disturbances. Quantify and model flood-specific disturbances (water surface turbulence texture, floating object occlusion, and low-light water reflection) separately. Transform the disturbances into computable feature coefficients, construct the corresponding loss function for the disturbances, ensure that the quantitative output approximates the physical reality, and adopt the disturbance weighted non-maximum suppression algorithm to dynamically adjust the suppression threshold, achieve anti-missed detection optimization, and complete the anti-interference modeling.

[0208] S3: Employing a non-contact spatial geometric measurement method, the pixel coordinates of the transmission line are extracted through image recognition. The weighted least squares method is used to fit the conductor parabola, establishing a geometric mapping relationship from two-dimensional image to three-dimensional space. This enables the calculation of tree-line safety distance and conductor sag inversion, completing the integrated detection and hazard assessment of visible faults and hidden hazards in transmission lines.

[0209] S4: Construct a multi-fault intelligent diagnosis and operation and maintenance instruction generation module, which integrates the flood interference visual features extracted in step S2 and the spatial geometric parameters calculated in step S3 to complete the accurate calibration of fault location (including latitude and longitude, tower position, and distance within the span), and automatically generates structured operation and maintenance instructions with priority based on the preset fault-handling measures knowledge base, realizing the detection-calibration-handling closed loop.

[0210] The feasibility of this embodiment will be verified below.

[0211] To verify the feasibility and superiority of this embodiment, the technical effects of each step were verified through actual flood scenarios and data comparison. The specific verification results are as follows.

[0212] (1) Verification of anti-interference modeling effect. 1000 images of transmission lines in flood scenes (including water surface turbulence, floating object obstruction, low light reflection and other interference) were selected. The anti-interference modeling method of this embodiment was compared with the existing conventional image recognition method. The results showed that the fault detection accuracy of the method of this embodiment reached 95.3%, the false detection rate was 3.2%, and the false detection rate was 1.5%; the detection accuracy of the existing conventional method was 78.6%, the false detection rate was 15.8%, and the false detection rate was 5.6%. The anti-interference ability of this embodiment is significantly better than that of the existing technology.

[0213] (2) Verification of integrated detection effect. Fifty transmission lines with a span of less than 500m were selected (covering three voltage levels: low voltage, 10kV, and 35kV). A flood scenario was simulated, and the method of this embodiment was used to detect visible faults and hidden hazards. The results showed that the accuracy rate of visible fault detection was 96.7%, and the accuracy rate of hidden hazard detection was 94.2%. It can effectively identify hidden hazards such as insufficient tree-line distance and abnormal sag. The detection range is comprehensive, making up for the deficiency of existing technologies that can only detect visible faults.

[0214] (3) Verification of location calibration accuracy. Twenty fault points with known coordinates were selected and their locations were calibrated using the method of this embodiment. The results showed that the calibration accuracy error was ≤0.5m and the average error was 0.32m, which meets the actual needs of the project and can ensure that maintenance personnel can quickly locate the fault location.

[0215] (4) Verification of emergency response efficiency. Ten simulated flood scenarios (covering emergency, high, medium and low levels) were selected. The method of this embodiment was used to generate operation and maintenance instructions and carry out the response. Compared with the existing manual response method, the results showed that the average response time of the method of this embodiment was 1.2 hours and the average completion time was 3.5 hours; the average response time of the existing manual response method was 4.8 hours and the average completion time was 8.2 hours. The method of this embodiment significantly improves the emergency response efficiency and reduces disaster losses.

[0216] The method for detecting transmission line faults under flood disasters in this embodiment has the following advantages.

[0217] (1) Based on the positive and negative sample screening mechanism of flood interference characteristics, combined with the weight optimization of blue-gray water and yellow-brown mud and sand and the positive and negative sample quantity balancing strategy, the training dataset is more in line with the actual interference characteristics of flood disaster scenarios. At the same time, through the multimodal quantitative modeling mechanism of flood interference, the accurate quantification and adversarial training of water surface turbulence, floating object occlusion and low light reflection are realized, which greatly improves the robustness of the model in fault identification under complex flood interference and effectively reduces the situation of missed detection and false detection.

[0218] (2) By using an image-based non-contact spatial geometric measurement method, a geometric mapping relationship from two-dimensional image to three-dimensional space is constructed to realize the calculation of tree line safety distance and conductor sag inversion. This breaks through the limitation that traditional simple visual recognition can only detect visible faults. It can accurately diagnose hidden structural hazards such as insufficient tree line distance and abnormal sag (see step four for specific hazard types), realize the integrated detection of visible faults and hidden hazards, and improve the comprehensiveness of transmission line fault detection.

[0219] (3) Based on the intelligent diagnostic output layer of multiple faults, the fault type identification, hidden danger status judgment and risk level assessment are completed simultaneously through the multi-branch output layer. Combined with the interference perception mechanism, the fault candidate box is corrected to achieve accurate location calibration. Finally, a graded early warning signal and a structured diagnostic report are generated, making the fault detection results more accurate and the output information more comprehensive, which greatly improves the response efficiency and decision-making scientificity of emergency inspection of transmission lines in flood scenarios.

[0220] The following is a more specific example illustrating the implementation process of a fault detection method for power transmission lines during floods.

[0221] In the context of flood disasters, the entire process of monitoring and handling various faults in transmission lines using the method in this embodiment is as follows.

[0222] The drone, equipped with a visible light camera, flies along the route to collect images in real time. After the images are transmitted to the ground workstation, they are first preprocessed by standardization and normalization, and then the positive and negative sample features in the anti-interference training dataset constructed in step one are used to perform preliminary screening and feature extraction on the images.

[0223] To address the unique interferences in flood scenarios, such as water surface ripple textures, floating object occlusion, and low-light reflections, step two uses a multimodal quantization module to calculate the texture ripple intensity, occlusion index, and illumination condition vector for each frame. Based on these quantization metrics, the model automatically adjusts the detection threshold and uses an interference-weighted non-maximum suppression algorithm to suppress dynamic water body interference, ensuring accurate extraction of fault features even in complex environments.

[0224] Step four's multi-task output layer classifies and identifies candidate regions, outputting the probability distribution of fault types. The model outputs a confidence score, which is then evaluated. If the confidence score is less than 0.7, occlusion exists, and the model indicates "partial occlusion, go-around recommended," arranging for the drone to go-around. If the confidence score is greater than 0.7, step three simultaneously calculates the three-dimensional spatial distance and inverted sag value using a non-contact spatial geometric measurement method. The system maps the candidate box coordinates to latitude and longitude and outputs the results.

[0225] The output results are shown in the following examples: {Result: Fault, Type: Arc flashover, Confidence: 0.95, Additional Hazard: Insufficient tree-line distance (1.2m)}; {Result: Hazard, Type: Abnormal sag, Severity: 0.75}.

[0226] The obtained output results and the feature-extracted acquired images (such as...) Figure 5 The issues (as shown in Table 2) are summarized to form an issue document.

[0227] Table 2 Problem Documents

[0228] After identifying and locating the fault, the system automatically triggers a tiered response mechanism. The core handling process is as follows: Based on the identified fault type, confidence level, and location information, the system generates a reminder message containing specific operating steps and pushes it to the relevant maintenance or emergency repair personnel. Following the handling recommendations, the relevant personnel, carrying the necessary tools and equipment, quickly arrive at the site to perform precise repairs or eliminate potential hazards. The fault types and the reminder message content containing operating steps are shown in Table 3.

[0229] Table 3 Reminder Message Content

[0230] All aspects of this application are feasible, and its detection accuracy, anti-interference ability, and emergency response efficiency are superior to existing technologies. It can meet the actual needs of emergency inspection and precise handling of transmission lines in flood disaster scenarios, and has good practicality and promotion value.

[0231] Based on the same inventive concept, this application also provides a flood-prone transmission line fault detection device for implementing the above-mentioned method for detecting transmission line faults under flood disasters. The solution provided by this device is similar to the solution described in the above-described method. Therefore, the specific limitations of one or more embodiments of the flood-prone transmission line fault detection device provided below can be found in the limitations of the flood-prone transmission line fault detection method described above, and will not be repeated here.

[0232] In one exemplary embodiment, such as Figure 6 As shown, a fault detection device for transmission lines under flood disasters is provided, including: a data acquisition module 601, used to acquire target transmission line images and target point cloud data in flood scenarios.

[0233] The quantization model construction module 602 is used to input the target transmission line image into the flood interference multimodal quantization model to obtain the flood interference features of the target transmission line image; the flood interference features include: texture fluctuation intensity, occlusion index and illumination statistics; the flood interference multimodal quantization model is obtained by training a deep feature extraction network based on a positive and negative sample selection mechanism.

[0234] The visible fault identification module 603 is used to identify visible faults in the target transmission line image based on the flood interference features and classification dynamic threshold, and obtain visible fault identification results.

[0235] The latent fault identification module 604 is used to detect latent faults in the target power transmission line image based on the target point cloud data using a non-contact spatial geometric measurement method, and obtain latent fault detection results; the latent fault identification results include: whether there are hidden dangers in the tree line safety distance, whether there are hidden dangers in the conductor sag, whether there are hidden dangers in the ground safety distance, and whether there are hidden dangers in the water safety distance.

[0236] The fault location calibration module 605 is used to calibrate the fault location based on the flood interference features of the target transmission line image, the target point cloud data, the visible fault identification result, and the hidden fault identification result, so as to obtain the multi-dimensional fault location calibration result.

[0237] This application constructs an anti-interference training dataset based on a negative sample screening mechanism using flood disturbance characteristics. Through multimodal quantization modeling, hydrological disturbances such as water surface turbulence texture, floating object occlusion, and low-light water reflection are transformed into computable features. Non-contact spatial geometric measurement methods are used to calculate tree-line safety distances and invert conductor sag. Finally, a multi-fault intelligent diagnosis output layer integrates visual features and geometric parameters to simultaneously output fault types, hazard status, and risk levels. This application effectively solves the problem that traditional inspection methods are difficult to implement in flood disaster scenarios due to high safety risks. Through intelligent image recognition and geometric measurement, it achieves remote and accurate diagnosis and location of faults such as tower collapse, line breakage, and abnormal sag, significantly improving the emergency response speed and operational safety of the power grid under extreme weather conditions.

[0238] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores multi-dimensional fault location calibration results. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a method for detecting transmission line faults during floods.

[0239] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment to which the present application is applied. Specific computer equipment may include, for example, [the following is a list of possible additional structures]. Figure 7 The embodiments show more or fewer components, combinations of certain components, or different component arrangements. In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, which the processor executes to implement the steps in the above-described method embodiments.

[0240] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0241] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0242] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0243] 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, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0244] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0245] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0246] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for detecting transmission line faults under flood disasters, characterized in that, include: Acquire images and point cloud data of target power transmission lines in flood scenarios; The target transmission line image is input into the flood interference multimodal quantization model to obtain the flood interference features of the target transmission line image; The flood disturbance features include: texture fluctuation intensity, occlusion index, and illumination statistics; the flood disturbance multimodal quantification model is obtained by training a deep feature extraction network based on a positive and negative sample selection mechanism; Based on the flood interference features and classification dynamic threshold of the target transmission line image, visible faults in the target transmission line image are identified to obtain visible fault identification results; Based on the target point cloud data, a non-contact spatial geometric measurement method is used to detect hidden faults in the target power transmission line image, and the hidden fault detection results are obtained. The hidden fault identification results include: whether there are hidden dangers in the tree line safety distance, whether there are hidden dangers in the conductor sag, whether there are hidden dangers in the ground safety distance, and whether there are hidden dangers in the water safety distance. Based on the flood interference features of the target transmission line image, the target point cloud data, the visible fault identification results, and the hidden fault identification results, the fault location is calibrated to obtain a multi-dimensional fault location calibration result.

2. The method for detecting transmission line faults under flood disasters according to claim 1, characterized in that, After obtaining the multi-dimensional fault location calibration results, the following is also included: Based on the visible fault identification results, the hidden fault identification results, the multi-dimensional fault location calibration results, and the established fault-handling measures knowledge base, structured operation and maintenance instructions are generated to handle the faults in the target transmission line image.

3. The method for detecting transmission line faults under flood disasters according to claim 2, characterized in that, Before generating structured operation and maintenance instructions, the following is also included: Calculate the confidence score based on the visible fault identification results and the flood interference features of the target transmission line image; Based on the confidence score and the set score value, it is determined whether to trigger the re-flight mechanism. If so, the target power line image and target point cloud data in the flood scene are reacquired.

4. The method for detecting transmission line faults under flood disasters according to claim 1, characterized in that, The method for determining the multimodal quantization model of flood disturbance includes: Acquire historical fault images of power transmission lines in flood scenarios; The historical fault images of the power transmission line are identified based on visual markers, and the fault rectangular area is selected. The fault rectangle region is cropped and its features are extracted to obtain a positive sample set; The region outside the fault rectangle region in the historical fault image of the power transmission line is identified as a negative sample candidate region. The negative sample candidate regions are scored and sorted based on the negative sample screening index to obtain a negative sample set. The negative sample screening index is determined based on the mean color feature vector of the positive samples in the positive sample set, the mean color feature vector of the negative sample candidate regions, the feature difference weight, the scene adaptation weight, and the color interference weight coefficient of the flood scene. The number of positive samples in the positive sample set and the number of negative samples in the negative sample set are balanced to obtain the training sample set; The training sample set and the corresponding texture fluctuation intensity are input into the water surface turbulence texture quantization model, the negative sample set and the corresponding occlusion index are input into the floating object occlusion quantization model, and the training sample set and the corresponding illumination statistics are input into the low-light water body reflection quantization model. Collaborative training is performed with the goal of minimizing the loss function value. The trained water surface turbulence texture quantization model, the trained floating object occlusion quantization model, and the trained low-light water body reflection quantization model are determined as the flood disturbance multimodal quantization model. The water surface turbulence texture quantization model, the floating object occlusion quantization model, and the low-light water body reflection quantization model are all deep feature extraction networks.

5. The method for detecting transmission line faults under flood disasters according to claim 1, characterized in that, Based on the flood interference features and classification dynamic threshold of the target transmission line image, visible faults in the target transmission line image are identified to obtain visible fault identification results, specifically including: Based on the aforementioned flood disturbance characteristics, the disturbance weighted nonmaximum suppression algorithm is used to dynamically adjust the suppression threshold to obtain the classification dynamic threshold; The flood interference features of the target transmission line image are input into the fault classifier to obtain preliminary identification results of visible faults; The preliminary identification results of the visible faults are filtered based on the classification dynamic threshold to obtain the final visible fault identification results.

6. The method for detecting transmission line faults under flood disasters according to claim 1, characterized in that, Based on the target point cloud data, a non-contact spatial geometric measurement method is used to detect latent faults in the target transmission line image, obtaining latent fault detection results, specifically including: Based on the target point cloud data, a non-contact spatial geometric measurement method is used to extract the coordinates of the power transmission line pixels in the target power transmission line image to obtain the power transmission line pixel coordinates. The weighted least squares method is used to perform curve fitting on the pixel coordinates of the transmission line to obtain the conductor parabola. Calculate the theoretical height of the power transmission line at the projection point of the vegetation point on the power transmission line based on the parabola of the conductor. Based on the theoretical height of the power transmission line, the Euclidean distance from the vegetation point to the corresponding projection point in the cross section is calculated. The minimum Euclidean distance from all vegetation points to their corresponding projection points is determined as the treeline safety distance, and the treeline safety distance and the set treeline safety threshold are used to determine whether there are any potential hazards. Based on the target point cloud data, the conductor sag is calculated according to the height difference of the transmission line suspension points, span, transmission line parameters and operating conditions, and the presence of potential hazards in the conductor sag is determined based on the conductor sag and the set sag safety threshold range. The ground safety distance and water safety distance are calculated based on the parabola of the conductor, and the presence of potential hazards in the ground safety distance and water safety distance are determined based on the ground safety distance, the water safety distance, the set ground safety threshold, and the set water safety distance threshold.

7. The method for detecting transmission line faults under flood disasters according to claim 1, characterized in that, Based on the flood interference features of the target transmission line image, the target point cloud data, the visible fault identification results, and the latent fault identification results, the fault location is calibrated to obtain a multi-dimensional fault location calibration result, specifically including: The GPS positioning information of the drone used to collect images of the target power transmission line was obtained. Establish the mapping relationship between the pixel coordinates of the target transmission line image and the latitude and longitude coordinates of the GPS positioning information, and obtain the coordinate mapping formula; The initial coordinates of the fault candidate boxes corresponding to the visible fault identification results and the latent fault identification results are determined, and the initial coordinates of the fault candidate boxes are corrected using the flood interference features to obtain the corrected coordinates of the fault candidate boxes. Substitute the center point coordinates in the fault candidate box correction coordinates into the coordinate mapping formula to obtain the latitude and longitude coordinates of the fault center point. Based on the latitude and longitude coordinates of the fault center point, the target point cloud data, and the transmission line ledger information, the location calibration result of the visible fault is determined; The location calibration result of the latent fault is determined based on the latitude and longitude coordinates of the fault center point; Based on the location calibration results of visible faults and latent faults, the multi-dimensional location calibration results of faults are determined.

8. A fault detection device for power transmission lines under flood disasters, characterized in that, include: The data acquisition module is used to acquire images of target power transmission lines and target point cloud data in flood scenarios; The quantization model construction module is used to input the target transmission line image into the flood interference multimodal quantization model to obtain the flood interference features of the target transmission line image; The flood disturbance features include: texture fluctuation intensity, occlusion index, and illumination statistics; the flood disturbance multimodal quantification model is obtained by training a deep feature extraction network based on a positive and negative sample selection mechanism; The visible fault identification module is used to identify visible faults in the target transmission line image based on the flood interference features and classification dynamic threshold, and obtain visible fault identification results. The latent fault identification module is used to detect latent faults in the target power transmission line image based on the target point cloud data using a non-contact spatial geometric measurement method, and obtain latent fault detection results; the latent fault identification results include: whether there are hidden dangers in the tree line safety distance, whether there are hidden dangers in the conductor sag, whether there are hidden dangers in the ground safety distance, and whether there are hidden dangers in the water safety distance; The fault location calibration module is used to calibrate the fault location based on the flood interference features of the target transmission line image, the target point cloud data, the visible fault identification results, and the hidden fault identification results, so as to obtain the multi-dimensional fault location calibration results.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the method for detecting transmission line faults under flood disasters as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for detecting transmission line faults under flood disasters as described in any one of claims 1-7.

Images