A method, system, and medium for integrated intelligent inspection of power transmission, distribution, and transformation systems using unmanned aerial vehicles (UAVs) based on image processing algorithms.

Abstract

This invention discloses an intelligent inspection method, system, and medium for integrated power transmission and distribution equipment using unmanned aerial vehicles (UAVs) based on image processing algorithms. The method includes: constructing a gridded intelligent inspection network and configuring UAV nests integrating charging, weather monitoring, and 5G communication; a central control platform integrating power transmission and distribution inspection needs and dynamically optimizing task sequences; UAVs acquiring equipment images and generating optimized images through multi-dimensional feature fusion; using an improved MFB-Otsu algorithm combined with Gaussian process modeling to locate the optimal segmentation threshold; utilizing multi-scale deep convolutional neural networks and cross-disciplinary attention mechanisms to achieve intelligent defect identification, outputting a three-dimensional evaluation result including type, location, and severity; and implementing real-time edge processing and closed-loop optimization of the central platform through a two-level architecture. This invention achieves efficient collaborative inspection of power transmission and distribution equipment, improves the accuracy and efficiency of defect identification, and features adaptive optimization and cross-disciplinary integration.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for power equipment, specifically to an integrated intelligent inspection method, system, and medium for power transmission, transformation, and distribution using unmanned aerial vehicles (UAVs) based on image processing algorithms. It is particularly suitable for automated inspection of power grid transmission lines, substation equipment, and distribution lines. Background Technology

[0002] With the rapid development of power systems, the scale of transmission, substation, and distribution equipment is constantly expanding, making traditional manual inspection methods insufficient to meet the demands for efficient and accurate operation and maintenance. Unmanned aerial vehicle (UAV) inspection technology is gradually being adopted due to its flexibility and efficiency, but the following problems still exist in actual operation:

[0003] Insufficient multi-disciplinary collaboration: Existing inspection systems are typically designed for single power equipment (such as transmission lines or substations), lacking an integrated collaborative scheduling mechanism for transmission, transformation, and distribution, resulting in low resource utilization and limited inspection efficiency. Low image processing accuracy: In complex industrial environments, equipment images are subject to significant noise interference. Traditional segmentation algorithms (such as Otsu) rely on single grayscale features for threshold selection, making it difficult to adapt to changing scenarios and leading to decreased defect recognition accuracy. Insufficient real-time performance and intelligence: UAVs have limited edge computing capabilities, and existing methods largely rely on cloud processing, making real-time defect detection difficult. Furthermore, fixed model parameters cannot be dynamically optimized based on equipment type and environment, affecting inspection adaptability. Lack of closed-loop optimization: Most systems only complete defect detection without feeding back the recognition results to the scheduling and image processing stages, making it difficult to achieve adaptive improvement of inspection strategies.

[0004] To address the aforementioned issues, there is an urgent need for an integrated intelligent inspection method, system, and medium for power transmission, transformation, and distribution based on image processing algorithms, in order to improve the level of intelligent operation and maintenance of power equipment. Summary of the Invention

[0005] This invention proposes an intelligent inspection method, system, and medium for power transmission, transformation, and distribution equipment based on image processing algorithms. By integrating multi-dimensional feature perception with mathematical acceleration models, it significantly improves the image segmentation accuracy and computational efficiency of power equipment, thereby realizing intelligent joint inspection of power transmission, transformation, and distribution equipment.

[0006] In a first aspect, embodiments of this application provide an intelligent inspection method for integrated power transmission, distribution, and transformation of unmanned aerial vehicles (UAVs) based on image processing algorithms, the method comprising:

[0007] S1. Construct an intelligent inspection network by dividing the inspection area into grids and determining the size of each grid cell based on equipment distribution density and inspection frequency requirements. Configure an intelligent drone nest in each grid cell. The nest integrates an autonomous charging module, a meteorological monitoring unit, and a 5G communication module. Establish a real-time data connection between the nest and the central control platform through the communication module to form an intelligent inspection network covering the entire area.

[0008] S2. The central control platform receives and integrates inspection requirements from various disciplines, including power transmission, substation, and distribution; it adopts an intelligent scheduling algorithm based on equipment status assessment to dynamically calculate task priorities; and it optimizes and generates cross-disciplinary joint inspection task sequences based on the drone's endurance and grid coverage.

[0009] S3. Use an industrial camera mounted on a drone to capture images of power transmission and distribution equipment, and simultaneously perform multi-dimensional image feature fusion on the captured images to generate an optimized fused image;

[0010] S4. The optimized fused image is segmented based on the MFB-Otsu algorithm, and the optimal segmentation threshold is quickly located by Gaussian process modeling and expectation improvement function.

[0011] S5. Construct a deep convolutional neural network containing multi-scale convolutional kernel groups, and optimize feature extraction strategies for transmission lines, substation equipment, and power distribution equipment respectively through a cross-professional attention mechanism; perform multi-scale intelligent defect recognition on the segmented images through the deep convolutional neural network, and output a three-dimensional evaluation result containing defect type, location coordinates, and severity score;

[0012] S6. Optimize and adjust the recognition results, dynamically adjust the feature fusion weight coefficients in S3 according to the equipment type; optimize the segmentation parameters in S4 by combining real-time meteorological data; update the defect recognition model parameters in S5 through an online learning mechanism of historical segmentation results, and perform adaptive closed-loop optimization.

[0013] Among them, steps S3-S6 are implemented through a two-level architecture deployed on the edge computing nodes of the drone and the central control platform. The drone nodes are responsible for the real-time processing of S3-S5, while the central platform is responsible for the model optimization and result verification of S6, forming a complete intelligent inspection closed-loop system.

[0014] S5 includes: S5-1, constructing a multi-scale feature extraction network:

[0015] Parallel convolutional groups containing 3×3, 5×5, and 7×7 convolutional kernels are used to extract local detail features and global structural features of transmission lines, substation equipment, and power distribution equipment, respectively.

[0016] S5-2, Designing a cross-disciplinary attention module:

[0017] The feature weights of the three professional fields of power transmission, substation and distribution are dynamically adjusted by the channel attention mechanism to achieve adaptive fusion of professional features;

[0018] S5-3. Implement graded defect identification:

[0019] a) The primary network layer identifies the coordinates of the defect location and outputs a two-dimensional localization heatmap;

[0020] b) The intermediate network layer classifies the defect types and generates a classification probability distribution;

[0021] c) The high-level network layer assesses the severity of the defect and outputs a score of 1-5.

[0022] Optionally, in one implementation of the first aspect of the present invention, step S2 specifically includes:

[0023] S2-1: Receive inspection requests from various power transmission, substation, and distribution disciplines, and establish a multi-dimensional inspection request matrix based on equipment type, defect level, and operating data.

[0024] S2-2. An intelligent scheduling algorithm based on equipment status assessment is adopted, which combines real-time equipment health, historical failure probability and maintenance urgency to dynamically calculate the comprehensive priority weight of each inspection task.

[0025] S2-3. Based on the real-time endurance of the UAV, the distribution of charging stations, and the coverage of the gridded zones, construct a task allocation optimization model to generate the optimal cross-professional joint inspection task sequence with the goal of minimizing the total inspection time and maximizing resource utilization.

[0026] Optionally, in one implementation of the first aspect of the present invention, step S3, which involves synchronously performing multi-dimensional image feature fusion on the acquired image, includes:

[0027] Gray-level histogram equalization is used to extract gray-level features while preserving the global brightness distribution of the image.

[0028] The Sobel operator is applied to calculate gradient magnitude characteristics, thereby enhancing the edge contours of key parts of equipment insulators and conductor connections.

[0029] Extract LBP texture features and suppress background noise in complex industrial environments;

[0030] The three features are weighted and fused according to a weighting coefficient of 0.4:0.4:0.2 to generate an optimized fused image.

[0031] Optionally, in one implementation of the first aspect of the present invention, step S4, segmenting the optimized fused image based on the MFB-Otsu algorithm, and quickly locating the optimal segmentation threshold through Gaussian process modeling and expectation improvement function, includes:

[0032] S4-1. An improved Otsu algorithm with multi-feature fusion is adopted, which combines the gray-level distribution, edge gradient and texture features of the image to construct a multi-dimensional segmentation evaluation function;

[0033] S4-2. The segmentation threshold search space is probabilistically fitted by Gaussian process modeling, and the segmentation performance of candidate thresholds is dynamically evaluated using the expectation improvement function EI.

[0034] S4-3. Based on the Bayesian optimization strategy, the Gaussian process model is iteratively updated to quickly converge to the global optimal segmentation threshold, thereby achieving high-precision extraction of inspection targets.

[0035] Optionally, in one implementation of the first aspect of the present invention, step S4-1 employs an improved Otsu algorithm with multi-feature fusion, combining the image's grayscale distribution, edge gradient, and texture features to construct a multi-dimensional segmentation evaluation function, including:

[0036] S4-1-1. Calculating inter-class variance of grayscale based on recursive Otsu's algorithm: Dynamically updating the statistical parameters of foreground and background using a recursive formula.

[0037] ,

[0038] ,

[0039] in, To be at the threshold The pixel ratio of the background area. Represented as at the threshold The average gray level of the background region. grayscale The probability of a pixel appearing;

[0040] S4-1-2. Enhance segmentation boundaries by combining edge gradient features:

[0041] Calculate the gradient magnitude of the image using the Sobel operator. And construct the edge energy term:

[0042] ,

[0043] in Threshold The segmented set of foreground and background pixels. This represents the gradient magnitude of a pixel in an image. This represents the edge energy term, used to measure the sum of edge strengths at the dividing boundary;

[0044] S4-1-3. Introducing texture features to optimize segmentation robustness:

[0045] Calculating texture contrast features based on the gray-level co-occurrence matrix (GLCM):

[0046] ,

[0047] in, Threshold Below, gray levels in the Gray-Level Co-occurrence Matrix (GLCM) and The joint probability;

[0048] S4-1-4. Construct a multidimensional evaluation function and optimize the threshold:

[0049] In grayscale search range Internally, the optimal threshold is solved by fusing three types of features. :

[0050] ,

[0051] in, This represents the inter-class variance of traditional Otsu's algorithm, measuring the separation between foreground and background, with weighting coefficients. The contribution weights for grayscale, edge, and texture features are respectively satisfied. Furthermore, it dynamically adjusts based on the image signal-to-noise ratio. This represents the average grayscale value of the entire image, used to limit the threshold search range. The optimal threshold is determined by maximizing the fusion evaluation function. contrast(T) represents the texture contrast feature calculated based on the gray-level co-occurrence matrix under a given segmentation threshold T.

[0052] Optionally, in one implementation of the first aspect of the present invention, step S4-2, probabilistically fitting the segmentation threshold search space through Gaussian process modeling and dynamically evaluating the segmentation performance of candidate thresholds using the expectation improvement function EI, includes:

[0053] S4-2-1. Establish a Gaussian process proxy model for segmentation performance:

[0054] Using the threshold T as the input variable and the corresponding multidimensional evaluation function value F(T) as the output, a Gaussian process regression model is constructed:

[0055] ,

[0056] in, The value of the multidimensional evaluation function represents the threshold. The corresponding segmentation performance, Let be the mean function of a Gaussian process, representing the expression for the mean of a Gaussian process. The initial prediction For the covariance kernel function based on radial basis function (RBF), a threshold is used to measure the value. and The similarity between them This represents a Gaussian process model used to fit the nonlinear relationship between the threshold and segmentation performance. Represents the Gaussian process function in a Gaussian process model;

[0057] S4-2-2, Design an adaptive sampling strategy:

[0058] a) In the initial stage, uniformly sample N candidate thresholds. Calculate its evaluation function value ;

[0059] b) Update the hyperparameters of the Gaussian process based on the sampled data to obtain the posterior probability distribution;

[0060] S4-2-3, Using the expectation function to improve the function Guidance threshold:

[0061] Define the improvement function , indicating the current threshold Relative to the known optimal The improvement, among which This is the current optimal evaluation function value.

[0062] Expected improvement in computation :

[0063] ,

[0064] in, Used to select the most promising candidate threshold. and Predict the mean and standard deviation for a Gaussian process. To balance the parameters and control the weight of exploration and utilization, These are the cumulative distribution function (CDF) and probability density function (PDF) of the standard normal distribution, respectively.

[0065] S4-2-4, Iterative optimization until convergence:

[0066] a) Choose to make Maximum candidate threshold Conduct an actual assessment;

[0067] b) Update the Gaussian process model by adding it to the training set;

[0068] c) Repeat the above process until Less than the set threshold Or it may reach the maximum number of iterations.

[0069] Optionally, in one implementation of the first aspect of the present invention, step S4-3, iteratively updating the Gaussian process model based on a Bayesian optimization strategy to quickly converge to the globally optimal segmentation threshold and achieve high-precision extraction of the inspection target, includes:

[0070] S4-3-1. Construct an adaptive optimization framework: Establish a Bayesian optimization framework with image segmentation evaluation index as the objective function and threshold as the optimization variable, and establish the mapping relationship between threshold and segmentation performance through Gaussian process model;

[0071] S4-3-2, Implement intelligent iterative optimization: a) Generate a set of candidate thresholds based on the current model prediction; b) Select the most promising evaluation point by balancing the acquisition function of exploration and development; c) Dynamically adjust the search strategy, focusing on global exploration in the early stage and local fine-grained search in the later stage.

[0072] S4-3-3, Dynamic model update mechanism: After each evaluation of a new threshold, the parameters and hyperparameters of the Gaussian process model are updated in real time to gradually improve the model's fitting accuracy to the objective function;

[0073] S4-3-4. Set intelligent termination conditions: When the stability of the optimal solution and the search interval are fully converged at the same time, the optimization process will automatically terminate.

[0074] S4-3-5, Output the optimal segmentation scheme: Output the optimal segmentation threshold determined by Bayesian optimization to achieve accurate extraction of inspection targets.

[0075] Optionally, in one implementation of the first aspect of the present invention, step S5 further includes:

[0076] S5-4. Generate three-dimensional evaluation results:

[0077] The location coordinates, defect type, and severity score are integrated into structured data to form a three-dimensional output that includes spatial location, attribute information, and status assessment.

[0078] S5-5, Optimize network training strategies:

[0079] A multi-task joint loss function is adopted to balance the three indicators of positioning accuracy, classification accuracy, and scoring reliability.

[0080] In a second aspect, embodiments of this application provide an integrated intelligent inspection system for power transmission, distribution, and transformation of unmanned aerial vehicles (UAVs) based on image processing algorithms, applied to the integrated intelligent inspection method for power transmission, distribution, and transformation of unmanned aerial vehicles (UAVs) based on image processing algorithms as described in the first aspect. The system includes:

[0081] The intelligent inspection network module is used to construct an intelligent inspection network. It divides the inspection area into grids and determines the size of the grid cells based on the equipment distribution density and inspection frequency requirements. Each grid cell is equipped with an intelligent drone nest, which integrates an autonomous charging module, a meteorological monitoring unit, and a 5G communication module. The communication module establishes a real-time data connection between the drone nest and the central control platform, forming an intelligent inspection network covering the entire area.

[0082] The multi-disciplinary collaborative inspection and scheduling module is used by the central control platform to receive and integrate inspection requirements from various disciplines, including power transmission, substation, and distribution; it adopts an intelligent scheduling algorithm based on equipment status assessment to dynamically calculate task priorities; and it optimizes and generates cross-disciplinary joint inspection task sequences based on the drone's endurance and grid coverage.

[0083] The multi-dimensional feature fusion processing module is used to acquire images of power transmission and distribution equipment through an industrial camera mounted on a drone, and simultaneously perform multi-dimensional image feature fusion on the acquired images to generate an optimized fused image.

[0084] An improved image segmentation processing module is used to segment the optimized fused image based on the MFB-Otsu algorithm, and to quickly locate the optimal segmentation threshold by modeling Gaussian process and expectation improvement function;

[0085] The multi-scale defect recognition module is used to construct a deep convolutional neural network containing multi-scale convolutional kernel groups. Through a cross-professional attention mechanism, it optimizes feature extraction strategies for transmission lines, substation equipment, and power distribution equipment respectively. The segmented image is then subjected to multi-scale defect intelligent recognition through the deep convolutional neural network, and a three-dimensional evaluation result including defect type, location coordinates, and severity score is output.

[0086] The adaptive optimization processing module is used to perform optimization adjustments on the identification results, dynamically adjust the feature fusion weight coefficients in S3 according to the equipment type, optimize the segmentation parameters in S4 by combining real-time meteorological data, and update the defect identification model parameters in S5 through an online learning mechanism to perform adaptive closed-loop optimization.

[0087] Among them, steps S3-S6 are implemented through a two-level architecture deployed on the edge computing nodes of the drone and the central control platform. The drone nodes are responsible for the real-time processing of S3-S5, while the central platform is responsible for the model optimization and result verification of S6, forming a complete intelligent inspection closed-loop system.

[0088] The multi-scale defect identification module also includes: constructing a multi-scale feature extraction network: using parallel convolutional groups containing 3×3, 5×5, and 7×7 convolutional kernels to extract local detail features and global structural features of transmission lines, substation equipment, and distribution equipment, respectively; designing a cross-professional attention module: dynamically adjusting the feature weights of the three professional fields of power transmission, substation, and distribution through a channel attention mechanism to achieve adaptive fusion of professional features; and implementing hierarchical defect identification: a) the primary network layer identifies the defect location coordinates and outputs a two-dimensional location heatmap; b) the intermediate network layer classifies the defect type and generates a classification probability distribution; and c) the advanced network layer evaluates the severity of the defect and outputs a score of 1-5.

[0089] Thirdly, embodiments of this application provide an electronic device, including:

[0090] processor;

[0091] Memory used to store processor-executable instructions;

[0092] The processor is configured to implement the image processing algorithm-based intelligent inspection method for unmanned aerial vehicle (UAV) power transmission and distribution integration when executing the instructions.

[0093] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program that instructs a device to execute the image processing algorithm-based intelligent inspection method for unmanned aerial vehicle power transmission and distribution systems as described in the first aspect.

[0094] This invention discloses an intelligent inspection method and system for power transmission and distribution equipment using unmanned aerial vehicles (UAVs) based on image processing algorithms. The method includes: constructing a gridded intelligent inspection network and configuring UAV nests integrating charging, weather monitoring, and 5G communication; a central control platform integrating power transmission and distribution inspection needs and dynamically optimizing task sequences; UAVs acquiring equipment images and generating optimized images through multi-dimensional feature fusion; using an improved MFB-Otsu algorithm combined with Gaussian process modeling to locate the optimal segmentation threshold; utilizing multi-scale deep convolutional neural networks and cross-disciplinary attention mechanisms to achieve intelligent defect identification, outputting a three-dimensional evaluation result including type, location, and severity; and implementing real-time edge processing and closed-loop optimization of the central platform through a two-level architecture. This invention achieves efficient collaborative inspection of power transmission and distribution equipment, improves the accuracy and efficiency of defect identification, and features adaptive optimization and cross-disciplinary integration.

[0095] Beneficial effects:

[0096] 1. Multi-disciplinary collaborative and efficient inspection: By dividing the inspection area into grids and configuring intelligent drone nests, combined with the dynamic task scheduling algorithm of the central control platform, cross-disciplinary joint inspection of power transmission, substation and distribution equipment can be realized, which significantly improves resource utilization and operation and maintenance efficiency.

[0097] 2. High-precision image processing and defect recognition. Multi-dimensional feature fusion technology (grayscale, edge, texture) is used to optimize image quality. Combined with the improved MFB-Otsu algorithm and Gaussian process modeling, the optimal segmentation threshold is quickly located, improving the accuracy of target extraction in complex environments.

[0098] 3. Real-time processing and intelligent optimization. Through a two-tier architecture of UAV edge computing nodes and a central platform, real-time processing of image acquisition, segmentation, and recognition is achieved, reducing data transmission latency.

[0099] 4. Enhanced Adaptability and Robustness. By combining real-time meteorological data and equipment status feedback, the system dynamically optimizes the inspection task sequence and image processing parameters to ensure stability and reliability in changing environments.

[0100] 5. Reduced operation and maintenance costs. Automated inspections and intelligent analysis reduce manual intervention and shorten defect detection cycles. At the same time, the integration of cross-disciplinary tasks reduces the cost of repetitive inspections, providing efficient and economical technical support for power system operation and maintenance. Attached Figure Description

[0101] Figure 1 This is a schematic diagram of the process of an intelligent inspection method for unmanned aerial vehicle (UAV) power transmission, distribution and transformation based on image processing algorithms, provided in an embodiment of this application.

[0102] Figure 2 This is a structural diagram of an improved image segmentation processing module provided in an embodiment of this application.

[0103] Figure 3 This is an architecture diagram of an integrated intelligent inspection system for power transmission, distribution and transformation based on image processing algorithms provided in one embodiment of this application.

[0104] Figure 4 A schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0105] 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 a part of the embodiments of this application, and not all of them.

[0106] It should be noted that, in the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0107] It should be noted that in the embodiments of this application, the terms "first," "second," etc., are used only for descriptive purposes and should not be construed as indicating or implying relative importance, nor as indicating or implying order. Features specified as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0108] Based on the embodiments described in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0109] Example 1

[0110] The core of this invention lies in building an intelligent inspection network, which enables intelligent inspection and defect identification of power transmission and distribution equipment through the collaborative work of drones, edge computing nodes and a central control platform.

[0111] Figure 1 This is a schematic flowchart of an integrated intelligent inspection method for power transmission, distribution, and transformation using an image processing algorithm, provided in one embodiment of this application. Figure 1 As shown, the specific steps include:

[0112] S1. Construct an intelligent inspection network by dividing the inspection area into grids and determining the size of each grid cell based on equipment distribution density and inspection frequency requirements. Configure an intelligent drone nest in each grid cell. The nest integrates an autonomous charging module, a meteorological monitoring unit, and a 5G communication module. Establish a real-time data connection between the nest and the central control platform through the communication module to form an intelligent inspection network covering the entire area.

[0113] Building an intelligent inspection network is the core component of the entire intelligent inspection system. By dividing the inspection area into grids and determining the grid unit size based on equipment distribution density and inspection frequency requirements, refined management of the inspection area is achieved. Each grid unit is equipped with intelligent drone nests, which integrate autonomous charging modules, meteorological monitoring units, and 5G communication modules. These nests establish real-time data connections with the central control platform through the communication modules, forming an intelligent inspection network covering the entire area.

[0114] This involves utilizing dynamic gridding partitioning technology to develop a dynamic partitioning algorithm based on Voronoi diagrams. This algorithm automatically adjusts the grid shape according to equipment distribution density, establishing a grid parameter optimization model with the objective function: min(inspection blind zone) + max(resource balance). Constraints include equipment density gradient, terrain complexity, and spatial limitations. An improved K-means clustering algorithm is employed, using equipment criticality as the weight for region partitioning. Simultaneously, a grid self-healing mechanism is designed to dynamically adjust grid boundaries when equipment is added or removed.

[0115] Furthermore, according to the intelligent device nest configuration scheme, the modular device nest design includes: an energy module, such as a fast charging pile + battery swapping compartment, supporting 15-minute rapid power replenishment; a sensing module, such as a six-element weather station + lidar obstacle avoidance; and a communication module, such as a 5G private network + LoRa dual-channel redundant transmission.

[0116] Adaptive network topology management is used to construct a hierarchical network architecture. For example, based on function, it can be divided into: backbone layer: fixed nested nodes; mobile layer: inspection drone nodes; and emergency layer: mobile supply nodes.

[0117] It also includes using a real-time data fusion system to build a unified data platform that integrates: equipment ledger data, real-time telemetry data, meteorological and environmental data, and drone status data; and developing an edge-cloud collaborative processing framework, such as edge nodes responsible for data preprocessing and real-time alarms, and a central platform responsible for in-depth analysis and decision support.

[0118] This intelligent inspection network breaks through the limitations of traditional fixed deployment. Through dynamic grid division and intelligent nest configuration, it achieves precise coverage and efficient operation and maintenance of complex power equipment, providing a solid network foundation for integrated inspection.

[0119] Specifically, in this embodiment,

[0120] S2. The central control platform receives and integrates inspection requirements from various disciplines, including power transmission, substation, and distribution; it adopts an intelligent scheduling algorithm based on equipment status assessment to dynamically calculate task priorities; and it optimizes and generates cross-disciplinary joint inspection task sequences based on the drone's endurance and grid coverage.

[0121] Specifically, in this embodiment, step S2 includes:

[0122] S2-1: Receive inspection requests from various power transmission, substation, and distribution disciplines, and establish a multi-dimensional inspection request matrix based on equipment type, defect level, and operational data. Specifically, using multi-dimensional request integration technology, an inspection request matrix is ​​constructed to integrate the differentiated requirements of power transmission, substation, and distribution disciplines. Matrix dimensions include: equipment type (conductors / insulators / transformers, etc.), defect level (urgent / important / general), and operational data (temperature / load / historical failure rate). Tensor decomposition technology is used to extract the correlation features of the requirements from each discipline, achieving effective fusion of requirements.

[0123] S2-2. An intelligent scheduling algorithm based on equipment status assessment is adopted, combining real-time equipment health, historical failure probability, and maintenance urgency to dynamically calculate the comprehensive priority weight of each inspection task. Specifically, a dynamic priority assessment model can be set up, designing a three-layer assessment system, including: a) Basic layer: real-time equipment health (based on sensor data), b) Historical layer: equipment failure probability (based on maintenance big data analysis), c) Strategy layer: maintenance urgency (based on power grid operating conditions). Simultaneously, a fuzzy comprehensive evaluation algorithm is developed to quantify the influence weight of each factor through a membership function. A time decay factor can also be introduced to dynamically adjust the contribution of historical data.

[0124] S2-3. Based on the real-time endurance of the UAV, the distribution of charging stations, and the coverage of the gridded zones, construct a task allocation optimization model to generate the optimal cross-professional joint inspection task sequence with the goal of minimizing the total inspection time and maximizing resource utilization.

[0125] Specifically, a multi-objective optimization model is established using an intelligent optimization scheduling engine. The objective function is: min(total inspection time) + max(resource utilization). Constraints include: UAV endurance, charging station distribution, and weather restrictions. An improved NSGA-II algorithm is used to solve for the Pareto optimal solution set. An adaptive mutation operator is designed to enhance the algorithm's convergence in complex scenarios.

[0126] A simulation verification module based on digital twins was developed using a real-time dynamic adjustment mechanism to pre-evaluate the feasibility of the task sequence. Multiple feedback channels were set up: a) real-time monitoring of UAV status; b) early warning of sudden defects; c) early warning of weather changes. A rolling time-domain optimization strategy was adopted, replanning the task sequence every 15 minutes.

[0127] This intelligent scheduling method effectively solves the problems of uneven resource allocation, delayed response, and professional barriers in traditional inspections, providing core support for integrated inspection of power transmission, transformation, and distribution.

[0128] S3. Images of power transmission and distribution equipment are acquired using an industrial camera mounted on a drone. Multi-dimensional image feature fusion is then performed simultaneously on the acquired images to generate an optimized fused image. Specifically, in this embodiment, the image processing steps include grayscale histogram equalization, Sobel operator, LBP texture feature extraction, and feature fusion based on specific weight coefficients.

[0129] Specifically, the implementation steps include:

[0130] Gray-level histogram equalization is employed to extract gray-level features while preserving the global brightness distribution of the image. Gray-level histogram equalization is an image enhancement technique used to improve the contrast and brightness distribution of an image. By adjusting the gray-level value distribution, it makes the gray-level values ​​more uniform, thereby enhancing the visual effect of the image and the effectiveness of subsequent processing. In this method, gray-level histogram equalization calculates the gray-level histogram of the image and performs normalization processing to enhance the contrast and brightness distribution of the image.

[0131] The Sobel operator is applied to calculate gradient magnitude features, enhancing the edge contours of key areas such as equipment insulators and conductor connections. The Sobel operator is an edge detection operator that extracts edge information from an image by calculating the gradients in the horizontal and vertical directions. In this method, the Sobel operator is used to calculate gradient magnitude features to enhance the edge contours of key areas such as equipment insulators and conductor connections. The Sobel operator uses two 3×3 convolution kernels to calculate the gradients in the horizontal and vertical directions respectively, thereby extracting the edge information of the image.

[0132] Extracting LBP texture features to suppress background noise in complex industrial environments. LBP (Local Binary Pattern) is a texture feature extraction method that generates a binary pattern by comparing the grayscale values ​​of each pixel with its neighboring pixels to extract the texture features of an image. In this method, LBP texture features are used to suppress background noise in complex industrial environments.

[0133] LBP features extract texture features from an image by comparing the gray values ​​of the center pixel with those of its neighboring pixels to generate a binary pattern.

[0134] The three features are weighted and fused using a weighting coefficient of 0.4:0.4:0.2 to generate an optimized fused image. This weighted fusion method can combine the advantages of different features to improve the image processing effect.

[0135] S4 involves segmenting the optimized fused image using the MFB-Otsu algorithm (Multi-Feature Boosted Otsu algorithm), and quickly locating the optimal segmentation threshold through Gaussian process modeling and expectation improvement function.

[0136] Figure 2 This is a structural diagram of an improved image segmentation processing module provided in an embodiment of this application. Figure 2 As shown, the improved image segmentation processing module includes an input module, a core processing module, a dynamic optimization module, an image segmentation result output module, and a multi-scale defect intelligent recognition module. Specifically, S4 specifically includes:

[0137] S4-1. An improved Otsu algorithm with multi-feature fusion is adopted, which combines the gray-level distribution, edge gradient and texture features of the image to construct a multi-dimensional segmentation evaluation function.

[0138] Specifically, S4-1 employs an improved Otsu algorithm with multi-feature fusion, combining image grayscale distribution, edge gradient, and texture features to construct a multi-dimensional segmentation evaluation function, including:

[0139] S4-1-1. Calculating inter-class variance of grayscale based on recursive Otsu's algorithm: Dynamically updating the statistical parameters of foreground and background using a recursive formula.

[0140] ,

[0141] ,

[0142] in, To be at the threshold The pixel ratio of the background area. Represented as at the threshold The average gray level of the background region. grayscale The probability of a pixel appearing is calculated. This step involves the recursive Otsu's algorithm, which calculates the inter-class variance by dynamically updating the statistical parameters of the foreground and background. The core idea of ​​the Otsu's algorithm is to determine the optimal segmentation threshold by maximizing the inter-class variance.

[0143] S4-1-2. Enhance segmentation boundaries by combining edge gradient features:

[0144] Calculate the gradient magnitude of the image using the Sobel operator. And construct the edge energy term:

[0145] ,

[0146] in Threshold The segmented set of foreground and background pixels. This represents the gradient magnitude of a pixel in an image. The edge energy term measures the sum of edge intensities at the segmentation boundaries. This step calculates the image gradient magnitude using the Sobel operator and constructs the edge energy term to enhance the segmentation boundaries.

[0147] S4-1-3. Introducing texture features to optimize segmentation robustness:

[0148] Calculating texture contrast features based on the gray-level co-occurrence matrix (GLCM):

[0149] ,

[0150] in, Threshold Below, gray levels in the Gray-Level Co-occurrence Matrix (GLCM) and The joint probability; this step calculates texture contrast features through the gray-level co-occurrence matrix (GLCM) to enhance the robustness of segmentation.

[0151] S4-1-4. Construct a multidimensional evaluation function and optimize the threshold:

[0152] In grayscale search range Internally, the optimal threshold is solved by fusing three types of features. :

[0153] ,

[0154] in, This represents the inter-class variance of traditional Otsu's algorithm, measuring the separation between foreground and background, with weighting coefficients. The contribution weights for grayscale, edge, and texture features are respectively satisfied. Furthermore, it dynamically adjusts based on the image signal-to-noise ratio. This represents the average grayscale value of the entire image, used to limit the threshold search range. The optimal threshold is determined by maximizing the fusion evaluation function, where contrast(T) represents the texture contrast feature calculated based on the gray-level co-occurrence matrix at a given segmentation threshold T. This step constructs a multi-dimensional evaluation function by fusing gray-level, edge, and texture features, and determines the optimal segmentation result by optimizing the threshold.

[0155] The S4-1 method combines the Otsu algorithm, edge detection, texture features, and multi-feature fusion to construct a multi-dimensional segmentation evaluation function to optimize image segmentation results.

[0156] S4-2. The segmentation threshold search space is probabilistically fitted by Gaussian process modeling, and the segmentation performance of candidate thresholds is dynamically evaluated using the expectation improvement function EI.

[0157] Specifically, S4-2, which involves probabilistically fitting the segmentation threshold search space through Gaussian process modeling and dynamically evaluating the segmentation performance of candidate thresholds using the expectation improvement function EI, includes:

[0158] S4-2-1. Establish a Gaussian process proxy model for segmentation performance:

[0159] Using the threshold T as the input variable and the corresponding multidimensional evaluation function value F(T) as the output, a Gaussian process regression model is constructed:

[0160] ,

[0161] in, The value of the multidimensional evaluation function represents the threshold. The corresponding segmentation performance, Let be the mean function of a Gaussian process, representing the expression for the mean of a Gaussian process. The initial prediction For the covariance kernel function based on radial basis function (RBF), a threshold is used to measure the value. and The similarity between them This represents a Gaussian process model used to fit the nonlinear relationship between the threshold and segmentation performance. This represents the Gaussian process function in the Gaussian Process model; a Gaussian process regression model is constructed to fit the nonlinear relationship between the threshold and segmentation performance. The Gaussian process is a nonlinear modeling method widely used in machine learning and optimization problems.

[0162] S4-2-2, Design an adaptive sampling strategy. This step involves an adaptive sampling strategy, including initial sampling and updating the hyperparameters of the Gaussian process. The specific steps are as follows:

[0163] a) In the initial stage, uniformly sample N candidate thresholds. Calculate its evaluation function value b) Update the hyperparameters of the Gaussian process based on the sampled data to obtain the posterior probability distribution.

[0164] S4-2-3, Using the expectation function to improve the function Guiding threshold. The Expected Improvement (EI) function is used to guide the selection of candidate thresholds. The EI function is a sampling function used to select the next sampling point in Bayesian optimization. The specific steps are as follows:

[0165] Define the improvement function , indicating the current threshold Relative to the known optimal The improvement, among which This is the current optimal evaluation function value.

[0166] Expected improvement in computation :

[0167] ,

[0168] in, Used to select the most promising candidate threshold. and Predict the mean and standard deviation for a Gaussian process. To balance the parameters and control the weight of exploration and utilization, These are the cumulative distribution function (CDF) and probability density function (PDF) of the standard normal distribution, respectively.

[0169] S4-2-4. Iterative optimization until convergence. This involves continuously updating the Gaussian process model and selecting the optimal threshold until convergence. The specific steps are as follows:

[0170] a) Choose to make Maximum candidate threshold Conduct an actual assessment; b) will c) Update the Gaussian process model by adding the training set; d) Repeat the above process until... Less than the set threshold Or it may reach the maximum number of iterations.

[0171] The S4-2 method combines Gaussian process modeling, adaptive sampling strategy, expectation improvement function and iterative optimization to construct an efficient threshold search method.

[0172] S4-3. Based on the Bayesian optimization strategy, the Gaussian process model is iteratively updated to quickly converge to the global optimal segmentation threshold, thereby achieving high-precision extraction of inspection targets.

[0173] Specifically, S4-3, which iteratively updates the Gaussian process model based on a Bayesian optimization strategy to quickly converge to the globally optimal segmentation threshold and achieve high-precision extraction of inspection targets, includes:

[0174] S4-3-1. Constructing an Adaptive Optimization Framework: Establish a Bayesian optimization framework with the image segmentation evaluation index as the objective function and the threshold as the optimization variable. Establish the mapping relationship between the threshold and segmentation performance through a Gaussian process model. This step involves establishing a Bayesian optimization framework with the image segmentation evaluation index as the objective function and the threshold as the optimization variable, and establishing the mapping relationship between the threshold and segmentation performance through a Gaussian process model.

[0175] S4-3-2, Implement Intelligent Iterative Optimization: a) Generate a candidate threshold set based on the current model prediction; b) Select the most promising evaluation point through a data collection function that balances exploration and development; c) Dynamically adjust the search strategy, focusing on global exploration in the early stages and local fine-grained search in the later stages. This step includes generating candidate thresholds, selecting the most promising evaluation point, and dynamically adjusting the search strategy.

[0176] S4-3-3, Dynamic Model Update Mechanism: After each evaluation of a new threshold, the parameters and hyperparameters of the Gaussian process model are updated in real time to gradually improve the model's fitting accuracy to the objective function. This step involves updating the parameters and hyperparameters of the Gaussian process model in real time to improve the model's fitting accuracy to the objective function.

[0177] S4-3-4. Set intelligent termination conditions: The optimization process automatically terminates when both optimal solution stability and search interval convergence are simultaneously satisfied. This step involves setting termination conditions for optimal solution stability and search interval convergence.

[0178] S4-3-5, Output the optimal segmentation scheme: Output the optimal segmentation threshold determined by Bayesian optimization to achieve accurate extraction of inspection targets. This step involves outputting the optimal segmentation threshold determined by Bayesian optimization to achieve accurate extraction of inspection targets.

[0179] S5. Construct a deep convolutional neural network containing multi-scale convolutional kernel groups. Through a cross-disciplinary attention mechanism, optimize feature extraction strategies for transmission lines, substation equipment, and power distribution equipment respectively. Perform multi-scale intelligent defect recognition on the segmented images through the deep convolutional neural network and output a three-dimensional evaluation result containing defect type, location coordinates, and severity score.

[0180] Specifically, in this embodiment, S5 includes:

[0181] S5-1, Constructing a multi-scale feature extraction network:

[0182] Parallel convolutional groups containing 3×3, 5×5, and 7×7 kernels are used to extract local detail features and global structural features of transmission lines, substation equipment, and power distribution equipment, respectively. Multi-scale feature fusion plays an important role in image recognition and object detection, and can improve the model's feature representation capability.

[0183] S5-2, Designing a cross-disciplinary attention module:

[0184] The feature weights of the three professional fields of power transmission, substation, and distribution are dynamically adjusted through a channel attention mechanism to achieve adaptive fusion of professional features.

[0185] S5-3. Implement graded defect identification to classify and identify defect location, type, and severity:

[0186] a) The primary network layer identifies the coordinates of the defect location and outputs a two-dimensional localization heatmap;

[0187] b) The intermediate network layer classifies the defect types and generates a classification probability distribution;

[0188] c) The high-level network layer assesses the severity of the defect and outputs a score of 1-5.

[0189] S5-4. Generate 3D assessment results. Simultaneously, integrate the location, classification, and scoring results into structured data. Merge the location coordinates, defect type, and severity score into structured data to form a 3D output containing spatial location, attribute information, and status assessment.

[0190] S5-5, Optimize network training strategy. Employ a multi-task joint loss function to balance localization, classification, and scoring metrics. Specifically, use a multi-task joint loss function to balance localization accuracy, classification accuracy, and scoring reliability.

[0191] S6. Optimize and adjust the recognition results, dynamically adjusting the feature fusion weight coefficients in S3 according to the equipment type; optimize the segmentation parameters in S4 by combining real-time meteorological data; and update the defect recognition model parameters in S5 through an online learning mechanism to perform adaptive closed-loop optimization. This approach combines adaptive feature fusion, online learning, real-time data optimization, and closed-loop optimization, demonstrating high technical feasibility and application prospects.

[0192] For example, dynamic feature fusion optimization driven by equipment type can construct an equipment feature knowledge base and establish a multi-dimensional equipment feature matrix, including: structural features (conductor diameter / number of insulator discs, etc.), material features (metal / ceramic / composite materials), and failure mode features (cracks / corrosion / loosening, etc.). Equipment association analysis based on graph neural networks can be developed. Furthermore, an adaptive weight adjustment engine and a real-time weight adjustment algorithm can be used, with an adjustment frequency supporting dynamic weight adjustments up to 1000 times per second.

[0193] For image segmentation enhancement in meteorological sensing, a multimodal meteorological impact model can be established, along with a meteorological-image quality mapping matrix. The pipeline can then be optimized through real-time parameter settings. The matrix structure is shown in Table 1 below.

[0194] Table 1. Mapping and Compensation Strategies for Meteorological Sensing Parameters

[0195]

[0196] This solution, by deeply integrating equipment characteristics, environmental factors, and a continuous learning mechanism, constructs an intelligent inspection system with autonomous evolution capabilities, achieving a breakthrough improvement in adaptability, accuracy, and reliability compared to traditional methods.

[0197] Steps S3-S6 are implemented through a two-tier architecture deployed on the edge computing nodes of the drone and the central control platform. The drone nodes are responsible for the real-time processing of S3-S5, while the central platform is responsible for the model optimization and result verification of S6, forming a complete intelligent inspection closed-loop system.

[0198] Specifically, in this embodiment, the collaborative work scheme can be displayed using a function allocation matrix as shown in Table 2:

[0199] Table 2. Functional Coordination Table for Edge-Center Two-Tier Architecture

[0200]

[0201] Example 2

[0202] like Figure 3 As shown in the figure, this application provides an architecture diagram of an integrated intelligent inspection system for UAV power transmission and distribution based on image processing algorithms. It is applied to the integrated intelligent inspection system for UAV power transmission and distribution based on image processing algorithms as described in Embodiment 1. It includes an intelligent inspection network module 11, a multi-professional collaborative inspection scheduling module 12, a multi-dimensional feature fusion processing module 13, an improved image segmentation processing module 14, a multi-scale defect identification module 15, and an adaptive optimization processing module 16.

[0203] The intelligent inspection network module 11 is used to construct an intelligent inspection network, divide the inspection area into grids, and determine the size of the grid unit based on the equipment distribution density and inspection frequency requirements; an intelligent drone nest is configured in each grid unit, the nest integrating an autonomous charging module, a meteorological monitoring unit and a 5G communication module; a real-time data connection is established between the nest and the central control platform through the communication module to form an intelligent inspection network covering the entire area.

[0204] The multi-disciplinary collaborative inspection and scheduling module 12 is used by the central control platform to receive and integrate the inspection requirements of various disciplines such as power transmission, substation, and distribution; adopts an intelligent scheduling algorithm based on equipment status assessment to dynamically calculate task priorities; and optimizes and generates cross-disciplinary joint inspection task sequences based on the drone's endurance and grid coverage.

[0205] The multi-dimensional feature fusion processing module 13 is used to acquire images of power transmission and distribution equipment through an industrial camera mounted on a drone, and simultaneously perform multi-dimensional image feature fusion on the acquired images to generate an optimized fused image.

[0206] The improved image segmentation processing module 14 is used to segment the optimized fused image based on the MFB-Otsu algorithm, and quickly locates the optimal segmentation threshold by modeling the Gaussian process and the expectation improvement function.

[0207] The multi-scale defect recognition module 15 is used to construct a deep convolutional neural network containing multi-scale convolutional kernel groups. Through a cross-professional attention mechanism, it optimizes feature extraction strategies for transmission lines, substation equipment, and power distribution equipment respectively. The segmented image is then subjected to multi-scale intelligent defect recognition through the deep convolutional neural network, and a three-dimensional evaluation result including defect type, location coordinates, and severity score is output.

[0208] The adaptive optimization processing module 16 is used to perform optimization adjustments on the recognition results, dynamically adjust the feature fusion weight coefficients in S3 according to the equipment type, optimize the segmentation parameters in S4 by combining real-time meteorological data, and update the defect recognition model parameters in S5 through an online learning mechanism to perform adaptive closed-loop optimization.

[0209] Steps S3-S6 are implemented through a two-tier architecture deployed on the edge computing nodes of the drone and the central control platform. The drone nodes are responsible for the real-time processing of S3-S5, while the central platform is responsible for the model optimization and result verification of S6, forming a complete intelligent inspection closed-loop system.

[0210] Figure 4 This is an electronic device provided in one embodiment of this application. For example... Figure 4 As shown, the electronic device includes at least the following components: processor 101 and memory 100, communication interface 103, and bus 102.

[0211] In this embodiment of the application, memory 100 is used to store executable instructions of processor 101, which, when configured to execute instructions, implements the method as described in the first aspect.

[0212] In embodiments of this application, a computer-readable storage medium includes instructions that instruct a device to perform the method as described in the first aspect. For example, the instructions instruct the device to perform... Figure 1 The method is shown in the process steps.

[0213] In one embodiment of this application, the program operating in the electronic device may be a program that controls a central processing unit (CPU) or similar device to achieve the functions of the above-described embodiments of the present invention (a program that enables the computer to function). Information processed by these devices is then temporarily stored in random access memory (RAM) during processing, and subsequently stored in various ROMs such as read-only memory (FlashROM) and hard disk drives (HDDs), and read, corrected, and written by the CPU as needed.

[0214] It should be noted that a portion of the electronic device described above can also be implemented using a computer. In this case, the program for implementing the control function can be recorded on a computer-readable recording medium, and the program recorded on the recording medium can be read into the computer and executed.

[0215] It should be noted that the term "computer" as used here refers to a computer built into an electronic device, employing hardware including an operating system and peripheral devices. Furthermore, "computer-readable recording media" refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard drives built into a computer.

[0216] Furthermore, a "computer-readable recording medium" can include: a medium that dynamically stores a program for a short period of time, such as a communication line used when transmitting a program via a network such as the Internet or a communication line such as a telephone line; or a medium that stores a program for a fixed period of time, such as volatile memory inside a computer that serves as a server or client in this case. In addition, the aforementioned program can be a program used to implement the above-mentioned functions, or it can be a program that can implement the above-mentioned functions by combining with programs already recorded in the computer.

[0217] Furthermore, the electronic device in the above embodiments can also be implemented as an assembly (device group) composed of multiple devices. Each device constituting the device group can possess some or all of the functions or functional blocks of the electronic device in the above embodiments. As a device group, it is sufficient to have all the functions or functional blocks of the electronic device.

[0218] Those skilled in the art should recognize that the above embodiments are only used to illustrate this application and are not intended to limit this application. Any appropriate changes and variations made to the above embodiments within the essential spirit and scope of this application fall within the scope of protection claimed in this application.

Claims

1. A method for intelligent inspection of power transmission, distribution, and transformation systems using unmanned aerial vehicles (UAVs) based on image processing algorithms, characterized in that: The method includes: S1. Construct an intelligent inspection network by dividing the inspection area into grids and determining the size of each grid cell based on equipment distribution density and inspection frequency requirements. Configure an intelligent drone nest in each grid cell. The nest integrates an autonomous charging module, a meteorological monitoring unit, and a 5G communication module. Establish a real-time data connection between the nest and the central control platform through the communication module to form an intelligent inspection network covering the entire area. S2. The central control platform receives and integrates inspection requirements from various disciplines, including power transmission, substation, and distribution; it adopts an intelligent scheduling algorithm based on equipment status assessment to dynamically calculate task priorities; and it optimizes and generates cross-disciplinary joint inspection task sequences based on the drone's endurance and grid coverage. S3. Use an industrial camera mounted on a drone to capture images of power transmission and distribution equipment, and simultaneously perform multi-dimensional image feature fusion on the captured images to generate an optimized fused image; S4. The optimized fused image is segmented based on the MFB-Otsu algorithm, and the optimal segmentation threshold is quickly located by Gaussian process modeling and expectation improvement function. S5. Construct a deep convolutional neural network containing multi-scale convolutional kernel groups, and optimize feature extraction strategies for transmission lines, substation equipment, and power distribution equipment respectively through a cross-professional attention mechanism; perform multi-scale intelligent defect recognition on the segmented images through the deep convolutional neural network, and output a three-dimensional evaluation result containing defect type, location coordinates, and severity score; S6. Optimize and adjust the recognition results, dynamically adjust the feature fusion weight coefficients in S3 according to the equipment type; optimize the segmentation parameters in S4 by combining real-time meteorological data; update the defect recognition model parameters in S5 through an online learning mechanism, and perform adaptive closed-loop optimization. Among them, steps S3-S6 are implemented through a two-level architecture deployed on the edge computing nodes of the drone and the central control platform. The drone nodes are responsible for the real-time processing of S3-S5, while the central platform is responsible for the model optimization and result verification of S6, forming a complete intelligent inspection closed-loop system. S5 includes: S5-1, constructing a multi-scale feature extraction network: Parallel convolutional groups containing 3×3, 5×5, and 7×7 convolutional kernels are used to extract local detail features and global structural features of transmission lines, substation equipment, and power distribution equipment, respectively. S5-2, Designing a cross-disciplinary attention module: The feature weights of the three professional fields of power transmission, substation and distribution are dynamically adjusted by the channel attention mechanism to achieve adaptive fusion of professional features; S5-3. Implement graded defect identification: a) The primary network layer identifies the coordinates of the defect location and outputs a two-dimensional localization heatmap; b) The intermediate network layer classifies the defect types and generates a classification probability distribution; c) The high-level network layer assesses the severity of the defect and outputs a score of 1-5.

2. The intelligent inspection method for integrated power transmission, distribution, and transformation of unmanned aerial vehicles (UAVs) based on image processing algorithms according to claim 1, characterized in that, Step S2 specifically includes: S2-1: Receive inspection requests from various power transmission, substation, and distribution disciplines, and establish a multi-dimensional inspection request matrix based on equipment type, defect level, and operating data. S2-2. An intelligent scheduling algorithm based on equipment status assessment is adopted, which combines real-time equipment health, historical failure probability and maintenance urgency to dynamically calculate the comprehensive priority weight of each inspection task. S2-3. Based on the real-time endurance of the UAV, the distribution of charging stations, and the coverage of the gridded zones, construct a task allocation optimization model to generate the optimal cross-professional joint inspection task sequence with the goal of minimizing the total inspection time and maximizing resource utilization.

3. The intelligent inspection method for integrated power transmission, distribution, and transformation of unmanned aerial vehicles (UAVs) based on image processing algorithms according to claim 1, characterized in that, Step S3 involves synchronously performing multi-dimensional image feature fusion on the acquired images, including: Gray-level histogram equalization is used to extract gray-level features while preserving the global brightness distribution of the image. The Sobel operator is applied to calculate gradient magnitude characteristics, thereby enhancing the edge contours of key parts of equipment insulators and conductor connections. Extract LBP texture features and suppress background noise in complex industrial environments; The three features are weighted and fused according to a weighting coefficient of 0.4:0.4:0.2 to generate an optimized fused image.

4. The intelligent inspection method for integrated power transmission, distribution, and transformation of unmanned aerial vehicles based on image processing algorithms according to claim 2, characterized in that, S4 involves segmenting the optimized fused image based on the MFB-Otsu algorithm, and quickly locating the optimal segmentation threshold using Gaussian process modeling and the expectation improvement function, including: S4-1. An improved Otsu algorithm with multi-feature fusion is adopted, which combines the gray-level distribution, edge gradient and texture features of the image to construct a multi-dimensional segmentation evaluation function; S4-2. The segmentation threshold search space is probabilistically fitted by Gaussian process modeling, and the segmentation performance of candidate thresholds is dynamically evaluated using the expectation improvement function EI. S4-3. Based on the Bayesian optimization strategy, the Gaussian process model is iteratively updated to quickly converge to the global optimal segmentation threshold, thereby achieving high-precision extraction of inspection targets.

5. The intelligent inspection method for integrated power transmission, distribution, and transformation of unmanned aerial vehicles based on image processing algorithms according to claim 4, characterized in that, S4-1 employs an improved Otsu algorithm with multi-feature fusion, combining image grayscale distribution, edge gradient, and texture features to construct a multi-dimensional segmentation evaluation function, including: S4-1-1. Calculating inter-class variance of grayscale based on recursive Otsu's algorithm: Dynamically updating the statistical parameters of foreground and background using a recursive formula. ， ， in, To be at the threshold The pixel ratio of the background area. Represented as at the threshold The average gray level of the background region. grayscale The probability of a pixel appearing; S4-1-2. Enhance segmentation boundaries by combining edge gradient features: Calculate the gradient magnitude of the image using the Sobel operator. And construct the edge energy term: ， in Threshold The segmented set of foreground and background pixels. This represents the gradient magnitude of a pixel in an image. This represents the edge energy term, used to measure the sum of edge strengths at the dividing boundary; S4-1-3. Introducing texture features to optimize segmentation robustness: Calculating texture contrast features based on the gray-level co-occurrence matrix (GLCM): ， in, Threshold Below, gray levels in the Gray-Level Co-occurrence Matrix (GLCM) and The joint probability; S4-1-4. Construct a multidimensional evaluation function and optimize the threshold: In grayscale search range Internally, the optimal threshold is solved by fusing three types of features. : ， in, This represents the inter-class variance of traditional Otsu's algorithm, measuring the separation between foreground and background, with weighting coefficients. The contribution weights for grayscale, edge, and texture features are respectively satisfied. Furthermore, it dynamically adjusts based on the image signal-to-noise ratio. This represents the average grayscale value of the entire image, used to limit the threshold search range. The optimal threshold is determined by maximizing the fusion evaluation function. contrast(T) represents the texture contrast feature calculated based on the gray-level co-occurrence matrix under a given segmentation threshold T.

6. The intelligent inspection method for integrated power transmission, distribution, and transformation of unmanned aerial vehicles based on image processing algorithms according to claim 5, characterized in that, S4-2, which involves probabilistically fitting the segmentation threshold search space through Gaussian process modeling and dynamically evaluating the segmentation performance of candidate thresholds using the expectation improvement function EI, includes: S4-2-1. Establish a Gaussian process proxy model for segmentation performance: Using the threshold T as the input variable and the corresponding multidimensional evaluation function value F(T) as the output, a Gaussian process regression model is constructed: ， in, The value of the multidimensional evaluation function represents the threshold. The corresponding segmentation performance, Let be the mean function of a Gaussian process, representing the expression for the mean of a Gaussian process. The initial prediction For the covariance kernel function based on radial basis function (RBF), a threshold is used to measure the value. and The similarity between them This represents a Gaussian process model used to fit the nonlinear relationship between the threshold and segmentation performance. Represents the Gaussian process function in a Gaussian process model; S4-2-2, Design an adaptive sampling strategy: a) In the initial stage, uniformly sample N candidate thresholds. Calculate its evaluation function value ; b) Update the hyperparameters of the Gaussian process based on the sampled data to obtain the posterior probability distribution; S4-2-3, Using the expectation function to improve the function Guidance threshold: Define the improvement function , indicating the current threshold Relative to the known optimal The improvement, among which This is the current optimal evaluation function value. Expected improvement in computation : ， in, Used to select the most promising candidate threshold. and Predict the mean and standard deviation for a Gaussian process. To balance the parameters and control the weight of exploration and utilization, These are the cumulative distribution function (CDF) and probability density function (PDF) of the standard normal distribution, respectively. S4-2-4, Iterative optimization until convergence: a) Choose to make Maximum candidate threshold Conduct an actual assessment; b) Update the Gaussian process model by adding it to the training set; c) Repeat the above process until Less than the set threshold Or it may reach the maximum number of iterations.

7. The intelligent inspection method for integrated power transmission, distribution, and transformation of unmanned aerial vehicles based on image processing algorithms according to claim 6, characterized in that, S4-3, based on a Bayesian optimization strategy, iteratively updates the Gaussian process model, quickly converging to the globally optimal segmentation threshold to achieve high-precision extraction of inspection targets, including: S4-3-1. Construct an adaptive optimization framework: Establish a Bayesian optimization framework with image segmentation evaluation index as the objective function and threshold as the optimization variable, and establish the mapping relationship between threshold and segmentation performance through Gaussian process model; S4-3-2, Implement intelligent iterative optimization: a) Generate a set of candidate thresholds based on the current model prediction; b) Select the most promising evaluation point by balancing the acquisition function of exploration and development; c) Dynamically adjust the search strategy, focusing on global exploration in the early stage and local fine-grained search in the later stage. S4-3-3, Dynamic model update mechanism: After each evaluation of a new threshold, the parameters and hyperparameters of the Gaussian process model are updated in real time to gradually improve the model's fitting accuracy to the objective function; S4-3-4. Set intelligent termination conditions: When the stability of the optimal solution and the search interval are fully converged at the same time, the optimization process will be automatically terminated. S4-3-5, Output the optimal segmentation scheme: Output the optimal segmentation threshold determined by Bayesian optimization to achieve accurate extraction of inspection targets.

8. The intelligent inspection method for integrated power transmission, distribution, and transformation of unmanned aerial vehicles based on image processing algorithms according to claim 5, characterized in that, The S5 also includes: S5-4. Generate three-dimensional evaluation results: The location coordinates, defect type, and severity score are integrated into structured data to form a three-dimensional output that includes spatial location, attribute information, and status assessment. S5-5, Optimize network training strategies: A multi-task joint loss function is adopted to balance the three indicators of positioning accuracy, classification accuracy, and scoring reliability.

9. A UAV-based intelligent inspection system for power transmission, distribution, and transformation based on image processing algorithms, applied to the UAV-based intelligent inspection method for power transmission, distribution, and transformation based on image processing algorithms as described in any one of claims 1 to 8, characterized in that, The system includes: The intelligent inspection network module is used to construct an intelligent inspection network. It divides the inspection area into grids and determines the size of the grid cells based on the equipment distribution density and inspection frequency requirements. Each grid cell is equipped with an intelligent drone nest, which integrates an autonomous charging module, a meteorological monitoring unit, and a 5G communication module. The communication module establishes a real-time data connection between the drone nest and the central control platform, forming an intelligent inspection network covering the entire area. The multi-disciplinary collaborative inspection and scheduling module is used by the central control platform to receive and integrate inspection requirements from various disciplines, including power transmission, substation, and distribution; it adopts an intelligent scheduling algorithm based on equipment status assessment to dynamically calculate task priorities; and it optimizes and generates cross-disciplinary joint inspection task sequences based on the drone's endurance and grid coverage. The multi-dimensional feature fusion processing module is used to acquire images of power transmission and distribution equipment through an industrial camera mounted on a drone, and simultaneously perform multi-dimensional image feature fusion on the acquired images to generate an optimized fused image. An improved image segmentation processing module is used to segment the optimized fused image based on the MFB-Otsu algorithm, and to quickly locate the optimal segmentation threshold by modeling Gaussian process and expectation improvement function; The multi-scale defect recognition module is used to construct a deep convolutional neural network containing multi-scale convolutional kernel groups. Through a cross-professional attention mechanism, it optimizes feature extraction strategies for transmission lines, substation equipment, and power distribution equipment respectively. The segmented image is then subjected to multi-scale defect intelligent recognition through the deep convolutional neural network, and a three-dimensional evaluation result including defect type, location coordinates, and severity score is output. The adaptive optimization processing module is used to perform optimization adjustments on the identification results, dynamically adjust the feature fusion weight coefficients in S3 according to the equipment type, optimize the segmentation parameters in S4 by combining real-time meteorological data, and update the defect identification model parameters in S5 through an online learning mechanism to perform adaptive closed-loop optimization. Among them, steps S3-S6 are implemented through a two-level architecture deployed on the edge computing nodes of the drone and the central control platform. The drone nodes are responsible for the real-time processing of S3-S5, while the central platform is responsible for the model optimization and result verification of S6, forming a complete intelligent inspection closed-loop system. The multi-scale defect identification module also includes: constructing a multi-scale feature extraction network: using parallel convolutional groups containing 3×3, 5×5, and 7×7 convolutional kernels to extract local detail features and global structural features of transmission lines, substation equipment, and distribution equipment, respectively; designing a cross-professional attention module: dynamically adjusting the feature weights of the three professional fields of power transmission, substation, and distribution through a channel attention mechanism to achieve adaptive fusion of professional features; and implementing hierarchical defect identification: a) the primary network layer identifies the defect location coordinates and outputs a two-dimensional location heatmap; b) the intermediate network layer classifies the defect type and generates a classification probability distribution; and c) the advanced network layer evaluates the severity of the defect and outputs a score of 1-5 levels.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that instructs the device to execute the image processing algorithm-based intelligent inspection method for unmanned aerial vehicle (UAV) power transmission and distribution integration as described in any one of claims 1 to 8.

Images