A power transmission line fault identification and early warning method and system based on a UAV

Abstract

This invention provides a method and system for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs), relating to the field of image processing technology. The method includes: acquiring raw image data containing various components of the power transmission line; preprocessing the raw image data; extracting regions of interest from the preprocessed raw image data using a multi-stage component detection model to obtain multiple component sub-images; identifying faults in each component sub-image using a spatial attention convolutional neural network model to determine component-level fault categories; based on the component-level fault categories, fusing the spatial topology of the power transmission line and component risk characteristics to determine the fault type of the power transmission line; and performing early warning judgment and graded response based on the fault type. This invention can automatically identify fault categories without requiring manual frame-by-frame judgment, significantly improving work efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Traditional power transmission line inspections mainly rely on manual tower climbing or ground telescope observation, which has problems such as low efficiency, high labor intensity and high safety risks. Moreover, it is difficult to carry out in high-voltage, mountainous or harsh weather conditions. With the development of drone technology and artificial intelligence, using drones for line inspections has become a trend.

[0003] Drones can quickly cover large areas of power lines, reduce the risks of manual tower climbing operations, and significantly improve inspection efficiency and safety. Using drones for power line fault identification can provide data support for smart grid operation and maintenance, and promote the development of power inspection towards digitalization, intelligence and unmanned operation. It has broad application prospects and promotion value.

[0004] However, existing UAV-based power transmission line fault identification methods mostly rely on single-stage detection models, which struggle to accurately segment component areas in complex backgrounds, leading to false positives and false negatives. Secondly, existing algorithms generally lack comprehensive analysis of spatial topological relationships and inter-component correlations, failing to accurately determine fault types at the overall line level. This results in inaccurate responses, insufficient identification precision, and a lack of multi-component collaborative analysis and fault risk warning mechanisms, making it difficult to meet the accuracy requirements of modern fault identification. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs). This method addresses the shortcomings of existing UAV-based power transmission line fault identification, which often relies on single-stage detection models, making it difficult to accurately segment component regions in complex backgrounds and prone to false detections and missed detections. Furthermore, existing algorithms generally lack comprehensive analysis of spatial topological relationships and inter-component correlations, failing to accurately determine fault types at the overall line level. This results in inaccurate responses, insufficient identification precision, and a lack of multi-component collaborative analysis and fault risk early warning mechanisms, making it difficult to meet the technical requirements of modern fault identification accuracy.

[0006] A first aspect of this invention proposes a method for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs), comprising:

[0007] S1: Acquire raw image data containing various components of the transmission line;

[0008] S2: Preprocess the raw image data;

[0009] S3: Using a multi-stage component detection model, the region of interest in the preprocessed original image data is extracted to obtain multiple component sub-images;

[0010] S4: Using a spatial attention convolutional neural network model, fault identification is performed on each component sub-image to determine the component-level fault category;

[0011] S5: Based on the component-level fault categories, integrate the spatial topology of the transmission line with the risk characteristics of the components to determine the fault type of the transmission line;

[0012] S6: Based on the fault type, perform early warning judgment and graded response.

[0013] A second aspect of this invention provides a power transmission line fault identification and early warning system based on unmanned aerial vehicles (UAVs), comprising: a processor and a memory;

[0014] The memory stores programs or instructions that can run on a processor, and when the programs or instructions are executed by the processor, they implement the steps of the UAV-based power transmission line fault identification and early warning method of the first aspect.

[0015] A third aspect of the present invention provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the UAV-based power transmission line fault identification and early warning method of the first aspect.

[0016] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0017] In this embodiment of the invention, by introducing a multi-stage component detection model and a spatial attention convolutional neural network, key components such as towers, conductors, and insulators can be accurately segmented in complex backgrounds, achieving multi-level feature extraction and enhancement of key areas. Through the fault saliency index and probability adjustment mechanism, false alarms and missed alarms are effectively reduced. Furthermore, by combining the spatial topology relationship of the transmission line and environmental operating parameters, a Bayesian model is used for risk fusion and fault type reasoning, realizing intelligent judgment from the component level to the line level. Finally, through the early warning classification mechanism, potential risks can be identified in advance, maintenance decisions can be optimized, and the operational safety and maintenance efficiency of the transmission line can be improved. Attached Figure Description

[0018] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0019] Figure 1 This is a flowchart illustrating a method for identifying and warning of power transmission line faults based on unmanned aerial vehicles (UAVs) provided in an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of a power transmission line fault identification and early warning system based on unmanned aerial vehicles (UAVs) provided in an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] The following description, in conjunction with the accompanying drawings, details the method for identifying and warning of power transmission line faults based on unmanned aerial vehicles (UAVs) provided by the present invention through specific embodiments and application scenarios.

[0023] Reference manual attached Figure 1 The diagram shows a flowchart of a method for identifying and warning of power transmission line faults based on unmanned aerial vehicles (UAVs) according to an embodiment of the present invention.

[0024] This invention provides a method for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs), which may include the following steps:

[0025] S1: Obtain raw image data containing various components of the transmission line.

[0026] Raw image data refers to unfiltered and unprocessed image or video frame data collected by drones using high-definition cameras during inspections.

[0027] It should be noted that drones can fly flexibly in complex terrain, mountainous areas, and high-pressure environments, enabling multi-angle, full-field-of-view acquisition of towers and components, avoiding the safety hazards associated with manual tower climbing. The raw image data is uncompressed and unprocessed, preserving details such as lighting, texture, and color to the maximum extent possible. This provides a high-quality data foundation for subsequent image preprocessing and fault identification, thereby improving the accuracy and stability of model feature extraction and laying the foundation for the high-precision operation of the overall recognition system.

[0028] S2: Preprocess the raw image data.

[0029] Specifically, the preprocessing includes additive Gaussian noise preprocessing, Gaussian blur preprocessing, rotation preprocessing, and scaling preprocessing.

[0030] S3: By using a multi-stage component detection model, the region of interest in the preprocessed original image data is extracted to obtain multiple component sub-images.

[0031] Among them, the multi-stage component detection model refers to a layered detection structure, which usually includes two levels: tower detector and component detector. First, the overall tower is located, and then each component is identified in detail.

[0032] The Region of Interest (ROI) refers to the important area in the image related to fault identification, such as key parts like insulators and conductors. Component sub-images refer to independent image patches containing one or a few components cropped from the original image, used for subsequent depth feature analysis.

[0033] It should be noted that by first performing tower-level detection and then extracting component-level images, not only is the detection accuracy and robustness improved, but the impact of background interference on the recognition results is also effectively reduced. This process decomposes the entire image into multiple component sub-images, allowing the subsequent fault recognition model to focus on feature extraction of the target region, thereby improving feature representation ability and classification accuracy. Simultaneously, the multi-stage detection structure enables adaptive recognition of targets at different scales, enhancing the system's stability and generalization ability under different shooting angles, lighting conditions, and distances.

[0034] In one possible implementation, S3 specifically includes:

[0035] S301: Input the preprocessed raw image data into the multi-stage part detection model.

[0036] S302: Determine the tower boundary box image of the transmission line through the tower detector in the multi-stage component detection model.

[0037] S303: Cropping the tower boundary frame image to obtain the cropped boundary frame.

[0038] S304: Based on the cropping bounding box, the component detector in the multi-stage component detection model outputs a component sub-image containing multiple component regions. The component regions include the coordinate position, confidence score, and component category label of each component.

[0039] The components include: top cap, crossarm, insulator, and pole.

[0040] It should be noted that the target area is first located using a tower detector, and then the identification is refined using a component detector, significantly improving detection accuracy. Secondly, the bounding box cropping method effectively reduces background interference, lowers the computational complexity of the algorithm, and improves detection speed and stability. Furthermore, by outputting the component's coordinates, confidence level, and category label, the spatial localization and semantic annotation of the component are unified, providing high-quality input data for subsequent fault identification and risk analysis. Overall, it possesses comprehensive advantages of high accuracy, high robustness, and scalability.

[0041] S4: Using a spatial attention convolutional neural network model, fault identification is performed on each component sub-image to determine the component-level fault category.

[0042] Among them, the spatial attention convolutional neural network model is a deep learning structure that combines convolutional neural networks (CNN) with attention mechanisms, which can automatically focus on key regions in the image related to faults during the feature extraction process.

[0043] Among them, component-level fault categories refer to the specific fault type identification results for a single component, such as "damage", "discharge", "flashover", etc.

[0044] It should be noted that this model can adaptively focus on regions with abnormal features in the image, enhance the response of key features, and suppress irrelevant background interference, thereby significantly improving the accuracy and robustness of fault identification. Compared with traditional convolutional networks, the spatial attention mechanism enables the model to accurately identify different types of component faults even under complex lighting, angle, or noise conditions. In addition, the multi-layer convolution and feature fusion structure improves the sensitivity to subtle damage, making the identification results more stable and interpretable, and providing reliable data support for subsequent fault assessment and early warning.

[0045] In one possible implementation, the spatial attention convolutional neural network model includes an input layer, a feature extraction layer, a feature concatenation layer, a flattening layer, a fully connected layer, and an output layer connected in sequence.

[0046] S4 specifically includes:

[0047] S401: Input the sub-images of each component into the spatial attention convolutional neural network model through the input layer.

[0048] S402: Through the feature extraction layer, deep features of each component sub-image are extracted to generate multiple output feature maps.

[0049] In one possible implementation, S402 specifically includes:

[0050] S4021: The first convolutional unit of the feature extraction layer performs multi-channel convolution processing on the component sub-image to generate the first feature map.

[0051] The convolutional unit is a two-dimensional convolutional unit.

[0052] S4022: Calculate the attention weights between pixels in the first feature map through the first self-attention unit of the feature extraction layer.

[0053] In one possible implementation, S4022 specifically includes:

[0054] S4022A: Calculate the similarity between pixels in the first feature map using the first self-attention unit:

[0055]

[0056] Wherein, s(X) p ,X q X represents the similarity between the features of pixel p and the features of pixel q. p X represents the feature of pixel p in the first feature map. q This represents the feature of pixel q in the first feature map. This represents the Query transformation weight matrix. This represents the Key transformation weight matrix. T 'e' represents the transpose, and 'e' represents the exponential function.

[0057] S4022B: Normalize the similarity to obtain the normalized weights:

[0058]

[0059] Wherein, C(X) p ) represents the normalization factor of pixel p in the first feature map.

[0060] S4022C: Based on normalized weights, attention weights are determined through a weighted summation operation.

[0061]

[0062]

[0063] Among them, Y p C(X) represents the attention weight of pixel p. p h(X) represents the normalization factor for pixel p in the first feature map. q ) represents the feature map of pixel q. This represents the Value transformation weight matrix.

[0064] S4023: Based on attention weights, dimensionality reduction is performed on the first feature map through the first max pooling unit of the feature extraction layer.

[0065]

[0066] in, Let represent the output value of the i-th feature map at position (j,k) after pooling, and max indicates taking the maximum value. Let represent the input value of the i-th feature map at position (js+m, ks+n), where m represents the horizontal offset and n represents the vertical offset.

[0067] S4024: The second convolutional unit of the feature extraction layer performs multi-channel convolution processing on the first feature map after dimensionality reduction to generate the second feature map.

[0068] Specifically, a convolution unit refers to a neural network structure that calculates feature responses by sliding convolution kernels across an image, used to extract features such as local textures and edges.

[0069] S4025: Fault enhancement processing is performed on the second feature map through the second self-attention unit of the feature extraction layer.

[0070] The self-attention unit is used to calculate the correlation between different pixels or regions in an image, highlighting key region features and suppressing redundant information.

[0071] S4026: The second feature map after fault enhancement is reduced in dimensionality by using the minimum pooling unit of the feature extraction layer to generate a local minimum response feature map.

[0072] Among them, the min pooling unit takes the local minimum value to enhance the features of low response regions and help identify weak fault features.

[0073] S4027: The high-level semantic features of the local minimum response feature map are extracted through the third convolutional unit of the feature extraction layer to obtain the high-level semantic feature map.

[0074] Among them, high-level semantic feature maps refer to high-level abstract features formed after multiple layers of convolution and attention processing, which express the global semantic information of the fault.

[0075] S4028: The high-level semantic feature map is weighted and enhanced through the second self-attention unit of the feature extraction layer.

[0076] Specifically, by calculating the correlation weights between pixels or feature channels in the high-level semantic feature map, the significant response of the fault region is highlighted and the interference of the background or noise region is suppressed, thereby obtaining the attention-enhanced high-level semantic feature map.

[0077] S4029: The second max-pooling unit of the feature extraction layer performs dimensionality reduction on the processed high-level semantic feature map to generate multiple output feature maps:

[0078]

[0079] in, This represents the i-th output feature map in the r-th layer, and max indicates taking the maximum value. This represents the value of the j-th output feature map at layer r. Let represent the weights of the convolution kernel at layer r, which connects the i-th input local minimum response feature map to the j-th output feature map. This represents the value of the i-th local minimum response feature map at layer r.

[0080] Among them, the Max Pooling Unit refers to taking the maximum value of a local region of the feature map to achieve feature dimensionality reduction and preservation of salient features.

[0081] S403: The feature concatenation layer concatenates the output feature maps to generate a concatenated feature map.

[0082] S404: The spliced ​​feature map is expanded by a flattening layer to obtain a one-dimensional feature vector.

[0083] S405: A fully connected layer is used to map a one-dimensional feature vector to obtain multiple probability vectors. The fully connected layer includes a first fully connected layer and a second fully connected layer.

[0084] In one possible implementation, S405 specifically includes:

[0085] S4051: The first fully connected layer compresses the one-dimensional feature vector into low-dimensional semantic features.

[0086]

[0087] in, Indicates the first The output of the j-th neuron in the layer, Represents the ReLU activation function. Indicates the first The output of the i-th neuron in layer -1 Indicates the first -1 layer, from neuron i to the... The weight of the j-th neuron in the layer. Indicates the first The bias of the j-th neuron in the layer.

[0088] S4052: Through the second fully connected layer, low-dimensional semantic features are mapped to fault categories, resulting in multiple probability vectors:

[0089]

[0090] Among them, y i =m indicates that sample i belongs to fault category m, x i Let W represent the high-level features of sample i, and let W represent the weights of the Softmax layer. w represents the probability estimate that sample i belongs to fault category m. m w represents the learnable weight vector corresponding to fault category m in the Softmax layer. j Let K represent the learnable weight vector corresponding to fault category j in the Softmax layer, and K represent the total number of fault categories.

[0091] S406: The output layer adjusts each probability vector to determine the component-level fault category.

[0092] It should be noted that by combining feature extraction and stitching layers, the model can integrate low-level texture features with high-level semantic features, improving the accuracy of identifying subtle faults. Flattening and fully connected layers achieve efficient mapping of multi-dimensional feature vectors, enhancing the model's classification and discriminative power. Through probability vector adjustment and output layer decisions, the system can adaptively optimize confidence levels, reducing false positives. The overall structure ensures the model's stability and generalization performance in complex environments, providing a high-precision, end-to-end intelligent solution for fault identification of power transmission line components.

[0093] In one possible implementation, S406 specifically includes:

[0094] S4061: Calculate the fault significance index:

[0095]

[0096]

[0097]

[0098]

[0099] Among them, FSI i G represents the fault saliency index of the i-th component sub-image. i C represents the normalized value of the geometric gradient intensity of the i-th component sub-image. i L represents the foreground-background contrast index of the i-th component sub-image. i F represents the local deviation of the i-th component sub-image. i A represents the feature response map of the i-th component sub-image.i This represents the area of ​​the i-th component sub-image. and These represent the average gray levels of the foreground and background areas, respectively. and These represent the standard deviations of the foreground and background regions, respectively. Represents a minimal constant. Represents the set of foreground pixels. This represents the average of all response values ​​within the foreground region. This represents the response value of pixel p on the feature map. The weighting coefficients represent the normalized values ​​of the geometric gradient intensity. This represents the weighting coefficient of the foreground-background contrast index. The weighting coefficients represent the degree of local deviation. Let || denote the gradient, and ||2 denote the square of the norm.

[0100] S4062: Adjust each probability vector according to the fault significance index:

[0101]

[0102] in, This represents the adjusted probability vector. Represents a probability vector. This represents the significance adjustment coefficient.

[0103] S4063: Based on the adjusted probability vector, determine the fault classification result through the output layer.

[0104] S5: Based on the component-level fault category, integrate the spatial topology of the transmission line with the risk characteristics of the components to determine the fault type of the transmission line.

[0105] Spatial topology refers to the spatial connection relationship between the towers, conductors and insulators of the transmission line, and is used to describe the system structure.

[0106] Among them, the component risk characteristics refer to the risk score calculated by comprehensively considering factors such as component confidence level, location, aging, and environment.

[0107] Among them, the fault type refers to the fault category determined at the level of the entire line, such as three-phase grounding, two-phase short circuit, single-phase grounding, etc.

[0108] It should be noted that by fusing component-level identification results with the spatial topology information of transmission lines, intelligent reasoning from local faults to global line faults is achieved. This method not only considers the damage status of individual components but also integrates the spatial correlation between components and environmental risk characteristics, thereby improving the comprehensiveness and accuracy of fault diagnosis.

[0109] In one possible implementation, S5 specifically includes:

[0110] S501: Perform confidence calibration on component-level fault categories.

[0111] Among them, confidence calibration refers to adjusting the confidence level of component failure output by the model so that the predicted probability is more consistent with the actual situation. It is often achieved by combining power transformation and information entropy weight.

[0112] Specifically, the confidence level calibration is calculated as the product of the power-calibrated confidence level and the information entropy weighting factor.

[0113] S502: Based on the confidence calibration results, and using the UAV attitude parameters and camera intrinsic and extrinsic parameters, the component pixel coordinates in multiple component sub-images are projected onto the three-dimensional geographic coordinate system of the line.

[0114] Among them, the drone's attitude parameters describe the drone's attitude information (such as pitch angle, yaw angle, and roll angle) during the shooting process, which are used to accurately locate the position of components in the image.

[0115] Among the camera's intrinsic and extrinsic parameters, the intrinsic parameters represent the lens's imaging characteristics (focal length, distortion, etc.), while the extrinsic parameters describe the camera's position relative to world coordinates and are used for spatial projection.

[0116] Among them, the three-dimensional geographic coordinate system represents the three-dimensional coordinate system of the actual spatial location of the transmission line.

[0117] S503: Based on the three-dimensional geographic coordinate system of the route, the components are matched using the minimum distance matching method.

[0118] Specifically, matching components involves determining the pole or line segment number to which the component belongs.

[0119] S504: Based on the matching results, and combined with the spatial proximity relationships of towers, conductors, and insulators, construct a topology map.

[0120] Among them, a topology diagram refers to a graphical structure that describes the spatial connection relationships of components such as towers, conductors, and insulators.

[0121] S505: Based on the topology diagram, cross-domain environment and operating condition parameters are introduced to comprehensively correct component risks.

[0122]

[0123]

[0124]

[0125] in, This represents the corrected risk score for the i-th component. This represents the initial risk score of the i-th component calculated based on component-level fault categories. The topology enhancement weight coefficients are represented by , and w represents the fusion weight vector. This represents the neighborhood feature vector of the i-th component. This represents the environmental correction factor for the i-th component. This represents the weight of the m-th fault category. This represents the calibration confidence level of the i-th component. This represents the area weighting function for the i-th component. This indicates the weighting of correction coefficients for controlling meteorological factors. This represents the weighting of the correction coefficient for the control load factor. This indicates the weighting of the correction coefficients used to control for aging factors. Indicates meteorological factors, Indicates the load factor. This indicates aging factors.

[0126] S506: Based on the comprehensive correction of component risks, construct a set of line-level fault assumptions.

[0127] S507: Based on the set of line-level fault assumptions, determine the fault type of the transmission line using a Bayesian model:

[0128]

[0129]

[0130] in, This represents the fault type of transmission line 'a', argmax indicates taking the maximum value, and h represents the fault hypothesis variable. This represents the set of risk score results for all components on the transmission line. This indicates the set of observed component risk score results. In the case of a fault type h in the transmission line, the posterior probability is... Indicates the prior probability of failure. This indicates the line number to which the i-th component belongs. The weighted likelihood term representing the risk of a component. This represents the compatibility coefficient matrix, i.e., the component category and fault category c of the i-th component. m The degree of support for line fault type h, This represents the set of transmission line fault types. The sign indicates proportionality.

[0131] Specifically, Bayesian models are probabilistic reasoning-based models used to fuse component risk information and infer the overall fault type of the line.

[0132] It should be noted that confidence level calibration and 3D projection improve the reliability of component location and identification results; the minimum distance matching method based on topological relationships ensures accurate component attribution. Introducing environmental and operating condition correction factors enables dynamic adaptability in risk assessment, reflecting the impact of weather, load, and aging on fault risk. Finally, using a Bayesian model to integrate multi-source information for global inference achieves probabilistic inference from local features to the overall fault type, significantly improving the accuracy, robustness, and interpretability of fault judgment, providing a scientific basis for intelligent early warning of transmission lines.

[0133] S6: Based on the fault type, perform early warning judgment and graded response.

[0134] Among them, the early warning determination is to assess the safety level of the line operation status based on the fault type and risk level, and determine whether an early warning signal needs to be issued.

[0135] Among them, the tiered response is to take different levels of response measures based on the warning level (such as red, orange, yellow), including on-site inspection, remote dispatch or emergency shutdown.

[0136] It should be noted that the fault type-driven early warning and response mechanism achieves closed-loop management of transmission lines from identification to decision-making. This step can automatically classify early warning levels according to the degree of risk, making the system real-time and targeted, avoiding delays caused by manual judgment. The tiered response mechanism allows power operation and maintenance departments to rationally allocate resources according to the risk level, achieving "early handling of minor faults and rapid response to major faults." At the same time, the early warning results can be linked with the intelligent monitoring platform to form a dynamic safety monitoring system, improving the safety and controllability of power grid operation.

[0137] In one possible implementation, the fault types include: three-phase ground fault, two-phase short-circuit fault, and single-phase ground fault.

[0138] Specifically, the warning levels can be divided into red, orange, and yellow.

[0139] Specifically, a red alert is activated when there is a three-phase grounding fault, an orange alert is activated when there is a two-phase short-circuit fault, and a yellow alert is activated when there is a single-phase grounding fault.

[0140] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0141] In this embodiment of the invention, by introducing a multi-stage component detection model and a spatial attention convolutional neural network, key components such as towers, conductors, and insulators can be accurately segmented in complex backgrounds, achieving multi-level feature extraction and enhancement of key areas. Through the fault saliency index and probability adjustment mechanism, false alarms and missed alarms are effectively reduced. Furthermore, by combining the spatial topology relationship of the transmission line and environmental operating parameters, a Bayesian model is used for risk fusion and fault type reasoning, realizing intelligent judgment from the component level to the line level. Finally, through the early warning classification mechanism, potential risks can be identified in advance, maintenance decisions can be optimized, and the operational safety and maintenance efficiency of the transmission line can be improved.

[0142] Reference manual attached Figure 2 The diagram shows a structural schematic of a power transmission line fault identification and early warning system based on an unmanned aerial vehicle (UAV) according to an embodiment of the present invention.

[0143] This invention provides a UAV-based power transmission line fault identification and early warning system 20, comprising: a processor 201 and a memory 202;

[0144] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-mentioned method for identifying and warning of power line faults based on UAVs and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0145] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0146] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM).

[0147] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0148] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0149] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0150] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0151] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0152] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0153] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0154] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0155] This invention provides a readable storage medium comprising: storing a program or instructions on the readable storage medium, wherein when the program or instructions are executed by a processor, the program or instructions implement the steps of the above-described method for identifying and warning of power line faults based on unmanned aerial vehicles, and can achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for power transmission line fault identification and early warning based on unmanned aerial vehicle, characterized in that, include: S1: Acquire raw image data containing various components of the transmission line; S2: Preprocess the original image data; S3: Using a multi-stage component detection model, the region of interest in the preprocessed original image data is extracted to obtain multiple component sub-images; S4: Using a spatial attention convolutional neural network model, perform fault identification on each component sub-image to determine the component-level fault category; S5: Based on the component-level fault categories, the fault type of the transmission line is determined by integrating the spatial topology of the transmission line and the risk characteristics of the components; S6: Based on the fault type, perform early warning determination and graded response; Specifically, S3 includes: S301: Input the preprocessed raw image data into the multi-stage component detection model; S302: Determine the tower boundary box image of the transmission line using the tower detector in the multi-stage component detection model; S303: Crop the tower boundary frame image to obtain a cropped boundary frame; S304: Based on the cropping bounding box, the component detector in the multi-stage component detection model outputs a component sub-image containing multiple component regions, wherein the component regions include the coordinate position, confidence level, and component category label of each component; S5 specifically includes: S501: Perform confidence calibration on the component-level fault category; S502: Based on the confidence calibration results, and based on the UAV attitude parameters and camera intrinsic and extrinsic parameters, project the component pixel coordinates in the multiple component sub-images onto the three-dimensional geographic coordinate system of the line; S503: Based on the three-dimensional geographic coordinate system of the route, the components are matched using the minimum distance matching method; S504: Based on the matching results, and combined with the spatial proximity relationships of towers, conductors and insulators, construct a topology map; S505: Based on the topology diagram, cross-domain environment and operating condition parameters are introduced to comprehensively correct component risks; S506: Based on the comprehensively corrected component risks, construct a set of line-level fault assumptions; S507: Based on the set of line-level fault assumptions, determine the fault type of the transmission line using a Bayesian model. 2.The UAV-based power transmission line fault identification and early warning method of claim 1, wherein, The spatial attention convolutional neural network model includes an input layer, a feature extraction layer, a feature concatenation layer, a flattening layer, a fully connected layer, and an output layer connected in sequence. S4 specifically includes: S401: The sub-images of each component are input to the spatial attention convolutional neural network model through the input layer; S402: Through the feature extraction layer, extract the deep features of each component sub-image and generate multiple output feature maps; S403: The output feature maps are spliced ​​together through the feature splicing layer to generate a spliced ​​feature map; S404: The spliced ​​feature map is expanded through the flattening layer to obtain a one-dimensional feature vector; S405: The one-dimensional feature vector is mapped through the fully connected layer to obtain multiple probability vectors, wherein the fully connected layer includes a first fully connected layer and a second fully connected layer; S406: The output layer is used to adjust each probability vector to determine the component-level fault category.

3. The method for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, Specifically, S402 includes: S4021: The component sub-image is processed by multi-channel convolution through the first convolution unit of the feature extraction layer to generate a first feature map; S4022: Calculate the attention weights between pixels in the first feature map through the first self-attention unit of the feature extraction layer; S4023: Based on the attention weights, the first feature map is reduced in dimensionality using the first max pooling unit of the feature extraction layer; S4024: The second convolutional unit of the feature extraction layer performs multi-channel convolution processing on the first feature map after dimensionality reduction to generate a second feature map; S4025: The second feature map is subjected to fault enhancement processing through the second self-attention unit of the feature extraction layer; S4026: The second feature map after fault enhancement is reduced in dimensionality by the minimum pooling unit of the feature extraction layer to generate a local minimum response feature map. S4027: Extract the high-level semantic features of the local minimum response feature map through the third convolutional unit of the feature extraction layer to obtain the high-level semantic feature map; S4028: The high-level semantic feature map is weighted and enhanced through the second self-attention unit of the feature extraction layer; S4029: The second max pooling unit of the feature extraction layer is used to perform dimensionality reduction on the processed high-level semantic feature map to generate multiple output feature maps.

4. The method for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs) according to claim 3, characterized in that, Specifically, S4022 includes: S4022A: Calculate the similarity between pixels in the first feature map using the first self-attention unit; S4022B: Normalize the similarity to obtain normalized weights; S4022C: Based on the normalized weights, the attention weights are determined through a weighted summation operation.

5. The method for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, Specifically, S405 includes: S4051: The one-dimensional feature vector is compressed into low-dimensional semantic features through the first fully connected layer; S4052: The low-dimensional semantic features are mapped to fault categories through the second fully connected layer to obtain multiple probability vectors.

6. The method for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, Specifically, S406 includes: S4061: Calculate the fault significance index; S4062: Adjust each of the probability vectors according to the fault significance index; S4063: Based on the adjusted probability vector, the fault classification result is determined through the output layer.

7. The method for power transmission line fault identification and early warning based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The fault types include: three-phase grounding fault, two-phase short-circuit fault, and single-phase grounding fault.

8. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the UAV-based power line fault identification and early warning method as described in any one of claims 1 to 7.

Images