A cable laying state unmanned aerial vehicle intelligent inspection method and system

Abstract

The application discloses a kind of cable laying state unmanned plane intelligent inspection method and system, comprising: receiving flight control instruction and multi-source sensor configuration parameter, it is perceived with task analysis generation inspection task context based on this, through adaptive planning algorithm, preset inspection target is decomposed into multiple acquisition subtasks Form structured acquisition scheme;Synchronous acquisition three-dimensional point cloud data, high-definition image sequence, infrared temperature field data and acoustic vibration signal data are sequentially executed each acquisition subtask, and initial inspection report is generated by space-time registration and multi-modal fusion analysis;Report is evaluated, if not passed, then backtrack execution link positioning analysis defect link and generate optimization acquisition instruction;According to instruction, update historical inspection case library or defect feature knowledge base;Re-execute until report passes evaluation, output final inspection report and risk warning information.The application realizes the closed-loop dynamic optimization of inspection task, improves the accuracy and reliability of defect detection.

Description

Technical Field

[0001] This invention relates to the field of power facility inspection technology, and in particular to an intelligent inspection method and system for cable laying status using unmanned aerial vehicles (UAVs). Background Technology

[0002] Regular inspections of cable laying conditions are fundamental to ensuring safe power transmission. Currently, inspection methods have evolved from traditional manual ground patrols to aerial photography using drones equipped with optical cameras. This method can quickly acquire image data of large areas of the cable network, but the information obtained is relatively limited, relying mainly on visual inspection. Its ability to identify non-visual defects such as cable sag, internal damage, and overheating joints is limited. With the development of sensor technology, some solutions have attempted to integrate multiple detection devices such as laser scanning and infrared thermography onto drones to obtain more comprehensive status information. However, in practical applications, intelligently planning flight paths and coordinating multiple sensors for efficient and collaborative data acquisition based on the specific inspection environment and targets remains a challenge. Furthermore, the fusion analysis and comprehensive diagnosis of multi-source heterogeneous data are also complex. Existing methods often rely on fixed analysis processes and lack the ability to dynamically feedback and self-optimize based on on-site data collection results. This leads to insufficient intelligence in the entire inspection process, making it difficult to effectively guarantee the accuracy and reliability of inspection results when facing complex corridor environments or hidden defects. Summary of the Invention

[0003] In view of this, the purpose of this invention is to propose an intelligent inspection method and system for cable laying status using unmanned aerial vehicles (UAVs). By constructing an intelligent inspection process with adaptive task planning and closed-loop feedback optimization, this invention solves the problem of insufficient detection accuracy and reliability caused by the lack of dynamic coordination and self-optimization capabilities in existing inspection methods.

[0004] To achieve the aforementioned technical objectives, in a first aspect, the technical solution adopted by the present invention is: a method for intelligent inspection of cable laying status using unmanned aerial vehicles, comprising: Receive flight control commands containing cable corridor geographic information and preset inspection targets, as well as associated multi-source sensor configuration parameters; Environmental perception and task analysis are performed on flight control commands and multi-source sensor configuration parameters to generate an inspection task context that includes terrain features, cable spatial location and a list of facilities to be inspected; Based on the inspection task context, the preset inspection target is decomposed into multiple collection sub-tasks with spatial order and sensor scheduling relationship through an adaptive planning algorithm, forming a structured collection scheme, and each collection sub-task is identified with its required data collection mode and positioning method. According to the structured acquisition scheme, each acquisition sub-task is executed in sequence to simultaneously acquire the three-dimensional point cloud data, high-definition image sequence, infrared temperature field data and acoustic vibration signal data of the cable corridor; Spatiotemporal registration and multimodal fusion analysis are performed on 3D point cloud data, high-definition image sequences, infrared temperature field data and acoustic vibration signal data to generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis and mechanical condition classification. The initial inspection report is evaluated for defect detection rate and false alarm rate. If the evaluation fails, the execution link of the structured data acquisition scheme is traced back to locate and analyze the defective link, and optimized data acquisition instructions are generated for analyzing the defective link. Based on the optimized data collection instructions, update the historical inspection case library used for task decomposition or the defect feature knowledge base used for multimodal fusion analysis; Based on the updated historical inspection case library or defect feature knowledge base, the steps from generating the inspection task context to generating the initial inspection report are re-executed until the initial inspection report passes the defect detection rate and false alarm rate assessment, and the final cable laying status inspection report and risk warning information are output.

[0005] In some embodiments, environmental perception and task parsing are performed on flight control commands and multi-source sensor configuration parameters to generate an inspection task context that includes terrain features, cable spatial locations, and a list of facilities to be inspected, including: Obtain raw point cloud data of the cable corridor area based on the geographic information of the cable corridor; The raw point cloud data is filtered and classified to separate the ground point cloud from the non-ground point cloud. Using ground point clouds, a digital elevation model of the cable corridor is generated through an interpolation algorithm. The digital elevation model represents the terrain features. Point cloud clusters belonging to the cable body are identified and extracted from non-ground point clouds using a density-based spatial clustering algorithm; The point cloud clusters of the cable body are fitted with three-dimensional spatial curves to calculate the continuous spatial position coordinate sequence of the cable in three-dimensional space. The spatial position of the cable is characterized by the continuous spatial position coordinate sequence. Based on the preset inspection targets, match the set of facility types associated with the geographical information of the cable corridor from the preset facility type template library; By combining the spatial location of cables with the set of facility types, spatial proximity analysis is used to locate the spatial anchor points of the facilities to be inspected in the scene composed of digital elevation model and non-ground point cloud, and a list of facilities to be inspected is formed. The digital elevation model, continuous spatial location coordinate sequence, and list of facilities to be inspected are integrated and associated, and encapsulated into an inspection task context.

[0006] In some embodiments, identifying and extracting point cloud clusters belonging to the cable body from non-ground point clouds using a density-based spatial clustering algorithm includes: Voxel mesh downsampling is performed on non-ground point clouds to reduce point cloud density while maintaining spatial structure features, thus generating downsampled non-ground point clouds. In the downsampled non-ground point cloud, calculate the average distance of several nearest neighbor points for each point; Based on the statistical distribution of the average distance of all points, determine the distance threshold and the minimum number of points threshold; The following method samples each unvisited point in the non-terrestrial point cloud as a seed point, and finds all points whose spatial distance from the seed point is less than a distance threshold. If the number of points found is greater than or equal to the minimum number of points threshold, these points are marked as a temporary point cloud cluster, and the seed point is marked as visited. Using each unvisited point in the temporary point cloud cluster as a new seed point, the search and marking process is iteratively executed until the temporary point cloud cluster no longer expands. At this point, the temporary point cloud cluster is marked as a final point cloud cluster. Traverse all unvisited points and repeat the process from selecting seed points to marking final point cloud clusters until all points have been visited, resulting in multiple final point cloud clusters. From multiple final point cloud clusters, select the final point cloud clusters whose geometric morphological features conform to the cable morphological features. The geometric morphological features include the ratio of the main direction length of the point cloud cluster to the cross-sectional diameter and the linearity of the point cloud cluster. The final point cloud clusters that conform to the cable morphological characteristics are defined as point cloud clusters belonging to the cable body.

[0007] In some embodiments, based on the inspection task context, an adaptive planning algorithm decomposes a preset inspection target into multiple data acquisition sub-tasks with spatial order and sensor scheduling relationships, forming a structured data acquisition scheme, including: Based on the cable spatial location contained in the inspection task context, an initial flight path covering the cable spatial location is generated through a path planning algorithm; Based on the terrain features and list of facilities to be inspected contained in the inspection mission context, collision detection and field of view analysis are performed on the initial flight trajectory. The initial flight trajectory is adjusted to avoid terrain obstacles and ensure that each facility to be inspected is within the effective observation range of the preset sensors, thereby generating an optimized flight trajectory. Based on the preset inspection targets and the type of each facility in the list of facilities to be inspected, the data acquisition mode set required for status detection of each facility is determined by querying the preset sensor-task mapping rule base. The data acquisition mode set includes one or more of visible light imaging, infrared thermal imaging, lidar scanning and acoustic and vibration signal acquisition. The optimized flight path is discretized into ordered path points, and each path point is associated with one or more facilities in the list of facilities to be inspected that are located in the preset neighborhood of that path point. For each track point and its associated facilities to be inspected, multiple sensor action commands to be executed at that track point are generated based on the data acquisition mode set. The sensor action commands include sensor type, acquisition parameters, acquisition duration and gimbal pointing angle. Based on the order of waypoints on the optimized flight path, the sensor action commands corresponding to all waypoints are sorted and combined to form multiple acquisition sub-tasks with spatial order and sensor scheduling relationship. Multiple data acquisition subtasks, optimized flight paths, and data acquisition modal sets are encapsulated in a structured manner to generate a structured data acquisition solution.

[0008] In some embodiments, and for each acquisition subtask, the required data acquisition mode and positioning method are identified, including: For each acquisition subtask in the structured acquisition scheme, extract the sensor action command corresponding to that acquisition subtask; Based on the sensor type contained in the sensor action command, the data acquisition mode required to perform the acquisition subtask is determined from the preset mode-capability mapping table. The data acquisition mode includes at least one of the following: three-dimensional geometric perception mode, visible light texture perception mode, infrared radiation perception mode, and acoustic vibration physical perception mode. Based on the spatial coordinates of the track point corresponding to the data collection subtask in the optimized flight path, and the terrain features contained in the inspection task context, the positioning environment category of the track point is determined. The positioning environment category includes open space positioning environment and satellite denial positioning environment. If the positioning environment category is open space positioning environment, then the positioning method of this acquisition sub-task will be identified as a fusion positioning method of global navigation satellite system positioning and inertial measurement unit positioning; If the positioning environment category is satellite denied positioning environment, then the positioning method of this acquisition sub-task will be identified as a fusion positioning method of visual inertial odometry positioning and ultra-wideband anchor point positioning. The identified data acquisition mode and positioning method are associated with metadata and written into the corresponding acquisition subtask.

[0009] In some embodiments, the simultaneous acquisition of three-dimensional point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data of the cable corridor includes: Upon arrival at each waypoint, execute the data collection sub-task associated with that waypoint; When the acquisition subtask includes a lidar scanning command, the environment around the current track point is scanned according to the preset scanning frequency and angle range to obtain the original lidar point cloud frame. The original lidar point cloud frames obtained from all track points are then timestamped and their coordinate systems are transformed and fused to generate three-dimensional point cloud data covering the cable corridor. When the acquisition subtask includes a visible light imaging instruction, one or more visible light images are acquired at the gimbal pointing angle specified by the acquisition subtask, and the acquired visible light images are arranged into a high-definition image sequence according to the track point order and timestamp. When the acquisition subtask includes an infrared thermal imaging command, a single or multiple frames of infrared thermal images are acquired at the gimbal pointing angle specified in the acquisition subtask, and the acquired infrared thermal images are arranged into infrared temperature field data according to the track point order and timestamp. When the acquisition subtask includes an instruction to acquire acoustic and vibration signals, the original audio signal and the original vibration signal are acquired within the acquisition duration specified by the acquisition subtask, and the original audio signal and the original vibration signal are subjected to noise reduction and filtering preprocessing to generate acoustic and vibration signal data. During the execution of the data acquisition sub-task at each track point, the real-time positioning data and attitude data output by the positioning sensor corresponding to the positioning method identified by the data acquisition sub-task are recorded synchronously. The real-time positioning data and attitude data are used as spatiotemporal references and are synchronized and spatially correlated with 3D point cloud data, high-definition image sequences, infrared temperature field data and acoustic vibration signal data, respectively.

[0010] In some embodiments, spatiotemporal registration and multimodal fusion analysis are performed on 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data to generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis, and mechanical condition classification, including: Based on the real-time positioning and attitude data associated with 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data, each frame of the high-definition image sequence, each frame of the infrared temperature field data, and each segment of the acoustic vibration signal data are uniformly registered to the world coordinate system defined by the 3D point cloud data through coordinate transformation and time interpolation algorithms. The mapping relationship between image pixels, heat map pixels, signal segments and 3D point cloud spatial points is established to complete the spatiotemporal registration. From the spatiotemporally registered 3D point cloud data, extract the cable point cloud segments that characterize the cable body; The catenary equation parameters of the cable are obtained by fitting a three-dimensional spatial curve to the point cloud segment of the cable. Based on the catenary equation parameters and the digital elevation model contained in the inspection task context, the vertical distance between the lowest point of the cable sag and the terrain surface below is calculated, and the cable sag safety distance value is generated. Along the cable route, the safe distance values ​​of multiple sampling points are calculated at preset intervals to form a cable sag safe distance distribution curve. The cable sag safety assessment consists of the cable sag safe distance values ​​and the cable sag safe distance distribution curve. From a high-resolution image sequence that has been spatiotemporally registered, the surface appearance features of cables and their ancillary facilities are identified by a target detection algorithm. These surface appearance features include insulator contamination, metal component corrosion, foreign objects hanging on conductors, and signs of loose connectors. From the spatiotemporally registered infrared temperature field data, the local temperature distribution corresponding to the spatial location of surface appearance features is extracted, and the local temperature rise, temperature gradient and hot spot characteristics are calculated. The surface appearance features are associated with the corresponding local temperature distribution. The matching reasoning is performed through a pre-set appearance-temperature association rule base to determine whether there is a temperature anomaly pattern induced or accompanied by the appearance anomaly, and to generate a latent defect diagnosis conclusion. The latent defect diagnosis conclusion includes the defect type, location and risk level. From the spatiotemporally registered acoustic and vibration signal data, the Mel frequency cepstral coefficient features of the audio signal and the spectral features of the vibration signal are extracted. The Mel frequency cepstral coefficient features and spectral features are input into the pre-trained acoustic vibration classification model. The acoustic vibration classification model outputs mechanical state classification labels for cables and their accessories. The mechanical state classification labels include normal tightness, bolt looseness, corona discharge, and vortex-induced vibration. The cable sag safety assessment, latent defect diagnosis conclusions, and mechanical condition classification labels are integrated and packaged into a structured initial inspection report.

[0011] In some embodiments, the initial inspection report is evaluated for defect detection rate and false alarm rate, including: The latent defect diagnosis conclusions and mechanical condition classification labels contained in the initial inspection report are combined and defined as the defect set automatically identified by the algorithm. From the preset cable defect standard library, retrieve the set of known standard defects that match the cable corridor geographic information and preset inspection targets corresponding to the current inspection task. The set of known standard defects includes the defect type, standard location and standard severity. The algorithm automatically identifies a set of defects and compares their spatial location and type consistency with a set of known standard defects. The defect detection rate is obtained by dividing the number of defects in the defect set automatically identified by the calculation algorithm that match the known standard defect set in both spatial location and defect type by the total number of defects in the known standard defect set. The false alarm rate is calculated by dividing the number of defects in the set of defects automatically identified by the algorithm that cannot be matched with the known standard defect set in terms of spatial location or defect type by the total number of defects in the set of defects automatically identified by the algorithm. The defect detection rate is compared with the preset detection rate threshold, and the false alarm rate is compared with the preset false alarm rate threshold. If the defect detection rate is lower than the detection rate threshold, or the defect false alarm rate is higher than the false alarm rate threshold, the initial inspection report is deemed to have failed the defect detection rate and false alarm rate assessment. If the defect detection rate is not lower than the detection rate threshold and the false alarm rate is not higher than the false alarm rate threshold, then the initial inspection report is deemed to have passed the defect detection rate and false alarm rate assessment.

[0012] In some embodiments, the execution chain of the structured acquisition scheme is traced back to locate the defective link in the analysis, and optimized acquisition instructions for the defective link are generated, including: Based on the comparison between the defect set automatically identified by the algorithm and the known standard defect set, missed defects and false alarm defects are identified. Missed defects refer to defects that exist in the known standard defect set but are not identified in the defect set automatically identified by the algorithm; False alarm defects refer to defects that exist in the set of defects automatically identified by the algorithm but cannot be matched with the known standard set of defects; For each missed defect, based on the standard location of the missed defect recorded in the known standard defect set, the nearest track point to the standard location is found in the optimized flight track of the structured acquisition scheme. In the data acquisition sub-task associated with this track point, analyze whether the data acquisition mode specified by the sensor action command covers the sensing capability required to identify the missed defect; If not covered, the analysis defect corresponding to the missed detection defect is determined to be a missing sensor configuration step; If it has been covered, the quality of the raw data collected by the track point will be further checked. If the raw data has problems such as blurriness, occlusion or low signal-to-noise ratio, the analysis defect corresponding to the missed defect will be determined to be the data quality deficiency link. For each false alarm defect, based on the identification location of the false alarm defect recorded in the defect set automatically identified by the algorithm, the nearest track point to the identification location is found in the optimized flight track of the structured acquisition scheme. In the data acquisition subtask associated with the waypoint, analyze whether the data acquisition parameters specified by the sensor action command are within a range that is prone to interference, or whether there are known sources of interference in the environment where the waypoint is located. If there are improper parameters or environmental interference, the analysis defect corresponding to the false alarm defect is determined to be the interference introduction stage; The analysis defect links corresponding to all missed defects and all false alarm defects are summarized to form an analysis defect link set. Based on the type of each analytical defect in the set of analytical defect links, the corresponding optimization operation is matched from the pre-set optimization strategy library, including: Add or adjust the data acquisition mode to address missing components in the sensor configuration; Adjust flight paths, gimbal angles, or sensor parameters to improve data quality in areas with insufficient data quality. To mitigate interference, adjust data acquisition parameters or add filtering preprocessing steps. All matched optimization operations are converted into optimized acquisition instructions that can be parsed and executed by the structured acquisition scheme.

[0013] In a second aspect, the present invention also provides an intelligent UAV inspection system for cable laying status, applicable to the method described in the first aspect. The system includes a task instruction receiving module, an environmental perception and analysis module, a task planning and scheme generation module, a data synchronization acquisition module, a multimodal data analysis module, a report quality assessment module, a link backtracking and optimization instruction generation module, a knowledge base update module, and an iterative execution and output module. The task instruction receiving module receives flight control instructions containing cable corridor geographic information and preset inspection targets, along with associated multi-source sensor configuration parameters. The environmental perception and analysis module performs environmental perception and task analysis on the flight control instructions and multi-source sensor configuration parameters, generating an inspection task context containing terrain features, cable spatial location, and a list of facilities to be inspected. The task planning and scheme generation module, based on the inspection task context, decomposes the preset inspection targets into multiple acquisition sub-tasks with spatial order and sensor scheduling relationships using an adaptive planning algorithm, forming a structured acquisition scheme, and identifying the required data acquisition modality and positioning method for each acquisition sub-task. The data synchronization acquisition module executes each acquisition sub-task sequentially according to the structured acquisition scheme, synchronously acquiring data. The system acquires 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic and vibration signal data of the cable corridor. A multimodal data analysis module performs spatiotemporal registration and multimodal fusion analysis on the 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic and vibration signal data to generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis, and mechanical condition classification. A report quality assessment module evaluates the defect detection rate and false alarm rate of the initial inspection report. A link backtracking and optimization instruction generation module is used to backtrack the results when the initial inspection report fails the defect detection rate and false alarm rate assessment. The structured acquisition scheme execution chain is used to locate and analyze defective links and generate optimized acquisition instructions for analyzing defective links; the knowledge base update module is used to update the historical inspection case library for task decomposition or the defect feature knowledge base for multimodal fusion analysis based on the optimized acquisition instructions; the iterative execution and output module is used to re-execute the steps from generating the inspection task context to generating the initial inspection report based on the updated historical inspection case library or defect feature knowledge base, until the initial inspection report passes the defect detection rate and false alarm rate evaluation, and outputs the final cable laying status inspection report and risk warning information.

[0014] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: By receiving flight control commands and multi-source sensor configuration parameters, it performs environmental perception and task analysis to generate an inspection task context. Based on this, an adaptive planning algorithm decomposes the preset inspection target into multiple acquisition sub-tasks to form a structured acquisition scheme. Each acquisition sub-task is executed sequentially to simultaneously acquire 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data. Spatiotemporal registration and multimodal fusion analysis are performed to generate an initial inspection report. The report is evaluated; if it fails, the process backtracks to analyze the defective link and generates optimized acquisition commands. The historical inspection case library or defect feature knowledge base is updated according to the commands. The process is repeated until the report passes evaluation, and the final inspection report and risk warning information are output. This invention achieves closed-loop dynamic optimization of the inspection task, improving the accuracy and reliability of defect detection. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of steps S101 to S108 of the inspection method described in the specific implementation embodiment; Figure 2 This is a schematic diagram of steps S201 to S208 of the inspection method described in the specific implementation. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 In a first aspect, this embodiment provides a method for intelligent inspection of cable laying status using unmanned aerial vehicles (UAVs), including: S101, Receive flight control commands containing cable corridor geographic information and preset inspection targets, as well as associated multi-source sensor configuration parameters; S102. Perform environmental perception and task analysis on flight control commands and multi-source sensor configuration parameters to generate an inspection task context that includes terrain features, cable spatial location and a list of facilities to be inspected. S103. Based on the inspection task context, the preset inspection target is decomposed into multiple acquisition sub-tasks with spatial order and sensor scheduling relationship through an adaptive planning algorithm to form a structured acquisition scheme, and the required data acquisition mode and positioning method are identified for each acquisition sub-task. S104. Execute each acquisition sub-task in sequence according to the structured acquisition scheme, and simultaneously acquire the three-dimensional point cloud data, high-definition image sequence, infrared temperature field data and acoustic vibration signal data of the cable corridor. S105. Perform spatiotemporal registration and multimodal fusion analysis on three-dimensional point cloud data, high-definition image sequences, infrared temperature field data and acoustic vibration signal data to generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis and mechanical condition classification. S106. Evaluate the defect detection rate and false alarm rate of the initial inspection report. If the evaluation fails, trace back the execution link of the structured acquisition scheme to locate and analyze the defect link, and generate optimized acquisition instructions for analyzing the defect link. S107. Based on the optimized data collection instructions, update the historical inspection case library used for task decomposition or the defect feature knowledge base used for multimodal fusion analysis. S108. Based on the updated historical inspection case library or defect feature knowledge base, re-execute the steps from generating the inspection task context to generating the initial inspection report until the initial inspection report passes the defect detection rate and false alarm rate assessment, and output the final cable laying status inspection report and risk warning information.

[0019] In step S101, the cable corridor geographic information included in the flight control commands comes from a geographic information system database or preliminary survey results, defining the spatial range in the form of coordinate boundaries or polylines. The preset inspection targets are clearly defined as specific inspection requirements, such as full-line sag measurement or temperature surveys of specific facilities. The associated multi-source sensor configuration parameters are a set of initial operating parameters set for lidar, visible light cameras, infrared thermal imagers, and acoustic vibration sensors, including scanning mode, resolution, range, and sampling rate.

[0020] In step S102, the environmental perception and task parsing process integrates geographic information and inspection targets to generate a structured inspection task context. The terrain features are initially described as an overview of the region's elevation distribution. Initial information on cable spatial locations can be obtained from two-dimensional line data in the design drawings. The list of facilities to be inspected is a list of specific types of facility identifiers and their approximate locations within the current corridor, matched from the line asset database based on the inspection targets. This step outputs a digital task blueprint that includes environmental constraints, target objects, and spatial references.

[0021] In step S103, the spatial order is determined by path planning to ensure flight efficiency; sensor scheduling coordinates the start-up, shutdown, and parameter switching of different sensors between waypoints. For each sub-task, the required combination of physical signal acquisition capabilities is specified, such as a combination of geometric sensing and thermal radiation sensing; the positioning method is pre-specified based on the expected environment (open ground or obstructed area) of the sub-task execution point, such as fusing satellite navigation and visual positioning. The resulting structured acquisition scheme is an executable script containing a waypoint sequence, sensor action commands for each waypoint, and global constraints. This step achieves the automatic generation from task requirements to a refined, executable operation sequence.

[0022] In step S104, the UAV autonomously navigates to various waypoints according to the structured data acquisition scheme, and the payload management system synchronously triggers preset sensor actions. Three-dimensional point cloud data is acquired through lidar scanning, forming a dense set of points describing the surface geometry. High-resolution image sequences are obtained by capturing images with a visible light camera at a specified gimbal orientation. Infrared temperature field data is acquired through an infrared thermal imager, reflecting the surface temperature distribution. Acoustic and vibration signal data are recorded through a microphone and vibration sensor, including audio and vibration timing signals. All data are embedded with a unified timestamp and pose information provided by the positioning module that identifies the positioning method.

[0023] In step S105, spatiotemporal registration utilizes the timestamps and pose information inherent in each data point to unify all data into a common world coordinate system through coordinate transformation. Multimodal fusion analysis is then performed: cable sag safety assessment calculates the safe distance to ground by analyzing the three-dimensional curve of the cable extracted from the point cloud; latent defect diagnosis infers the relationship between high-definition image texture anomalies (such as dirt or damage) and local temperature rise patterns in the infrared temperature field at the same location; mechanical condition classification is achieved by extracting features from acoustic and vibration signals and inputting them into a classification model. The generated initial inspection report is a structured document summarizing the above assessment results (numerical values, conclusions, labels, and locations). This step achieves a comprehensive assessment of the cable condition through cross-modal information correlation and complementarity.

[0024] In step S106, the defect detection rate and false alarm rate assessment requires comparing the identification results in the initial inspection report with a reliable benchmark. This benchmark can be a manually verified defect list, a historically verified defect database, or authoritative results obtained through higher-precision detection methods. The assessment process calculates the proportion of correctly identified defects (detection rate) and the proportion of defects misidentified in the identification results (false alarm rate) through spatial location matching and type comparison. If the assessment results do not reach the preset performance threshold (which can be set according to operation and maintenance standards), the execution chain of the structured acquisition scheme is traced back. By analyzing missed detections or false alarms, the entire process of generating corresponding acquisition sub-tasks, data acquisition, and data analysis is traced in reverse to locate the specific link causing the problem, i.e., the defect analysis link. The optimized acquisition instructions generated for this link may include adjusting the altitude of specific waypoints, modifying sensor parameters, or adding supplementary observations. This step realizes the quantitative monitoring of the quality of inspection results and the automatic diagnosis of the root cause of problems.

[0025] In step S107, the historical inspection case library stores the evaluation context of inspection tasks, the corresponding structured data collection schemes, and evaluation reports, forming a mapping of "task conditions - execution scheme - good results." The defect feature knowledge base stores the verified feature patterns and diagnostic rules of defects in multimodal data. The knowledge base is updated according to the optimized data collection instructions, that is, the diagnosed problems and their solutions are transformed into structured knowledge, such as storing optimized data collection parameters as new cases, or updating the feature library with more accurate identification features.

[0026] In step S108, the updated historical inspection case library optimizes adaptive planning and generates a better data collection scheme; the updated defect feature knowledge base enhances multimodal fusion analysis and generates a more accurate initial report. This iterative process is repeated cyclically, using previous experience to improve the current performance until the initial report passes evaluation. The final output inspection report and risk warning information are high-confidence results verified through closed-loop optimization. This step realizes the leap of the inspection system from fixed program execution to having online learning and adaptive evolution capabilities.

[0027] This embodiment constructs a complete intelligent inspection closed loop encompassing perception, decision-making, execution, evaluation, and optimization. By introducing a closed-loop feedback mechanism based on quantitative evaluation, the inspection results are transformed into a basis for driving task planning and data analysis model optimization. This enables the system to adapt to complex environments and changing objectives, and to automatically approach the optimal strategy through iteration. Consequently, it significantly improves the intelligence level of inspection operations, the accuracy of defect detection, and the autonomy of long-term system operation and maintenance.

[0028] Please see Figure 2In some embodiments, environmental perception and task parsing are performed on flight control commands and multi-source sensor configuration parameters to generate an inspection task context that includes terrain features, cable spatial locations, and a list of facilities to be inspected, including: S201. Obtain raw point cloud data of the cable corridor area based on the geographic information of the cable corridor; S202. Perform ground point filtering and classification processing on the raw point cloud data to separate the ground point cloud from the non-ground point cloud. S203. Using ground point clouds, a digital elevation model of the cable corridor is generated through an interpolation algorithm. The digital elevation model represents the terrain features. S204. From the non-ground point cloud, identify and extract point cloud clusters belonging to the cable body using a density-based spatial clustering algorithm; S205. Perform three-dimensional spatial curve fitting on the point cloud cluster of the cable body to calculate the continuous spatial position coordinate sequence of the cable in three-dimensional space. The spatial position of the cable is represented by the continuous spatial position coordinate sequence. S206. Based on the preset inspection targets, match the set of facility types associated with the cable corridor geographic information from the preset facility type template library; S207. Combining the spatial location of cables with the set of facility types, spatial proximity analysis is used to locate the spatial anchor points of the facilities to be inspected in the scene composed of digital elevation model and non-ground point cloud, and a list of facilities to be inspected is formed. S208. Integrate and associate the digital elevation model, continuous spatial location coordinate sequence, and list of facilities to be inspected, and encapsulate them into an inspection task context.

[0029] In step S201, the raw point cloud data of the cable corridor area can be obtained by actively scanning the inspection area using a LiDAR scanner mounted on a UAV platform, or by importing a point cloud dataset pre-acquired through aerial photogrammetry, ground laser scanning, or other methods. This data contains a large number of points with three-dimensional coordinates, representing the surface geometry of the ground, vegetation, buildings, cables, and other facilities within the corridor.

[0030] In step S202, the raw point cloud data undergoes ground point filtering and classification to distinguish points representing the ground from points representing objects above the ground. Ground point filtering methods can employ algorithms based on local slope variations or mathematical morphology operations. Through this processing, the raw point cloud is separated into ground point clouds and non-ground point clouds. The ground point clouds are primarily used to construct terrain models, while the non-ground point clouds contain targets requiring further analysis, such as cables, towers, and vegetation.

[0031] In step S203, a digital elevation model of the cable corridor is generated using the separated ground point cloud through a spatial interpolation algorithm. The interpolation algorithm estimates the elevation value of each regular grid point based on the distribution of the ground point cloud; common methods include inverse distance weighted interpolation or kriging interpolation. The generated digital elevation model is an elevation matrix in the form of a regular grid, with each grid cell storing an elevation value, thus continuously representing the terrain undulation characteristics, i.e., the terrain features, of the inspection area.

[0032] In step S204, point cloud clusters belonging to the cable body are identified and extracted from the non-ground point cloud using a density-based spatial clustering algorithm. This algorithm clusters points based on their spatial distribution density, grouping spatially adjacent points with the required density into the same cluster. The density-based spatial clustering algorithm effectively handles uneven point cloud density and aggregates point clouds exhibiting elongated, linear distribution characteristics into independent point cloud clusters. Clusters that conform to the cable's morphological characteristics are identified as point cloud clusters belonging to the cable body.

[0033] In step S205, a three-dimensional spatial curve is fitted to the identified point cloud cluster of the cable body. The fitting process aims to find a three-dimensional spatial curve that closely approximates the spatial distribution of all points in the point cloud cluster. The fitting method can be polynomial curve fitting or spline curve fitting. Through fitting, a curve representing the continuous spatial orientation of the cable can be calculated. This curve consists of a series of ordered three-dimensional spatial coordinate points, i.e., a continuous spatial position coordinate sequence. This sequence accurately represents the continuous position of the cable in three-dimensional space.

[0034] In step S206, a pre-set facility type template library stores the type definitions and related attribute information of common facilities in cable corridors, such as insulators, vibration dampers, and suspension clamps. Based on the preset inspection targets, a set of facility types associated with the current cable corridor's geographical information is matched from this library. The matching process can be based on attributes such as line voltage level and area type included in the geographical information, querying the template library for a list of facility types suitable for those conditions.

[0035] In step S207, the facilities to be inspected are located by combining the cable's spatial location (a continuous sequence of spatial coordinates) with the set of facility types through spatial proximity analysis. During the analysis, in the fused scene composed of the digital elevation model and non-ground point cloud, based on the typical relative position of each facility type to the cable (e.g., installed on a tower, suspended below the cable), the point cloud structure of the possible corresponding facility in the non-ground point cloud is searched near the cable's spatial location, and a spatial anchor point (three-dimensional coordinates) representing the location of each identified potential facility is determined. All the located spatial anchor points and their corresponding facility types constitute the list of facilities to be inspected.

[0036] In step S208, the digital elevation model, continuous spatial coordinate sequence, and list of facilities to be inspected are integrated and placed under a unified coordinate system. Association involves establishing relationships between the data structures, such as associating each spatial anchor point in the list of facilities to be inspected with its corresponding location in the digital elevation model, and indicating its adjacent cable segments (defined by the continuous spatial coordinate sequence). The encapsulated inspection task context is a structured data object containing terrain, cables, facilities, and their spatial relationships, providing a precise environmental and target perception foundation for subsequent inspection planning.

[0037] This embodiment separates the ground and ground features from the original point cloud, constructs a terrain model, extracts the 3D location of cables, and identifies associated facilities, achieving refined and automated 3D perception and semantic understanding of the cable corridor environment. This embodiment overcomes the shortcomings of relying solely on inaccurate and unintuitive information from 2D maps or design drawings, providing high-precision 3D spatial references and a clear target list for UAV inspections. This is a key prerequisite for subsequent adaptive path planning and precise data acquisition, significantly improving the environmental perception capability and task execution accuracy of the entire inspection system.

[0038] In some embodiments, identifying and extracting point cloud clusters belonging to the cable body from non-ground point clouds using a density-based spatial clustering algorithm includes: Voxel mesh downsampling is performed on non-ground point clouds to reduce point cloud density while maintaining spatial structure features, thus generating downsampled non-ground point clouds. In the downsampled non-ground point cloud, calculate the average distance of several nearest neighbor points for each point; Based on the statistical distribution of the average distance of all points, determine the distance threshold and the minimum number of points threshold; The following method samples each unvisited point in the non-terrestrial point cloud as a seed point, and finds all points whose spatial distance from the seed point is less than a distance threshold. If the number of points found is greater than or equal to the minimum number of points threshold, these points are marked as a temporary point cloud cluster, and the seed point is marked as visited. Using each unvisited point in the temporary point cloud cluster as a new seed point, the search and marking process is iteratively executed until the temporary point cloud cluster no longer expands. At this point, the temporary point cloud cluster is marked as a final point cloud cluster. Traverse all unvisited points and repeat the process from selecting seed points to marking final point cloud clusters until all points have been visited, resulting in multiple final point cloud clusters. From multiple final point cloud clusters, select the final point cloud clusters whose geometric morphological features conform to the cable morphological features. The geometric morphological features include the ratio of the main direction length of the point cloud cluster to the cross-sectional diameter and the linearity of the point cloud cluster. The final point cloud clusters that conform to the cable morphological characteristics are defined as point cloud clusters belonging to the cable body.

[0039] In this embodiment, voxel mesh downsampling process divides the three-dimensional space into cubic meshes of fixed size. In each non-empty voxel, a single representative point (such as the average coordinates of all points) is used to replace multiple original points, thereby compressing the amount of data and alleviating the uneven density of the point cloud. At the same time, the overall geometric contour of the scene is preserved, and the generated downsampled non-ground point cloud is more suitable for subsequent intensive computation.

[0040] When calculating the average distance for each point, the number of nearest neighbors selected can be preset to a fixed value based on the overall density of the point cloud; the Euclidean distance between the point and all its nearest neighbors is calculated and the arithmetic mean is obtained. The resulting average distance quantifies the spatial sparsity of the point's neighborhood.

[0041] The determination of the distance threshold and the minimum number of points threshold depends on the statistical distribution of the average distance of all points. The mean and standard deviation of all average distances can be calculated, and the distance threshold can be set as the mean plus a certain multiple of the standard deviation. The minimum number of points threshold can be set as an empirical value related to the size of typical objects in the scene, or dynamically adjusted inversely proportional to the distance threshold to ensure the formation of meaningful clusters.

[0042] Using any unvisited point as the seed point, find all points whose spatial distance to it is less than a distance threshold, which can be calculated using Euclidean distance. If the number of found points meets the minimum point threshold, initialize these points and the seed point together into a temporary point cloud cluster, and mark the state of the seed point as visited to prevent duplicate processing.

[0043] During the iteration process, each unvisited point within the temporary point cloud cluster is used as a new seed point, and the neighborhood search and point merging operations are repeatedly performed. If a newly found point meets the distance condition and has not been visited, it is merged into the current temporary point cloud cluster, and its status is marked as visited. This process is repeated recursively until the boundary points of the temporary point cloud cluster can no longer find new adjacent points that meet the conditions. At this point, the cluster stops growing and is marked as a final point cloud cluster.

[0044] Traverse all points and repeat the complete process from seed point selection to cluster labeling for each unvisited point to ensure that all density-connected regions are identified as independent final point cloud clusters, thereby obtaining multiple final point cloud clusters covering all potential objects.

[0045] During the screening process, a geometric morphology analysis is performed on each final point cloud cluster. The principal direction length is obtained by calculating the projection range of the point cloud along the first principal component direction obtained from principal component analysis. The cross-sectional diameter is approximately estimated by calculating the distribution range of the point cloud in a plane perpendicular to the principal direction. The linearity is calculated using the eigenvalues ​​from principal component analysis, typically the ratio of the first eigenvalue to the second eigenvalue. Cable morphology characteristics require the principal direction length to be significantly greater than the cross-sectional diameter (i.e., the ratio is greater than a preset threshold) and a high linearity (close to 1). Only final point cloud clusters that simultaneously meet both of these geometric conditions are retained.

[0046] The final point cloud clusters retained through the above morphological screening are formally defined as point cloud clusters belonging to the cable body identified by the algorithm, and are used for subsequent precise spatial curve fitting.

[0047] This embodiment extracts cable point clouds based on density clustering. Through adaptive dual threshold determination, an iterative expansion mechanism based on distance and state markers, and a dual geometric screening criterion combining principal direction length ratio and linearity, it can effectively distinguish between the linear structure of cables and blocky or scattered ground features such as towers and vegetation. It achieves high-precision and robust extraction of the cable body in complex point clouds, providing a reliable spatial perception foundation for the entire inspection process.

[0048] In some embodiments, based on the inspection task context, an adaptive planning algorithm decomposes a preset inspection target into multiple data acquisition sub-tasks with spatial order and sensor scheduling relationships, forming a structured data acquisition scheme, including: Based on the cable spatial location contained in the inspection task context, an initial flight path covering the cable spatial location is generated through a path planning algorithm; Based on the terrain features and list of facilities to be inspected contained in the inspection mission context, collision detection and field of view analysis are performed on the initial flight trajectory. The initial flight trajectory is adjusted to avoid terrain obstacles and ensure that each facility to be inspected is within the effective observation range of the preset sensors, thereby generating an optimized flight trajectory. Based on the preset inspection targets and the type of each facility in the list of facilities to be inspected, the data acquisition mode set required for status detection of each facility is determined by querying the preset sensor-task mapping rule base. The data acquisition mode set includes one or more of visible light imaging, infrared thermal imaging, lidar scanning and acoustic and vibration signal acquisition. The optimized flight path is discretized into ordered path points, and each path point is associated with one or more facilities in the list of facilities to be inspected that are located in the preset neighborhood of that path point. For each track point and its associated facilities to be inspected, multiple sensor action commands to be executed at that track point are generated based on the data acquisition mode set. The sensor action commands include sensor type, acquisition parameters, acquisition duration and gimbal pointing angle. Based on the order of waypoints on the optimized flight path, the sensor action commands corresponding to all waypoints are sorted and combined to form multiple acquisition sub-tasks with spatial order and sensor scheduling relationship. Multiple data acquisition subtasks, optimized flight paths, and data acquisition modal sets are encapsulated in a structured manner to generate a structured data acquisition solution.

[0049] In this embodiment, the path planning algorithm generates an initial flight path based on the cable's spatial location (a continuous sequence of spatial coordinates) within the context of the inspection task. The goal of this path is to closely follow the cable's spatial orientation while maintaining safe lateral and vertical distances. The algorithm can generate a smooth three-dimensional spatial curve as the initial flight path based on the cable coordinate sequence, using spline interpolation or piecewise linear connections.

[0050] Collision detection calculates the vertical distance between each point on the initial flight path and the terrain surface represented by the digital elevation model in the inspection mission context, and compares this distance with the UAV's safe flight altitude threshold to determine if there is a risk of collision. Field of view analysis, for each facility in the list of facilities to be inspected, calculates whether the facility is completely within the field of view cone when observed from the nearest point on the flight path at a typical field of view angle of a preset sensor. An optimized flight path is generated by adjusting the spatial position of the flight path points (such as elevation and lateral offset) to avoid terrain and optimize the observation perspective.

[0051] A pre-defined sensor-task mapping rule base stores the correspondence between facility types, inspection targets, and recommended data acquisition modes. For example, for a "tension clamp" facility and a "temperature detection" target, the rule might map to "infrared thermal imaging"; for an "insulator" facility and a "visual inspection" target, the rule might map to "visible light imaging." By querying this rule base based on the pre-defined inspection target and the type of facility to be inspected, the set of data acquisition modes required for condition monitoring of each facility can be determined.

[0052] The optimized flight path is discretized at fixed intervals or by curvature variations, resulting in a series of ordered path points. A preset neighborhood (such as a three-dimensional spherical space) is defined for each path point, and facilities in the list of facilities to be inspected whose spatial anchor points are located within this neighborhood are associated with that path point. A facility may be associated with multiple consecutive path points.

[0053] In the sensor action command, the sensor type specifies which physical sensor to use; the acquisition parameters are set according to factors such as ambient light and distance, such as the exposure parameters of a visible light camera and the temperature range of an infrared thermal imager; the acquisition duration is set according to the data stability requirements, such as the need to continuously acquire acoustic vibration signals for a certain period of time to obtain a stable spectrum; the gimbal pointing angle is obtained by calculating the spatial vector direction of the facility's spatial anchor point relative to the current flight path of the UAV.

[0054] Based on the optimized flight path, the sequence of waypoints is used to arrange all sensor action commands corresponding to each waypoint in chronological order, forming multiple data acquisition sub-tasks with a clear spatial order and sensor scheduling relationship. Each data acquisition sub-task specifies when (corresponding to the order of waypoints), where (waypoint coordinates), what target (associated facilities), and what type of sensor operation (action command) is performed.

[0055] The generated data acquisition subtasks, the complete coordinate sequence of the optimized flight path, and the globally used data acquisition mode set are then encapsulated in a structured manner. The encapsulation format organizes the above information into a hierarchical data structure, for example, using waypoints as indexes, with each index storing a list of associated facilities and a list of corresponding action commands, thereby generating a structured data acquisition scheme that can be directly parsed and executed by the flight control and payload systems.

[0056] This embodiment optimizes the flight path by combining cable location, terrain, and facility information, performs sensor task mapping based on a rule base, and generates and schedules action commands down to the level of flight path points. This transforms the macroscopic inspection target into a series of executable micro-operations that are precisely coordinated in space and time, ensuring the safety of UAV flight, the effectiveness of observation, and the timing correctness of multi-sensor collaborative data acquisition. This lays a solid foundation for subsequent efficient and reliable data synchronization and acquisition.

[0057] In some embodiments, and for each acquisition subtask, the required data acquisition mode and positioning method are identified, including: For each acquisition subtask in the structured acquisition scheme, extract the sensor action command corresponding to that acquisition subtask; Based on the sensor type contained in the sensor action command, the data acquisition mode required to perform the acquisition subtask is determined from the preset mode-capability mapping table. The data acquisition mode includes at least one of the following: three-dimensional geometric perception mode, visible light texture perception mode, infrared radiation perception mode, and acoustic vibration physical perception mode. Based on the spatial coordinates of the track point corresponding to the data collection subtask in the optimized flight path, and the terrain features contained in the inspection task context, the positioning environment category of the track point is determined. The positioning environment category includes open space positioning environment and satellite denial positioning environment. If the positioning environment category is open space positioning environment, then the positioning method of this acquisition sub-task will be identified as a fusion positioning method of global navigation satellite system positioning and inertial measurement unit positioning; If the positioning environment category is satellite denied positioning environment, then the positioning method of this acquisition sub-task will be identified as a fusion positioning method of visual inertial odometry positioning and ultra-wideband anchor point positioning. The identified data acquisition mode and positioning method are associated with metadata and written into the corresponding acquisition subtask.

[0058] In this embodiment, a pre-defined modality-capability mapping table defines a fixed correspondence between different sensor types and the categories of sensing capabilities they can provide. For example, LiDAR is mapped to a 3D geometric sensing mode, a visible light camera to a visible light texture sensing mode, an infrared thermal imager to an infrared radiation sensing mode, and an acoustic vibration sensor to an acoustic vibration physical sensing mode. By querying this table based on the sensor type in the sensor action command, the functional data acquisition mode required to perform the acquisition subtask can be determined.

[0059] Three-dimensional geometric perception mode refers to the ability to obtain the three-dimensional shape and spatial relationship of an object through measurement; visible light texture perception mode refers to the ability to record the appearance characteristics of an object in the visible light band; infrared radiation perception mode refers to the ability to perceive the infrared thermal radiation of an object's surface to reflect its temperature distribution; acoustic vibration physical perception mode refers to the ability to detect the sound waves or structural vibration signals emitted by an object.

[0060] Determining the positioning environment category of a waypoint requires combining its spatial coordinates with the terrain features within the context of the inspection mission. This can be achieved by analyzing the surrounding terrain features (digital elevation model) to identify steep terrain or tall objects that could obstruct satellite signals, or by using pre-defined geographic information layers (such as built-up areas or forest areas). The goal is to comprehensively determine whether the point belongs to an open spatial positioning environment conducive to satellite signal reception or a satellite-denied positioning environment where satellite signals are severely weakened.

[0061] In open space positioning environments, Global Navigation Satellite System (GNSS) positioning can provide accurate absolute geographical location, while Inertial Measurement Unit (INS) positioning provides high-frequency acceleration and angular velocity data to calculate short-term displacement. The two are fused through filtering algorithms (such as Kalman filtering) to output continuous, smooth, and high-precision position and attitude information, which is identified as the positioning method of this acquisition subtask.

[0062] In satellite-denied positioning environments, visual-inertial odometry (VIO) positioning utilizes onboard camera image sequences and inertial measurement unit (IMU) data to estimate the relative trajectory of the UAV through visual feature tracking and inertial navigation calculations. Ultra-wideband (UWB) anchor point positioning, on the other hand, requires the pre-deployment of UWB base stations (anchor points) at known locations within the inspection area. The UAV performs geometric positioning by measuring the distances between itself and multiple anchor points. By fusing and correcting the relative trajectory from the VIOMS with the absolute anchor point positions provided by the UWB, usable positioning accuracy can be maintained in areas without satellite signals, and this method is identified as the positioning method for this data acquisition sub-task.

[0063] The identified data acquisition modalities and positioning methods are used as descriptive metadata and are written into the corresponding acquisition subtask data structure in the form of key-value pairs or attribute fields. This allows the subsequent data acquisition execution module to clearly know which sensing capabilities each subtask needs to call and which positioning source should be relied upon for data spatiotemporal benchmark synchronization.

[0064] This embodiment defines the core sensing capabilities and spatiotemporal reference acquisition methods required for each acquisition subtask through modal mapping and adaptive positioning strategy identification. It not only plans the acquisition actions and spatial locations, but also pre-defines the positioning sources for functional requirements and environmental adaptation, thereby enhancing the system's adaptability to diverse tasks and complex environments. This provides a key pre-configuration basis for high-quality synchronous acquisition and accurate spatiotemporal registration of subsequent multi-source data.

[0065] In some embodiments, the simultaneous acquisition of three-dimensional point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data of the cable corridor includes: Upon arrival at each waypoint, execute the data collection sub-task associated with that waypoint; When the acquisition subtask includes a lidar scanning command, the environment around the current track point is scanned according to the preset scanning frequency and angle range to obtain the original lidar point cloud frame. The original lidar point cloud frames obtained from all track points are then timestamped and their coordinate systems are transformed and fused to generate three-dimensional point cloud data covering the cable corridor. When the acquisition subtask includes a visible light imaging instruction, one or more visible light images are acquired at the gimbal pointing angle specified by the acquisition subtask, and the acquired visible light images are arranged into a high-definition image sequence according to the track point order and timestamp. When the acquisition subtask includes an infrared thermal imaging command, a single or multiple frames of infrared thermal images are acquired at the gimbal pointing angle specified in the acquisition subtask, and the acquired infrared thermal images are arranged into infrared temperature field data according to the track point order and timestamp. When the acquisition subtask includes an instruction to acquire acoustic and vibration signals, the original audio signal and the original vibration signal are acquired within the acquisition duration specified by the acquisition subtask, and the original audio signal and the original vibration signal are subjected to noise reduction and filtering preprocessing to generate acoustic and vibration signal data. During the execution of the data acquisition sub-task at each track point, the real-time positioning data and attitude data output by the positioning sensor corresponding to the positioning method identified by the data acquisition sub-task are recorded synchronously. The real-time positioning data and attitude data are used as spatiotemporal references and are synchronized and spatially correlated with 3D point cloud data, high-definition image sequences, infrared temperature field data and acoustic vibration signal data, respectively.

[0066] In this embodiment, the original laser point cloud frame is a set of points acquired by the lidar within a single scan cycle, represented in its own coordinate system. Timestamp alignment marks the precise start time of acquisition for each frame of point cloud, image, heatmap, and signal segment. Coordinate system transformation utilizes real-time positioning and attitude data synchronously recorded during acquisition, transforming each frame of the original laser point cloud from the sensor coordinate system to a unified global world coordinate system via a coordinate transformation matrix. All transformed point cloud frames are then stitched and fused, removing overlapping portions to generate continuous and complete 3D point cloud data covering the cable corridor.

[0067] The acquisition of visible light images and infrared thermal images requires controlling the gimbal to rotate to the gimbal pointing angle specified by the acquisition subtask to ensure that the sensor's line of sight is aligned with the target. The images and thermal images are arranged according to the trackpoint sequence and timestamps. The resulting high-resolution image sequence and infrared temperature field data not only record visual and thermal radiation information but also implicitly contain the spatial order of the acquisition points.

[0068] The original audio and vibration signals are continuously sampled by a microphone and an accelerometer within a specified acquisition duration. Noise reduction can be achieved by using methods such as spectral subtraction to suppress steady-state environmental noise; filtering preprocessing can be performed using a bandpass filter to retain characteristic frequency components related to the cable's mechanical condition (such as loose bolts or corona discharge), thereby generating acoustic and vibration signal data that can be used for analysis.

[0069] Real-time positioning and attitude data are derived from the sensor combinations corresponding to the positioning method identified by the acquisition subtask identifier. Positioning data includes latitude, longitude, altitude, or 3D coordinates; attitude data includes roll, pitch, and yaw angles. Time synchronization ensures that the sampling clocks of all sensors are synchronized with the system master clock, or through interpolation alignment using high-precision timestamps. Spatial correlation involves calculating the 3D position and spatial orientation of the sensor in the global coordinate system at the time of acquisition for each frame of image, heatmap, and each segment of acoustic / vibration signal, based on the real-time positioning and attitude data corresponding to its timestamp. This establishes a mapping relationship between these data and the corresponding spatial regions in the 3D point cloud.

[0070] This embodiment ensures the consistency of multi-source heterogeneous data in both time and space dimensions through strictly synchronized trigger control, data acquisition and preprocessing based on a unified spatiotemporal reference, and precise spatial anchoring of all modal data using positioning information.

[0071] In some embodiments, spatiotemporal registration and multimodal fusion analysis are performed on 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data to generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis, and mechanical condition classification, including: Based on the real-time positioning and attitude data associated with 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data, each frame of the high-definition image sequence, each frame of the infrared temperature field data, and each segment of the acoustic vibration signal data are uniformly registered to the world coordinate system defined by the 3D point cloud data through coordinate transformation and time interpolation algorithms. The mapping relationship between image pixels, heat map pixels, signal segments and 3D point cloud spatial points is established to complete the spatiotemporal registration. From the spatiotemporally registered 3D point cloud data, extract the cable point cloud segments that characterize the cable body; The catenary equation parameters of the cable are obtained by fitting a three-dimensional spatial curve to the point cloud segment of the cable. Based on the catenary equation parameters and the digital elevation model contained in the inspection task context, the vertical distance between the lowest point of the cable sag and the terrain surface below is calculated, and the cable sag safety distance value is generated. Along the cable route, the safe distance values ​​of multiple sampling points are calculated at preset intervals to form a cable sag safe distance distribution curve. The cable sag safety assessment consists of the cable sag safe distance values ​​and the cable sag safe distance distribution curve. From a high-resolution image sequence that has been spatiotemporally registered, the surface appearance features of cables and their ancillary facilities are identified by a target detection algorithm. These surface appearance features include insulator contamination, metal component corrosion, foreign objects hanging on conductors, and signs of loose connectors. From the spatiotemporally registered infrared temperature field data, the local temperature distribution corresponding to the spatial location of surface appearance features is extracted, and the local temperature rise, temperature gradient and hot spot characteristics are calculated. The surface appearance features are associated with the corresponding local temperature distribution. The matching reasoning is performed through a pre-set appearance-temperature association rule base to determine whether there is a temperature anomaly pattern induced or accompanied by the appearance anomaly, and to generate a latent defect diagnosis conclusion. The latent defect diagnosis conclusion includes the defect type, location and risk level. From the spatiotemporally registered acoustic and vibration signal data, the Mel frequency cepstral coefficient features of the audio signal and the spectral features of the vibration signal are extracted. The Mel frequency cepstral coefficient features and spectral features are input into the pre-trained acoustic vibration classification model. The acoustic vibration classification model outputs mechanical state classification labels for cables and their accessories. The mechanical state classification labels include normal tightness, bolt looseness, corona discharge, and vortex-induced vibration. The cable sag safety assessment, latent defect diagnosis conclusions, and mechanical condition classification labels are integrated and packaged into a structured initial inspection report.

[0072] In this embodiment, coordinate transformation utilizes real-time positioning and attitude data corresponding to the acquisition time of each image or thermal image frame to calculate the position and attitude matrix of the sensor's optical center in the world coordinate system. Then, each image pixel is back-projected into a three-dimensional point cloud space using perspective projection or collinearity equations to determine its corresponding three-dimensional point. A time interpolation algorithm is used to handle the time alignment between continuous data such as acoustic and vibration signals and discrete image frames, obtaining the precise signal segment corresponding to the image acquisition time through interpolation.

[0073] The cable point cloud segment is extracted from the registered 3D point cloud using spatial indexing, capturing the point set belonging to the cable body. A 3D spatial curve is fitted to the cable point cloud segment, and the parameters of the catenary equation, which describes the natural sag curve of the cable under its own weight, are solved using the least squares method. Based on the solved catenary equation, the 3D coordinates of its lowest point are calculated, and the terrain elevation directly below this point is interpolated from the digital elevation model within the inspection task context. Subtracting the two yields the cable sag safety distance.

[0074] Target detection algorithms (such as deep learning-based models) are used to automatically select components such as insulators and clamps from high-resolution image sequences and identify abnormal signs such as dirt and corrosion on their surfaces. These identification results constitute surface appearance features. From infrared temperature field data, based on the spatial location corresponding to the surface appearance features, the temperature matrix of that region is extracted, and its temperature rise relative to the environment or adjacent normal regions (local temperature rise), the rate of temperature change in space (temperature gradient), and the presence of significant high-temperature regions (hot spot features) are calculated.

[0075] The pre-defined appearance-temperature association rule base contains a series of logical judgment rules. For example, a rule might stipulate that when rust is detected on a metal part and a significant hot spot is present in the corresponding area, an increased contact resistance defect is inferred; when a foreign object is detected dangling from a wire but the temperature is not abnormal, it may only be judged as an appearance hazard. Through matching reasoning, a latent defect diagnosis conclusion is generated, which includes the specific defect type (such as poor contact, contamination flashover risk), precise location, and estimated risk level.

[0076] Mel frequency cepstral coefficients are obtained by performing a series of processes on the audio signal, including framing, windowing, fast Fourier transform, Mel filter bank filtering, and discrete cosine transform, and can characterize the short-time power spectrum characteristics of sound. The spectral characteristics of vibration signals are obtained by performing a fast Fourier transform on the vibration acceleration signal and calculating its energy distribution at different frequency components.

[0077] Pre-trained acoustic vibration classification models are typically classifiers such as neural networks (e.g., convolutional neural networks) or support vector machines. The training process uses a large number of labeled acoustic vibration signal samples. The Mel-frequency cepstral coefficients and spectral features of the samples are used as inputs, while the corresponding mechanical state classification labels (e.g., "loose bolts") serve as supervision signals. Through iterative optimization, the model learns the mapping from features to states. The trained model can automatically classify new acoustic vibration signal features and output mechanical state classification labels.

[0078] The calculated cable sag safety distance values ​​and their distribution curves, the inferred latent defect diagnostic conclusions, and the classified mechanical condition labels are organized and filled according to a preset report template, and packaged into a structured initial inspection report. This report comprehensively presents the cable's geometric safety, potential defects, and mechanical health status in the form of a combination of data tables, text descriptions, and visual charts.

[0079] This embodiment establishes a unified analysis foundation for multi-source data through spatiotemporal registration, and uses geometric calculation, rule reasoning and machine learning models to automatically and quantitatively assess the sag safety, latent defects and mechanical condition of cables. This achieves comprehensive, in-depth and efficient intelligent diagnosis of cable laying status, providing a direct and reliable basis for operation and maintenance decisions.

[0080] In some embodiments, the initial inspection report is evaluated for defect detection rate and false alarm rate, including: The latent defect diagnosis conclusions and mechanical condition classification labels contained in the initial inspection report are combined and defined as the defect set automatically identified by the algorithm. From the preset cable defect standard library, retrieve the set of known standard defects that match the geographical information of the cable corridor and the preset inspection target corresponding to the current inspection task. The set of known standard defects includes the defect type, standard location and standard severity. The data in the cable defect standard library comes from defect records that have been manually verified and confirmed in the historical operation and maintenance of the cable corridor or higher-precision special inspection reports. The algorithm automatically identifies a set of defects and compares their spatial location and type consistency with a set of known standard defects. The defect detection rate is obtained by dividing the number of defects in the defect set automatically identified by the calculation algorithm that match the known standard defect set in both spatial location and defect type by the total number of defects in the known standard defect set. The false alarm rate is calculated by dividing the number of defects in the set of defects automatically identified by the algorithm that cannot be matched with the known standard defect set in terms of spatial location or defect type by the total number of defects in the set of defects automatically identified by the algorithm. The defect detection rate is compared with the preset detection rate threshold, and the false alarm rate is compared with the preset false alarm rate threshold. If the defect detection rate is lower than the detection rate threshold, or the defect false alarm rate is higher than the false alarm rate threshold, the initial inspection report is deemed to have failed the defect detection rate and false alarm rate assessment. If the defect detection rate is not lower than the detection rate threshold and the false alarm rate is not higher than the false alarm rate threshold, then the initial inspection report is deemed to have passed the defect detection rate and false alarm rate assessment.

[0081] In this embodiment, the defect set automatically identified by the algorithm is a summary of all suspected defects automatically generated through multimodal fusion analysis during this inspection. The known standard defect set serves as the benchmark truth value for evaluation and originates from a defect archive that is pre-established and maintained for the same inspection area and target using reliable means independent of this automatic inspection (such as historical manual inspection reports and offline special inspection data).

[0082] Spatial location matching requires setting a location tolerance threshold, which comprehensively considers the UAV positioning error, the location accuracy of defect reports, and the actual size of the facility. During matching, the three-dimensional Euclidean distance between the coordinates of two defect locations is calculated. If the distance is less than the tolerance threshold, the spatial location is considered to be successfully matched. Type consistency comparison requires predefining a defect type mapping table. This table specifies the equivalence or subordinate relationship between the defect type descriptions output by the automatic identification system and the type terms in the standard defect library, ensuring that the same type of defect with different descriptions can be correctly associated.

[0083] The defect detection rate measures the extent to which the algorithm covers known real defects, and its calculation depends on the number of successfully matched defects; the false positive rate measures the purity of the algorithm's results, and its calculation depends on the number of defects that the algorithm independently identifies but cannot find a corresponding item in the standard library.

[0084] The preset detection rate threshold and false alarm rate threshold are performance acceptable levels set based on actual operation and maintenance needs. The detection rate threshold is usually set higher to ensure that the vast majority of real defects are detected; the false alarm rate threshold is usually set lower to control the number of false alarms and avoid wasting operation and maintenance resources. The specific values ​​of these thresholds can be dynamically adjusted according to the importance of the line, the inspection cycle, and the subsequent processing costs.

[0085] The evaluation logic is based on the comparison between the two values ​​mentioned above and their corresponding thresholds. Failure to pass the evaluation indicates that the algorithm's performance has not met the preset minimum requirements in either the "discovery capability" or "judgment accuracy" dimension. The automatic analysis results of this inspection have systematic biases or deficiencies, and its directly output report cannot be directly accepted. Root cause analysis and optimization processes must be triggered.

[0086] This embodiment establishes a quantitative evaluation system based on benchmark truth comparison. It solves the data alignment problem through spatial tolerance matching and type mapping, and uses the detection rate and false alarm rate—two key indicators—to objectively measure the inspection analysis results. This system transforms the quality of inspection reports from subjective judgment to objective data, providing accurate and reliable trigger signals for the system to identify its own performance shortcomings and initiate targeted closed-loop optimization.

[0087] In some embodiments, the execution chain of the structured acquisition scheme is traced back to locate the defective link in the analysis, and optimized acquisition instructions for the defective link are generated, including: Based on the comparison between the defect set automatically identified by the algorithm and the known standard defect set, missed defects and false alarm defects are identified. Missed defects refer to defects that exist in the known standard defect set but are not identified in the defect set automatically identified by the algorithm; False alarm defects refer to defects that exist in the set of defects automatically identified by the algorithm but cannot be matched with the known standard set of defects; For each missed defect, based on the standard location of the missed defect recorded in the known standard defect set, the nearest track point to the standard location is found in the optimized flight track of the structured acquisition scheme. In the data acquisition sub-task associated with this track point, analyze whether the data acquisition mode specified by the sensor action command covers the sensing capability required to identify the missed defect; If not covered, the analysis defect corresponding to the missed detection defect is determined to be a missing sensor configuration step; If it has been covered, the quality of the raw data collected by the track point will be further checked. If the raw data has problems such as blurriness, occlusion or low signal-to-noise ratio, the analysis defect corresponding to the missed defect will be determined to be the data quality deficiency link. For each false alarm defect, based on the identification location of the false alarm defect recorded in the defect set automatically identified by the algorithm, the nearest track point to the identification location is found in the optimized flight track of the structured acquisition scheme. In the data acquisition subtask associated with the waypoint, analyze whether the data acquisition parameters specified by the sensor action command are within a range that is prone to interference, or whether there are known sources of interference in the environment where the waypoint is located. If there are improper parameters or environmental interference, the analysis defect corresponding to the false alarm defect is determined to be the interference introduction stage; The analysis defect links corresponding to all missed defects and all false alarm defects are summarized to form an analysis defect link set. Based on the type of each analytical defect in the set of analytical defect links, the corresponding optimization operation is matched from the pre-set optimization strategy library, including: Add or adjust the data acquisition mode to address missing components in the sensor configuration; Adjust flight paths, gimbal angles, or sensor parameters to improve data quality in areas with insufficient data quality. To mitigate interference, adjust data acquisition parameters or add filtering preprocessing steps. All matched optimization operations are converted into optimized acquisition instructions that can be parsed and executed by the structured acquisition scheme.

[0088] In this embodiment, the root cause analysis of missed defects follows the logic of inferring the perception chain from the result. If a defect is not identified, the first step is to infer whether it is due to a lack of corresponding perception capabilities. This requires mapping the core data modality on which the defect's detection depends based on its physical characteristics. For example, detecting metal corrosion depends on the visible light texture perception modality. If the acquisition subtask of the corresponding track point does not contain visible light imaging instructions, the root cause is determined to be a lack of sensor configuration.

[0089] If the perceptual modality is already covered, it is necessary to infer whether feature extraction failed due to poor quality of the perceptual data itself. The criteria for determining insufficient data quality include: whether the spatial resolution of the image is sufficient to distinguish details, whether the image is blurred due to motion or defocus, whether the target is partially or completely occluded by other objects in the image, and whether the signal-to-noise ratio of the signal is lower than the threshold for reliable analysis.

[0090] Root cause analysis of false alarm defects focuses on identifying factors that cause the algorithm to generate false alarms. Determining the interference introduction stage requires examining the following aspects: whether the acquisition parameters are improperly set, such as using an excessively low infrared thermal imager temperature range in a low-temperature environment, which may amplify slight fluctuations in normal temperature differences into abnormal hot spots; whether there are interference sources in the environment during acquisition that are similar to the defect characteristics, such as strong sunlight reflection on a metal surface potentially being misjudged as a high-temperature point, or the specific spectrum of background wind noise potentially being confused with the acoustic characteristics of corona discharge.

[0091] The pre-built optimization strategy library includes optimization operations defined for missing steps in sensor configuration. These include adding corresponding sensor action commands to the data acquisition subtask and setting reasonable initial values ​​for acquisition parameters. Optimization operations defined for insufficient data quality include adjusting trackpoint positions to obtain an unobstructed view, adjusting gimbal pitch angles to obtain frontal images, optimizing camera focal length and exposure parameters to improve sharpness, or extending the acquisition time of acoustic and vibration signals to improve the signal-to-noise ratio. Optimization operations defined for interference introduction include correcting sensor parameters prone to misjudgment (such as adjusting infrared emissivity or modifying the contrast threshold in image processing), or inserting filtering or suppression algorithms for known interference patterns into the data processing pipeline.

[0092] The generated optimized acquisition instructions need to accurately describe the modifications to specific elements in the structured acquisition scheme. Specifically, this can be achieved by specifying the index of the target track point, the type of sensor action instruction to be modified or added, and its specific parameter key-value pairs.

[0093] This embodiment establishes a reverse reasoning mechanism based on defect type and perception logic, as well as an operable optimization strategy mapping system for different problem root causes. This enables the system to not only discover performance deficiencies, but also accurately diagnose the root causes of problems and automatically generate improvement solutions. It realizes an intelligent closed loop from problem identification to solution generation, and improves the system's adaptability and long-term performance.

[0094] In a second aspect, this embodiment also provides an intelligent UAV inspection system for cable laying status, applicable to the method described in the first aspect. The system includes a task instruction receiving module, an environmental perception and analysis module, a task planning and scheme generation module, a data synchronization acquisition module, a multimodal data analysis module, a report quality assessment module, a link backtracking and optimization instruction generation module, a knowledge base update module, and an iterative execution and output module. The task instruction receiving module receives flight control instructions containing cable corridor geographic information and preset inspection targets, as well as associated multi-source sensor configuration parameters. The environmental perception and analysis module performs environmental perception and task analysis on the flight control instructions and multi-source sensor configuration parameters, generating an inspection task context containing terrain features, cable spatial location, and a list of facilities to be inspected. The task planning and scheme generation module, based on the inspection task context, decomposes the preset inspection targets into multiple acquisition sub-tasks with spatial order and sensor scheduling relationships using an adaptive planning algorithm, forming a structured acquisition scheme, and identifying the required data acquisition modality and positioning method for each acquisition sub-task. The data synchronization acquisition module executes each acquisition sub-task sequentially according to the structured acquisition scheme, synchronizing... The system acquires 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic and vibration signal data of the cable corridor. A multimodal data analysis module performs spatiotemporal registration and multimodal fusion analysis on the 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic and vibration signal data to generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis, and mechanical condition classification. A report quality assessment module evaluates the defect detection rate and false alarm rate of the initial inspection report. A link backtracking and optimization instruction generation module is used to backtrack the process when the initial inspection report fails the defect detection rate and false alarm rate assessment. The structured data acquisition scheme's execution chain locates and analyzes defective links, and generates optimized data acquisition instructions for analyzing these links. The knowledge base update module updates the historical inspection case library used for task decomposition or the defect feature knowledge base used for multimodal fusion analysis based on the optimized data acquisition instructions. The iterative execution and output module re-executes the steps from generating the inspection task context to generating the initial inspection report based on the updated historical inspection case library or defect feature knowledge base, until the initial inspection report passes the defect detection rate and false alarm rate evaluation, and outputs the final cable laying status inspection report and risk warning information.

[0095] In this embodiment, the system integrates intelligent perception, planning, analysis, evaluation and closed-loop self-optimization functions through the coordinated operation of various modules. This closed-loop design enables the system to continuously learn from historical experience and optimize itself, which significantly improves the automation level of UAV inspection, the accuracy of defect detection and the long-term adaptability and reliability of the entire system in complex environments.

[0096] By adopting the above technical solutions, this invention differs from existing technologies and possesses the following beneficial effects: By receiving flight control commands and sensor configuration parameters, it generates an inspection task context through environmental perception and task parsing. Based on this context, an adaptive planning algorithm forms a refined, structured data acquisition scheme, guiding the UAV to simultaneously acquire multimodal data. Through spatiotemporal registration and fusion analysis of multi-source data, it achieves automated and quantitative comprehensive evaluation of cable sag safety, latent defects, and mechanical condition, generating an initial inspection report. Furthermore, it introduces a report quality quantification evaluation mechanism based on defect detection rate and false alarm rate. When a report fails to meet standards, it can automatically trace back the acquisition and analysis links to locate the problem link and generate targeted optimization commands to update the planning case library or defect feature knowledge base, thereby driving a new round of optimized inspection processes. The above technical solutions construct a complete closed loop of "perception-planning-execution-analysis-evaluation-optimization." Through iterative feedback and knowledge accumulation, the system can adapt to complex environments and changing targets, significantly improving the intelligence level of UAV inspection operations, the accuracy and reliability of defect detection, and the long-term operational autonomy and decision-making credibility of the entire system.

[0097] Furthermore, 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0098] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part 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.) or processor to execute all or part of the steps of the methods of 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.

[0099] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for intelligent inspection of cable laying status using unmanned aerial vehicles (UAVs), characterized in that, include: Receive flight control commands containing cable corridor geographic information and preset inspection targets, as well as associated multi-source sensor configuration parameters; The flight control commands and the configuration parameters of the multi-source sensors are subjected to environmental perception and task analysis to generate an inspection task context that includes terrain features, cable spatial location and a list of facilities to be inspected; Based on the inspection task context, the preset inspection target is decomposed into multiple acquisition sub-tasks with spatial order and sensor scheduling relationship through an adaptive planning algorithm, forming a structured acquisition scheme, and each acquisition sub-task is identified with its required data acquisition mode and positioning method. According to the structured acquisition scheme, each acquisition sub-task is executed sequentially to simultaneously acquire the three-dimensional point cloud data, high-definition image sequence, infrared temperature field data and acoustic vibration signal data of the cable corridor; Spatiotemporal registration and multimodal fusion analysis are performed on the three-dimensional point cloud data, high-definition image sequences, infrared temperature field data and acoustic vibration signal data to generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis and mechanical condition classification. The initial inspection report is evaluated for defect detection rate and false alarm rate. If the evaluation fails, the execution link of the structured acquisition scheme is traced back to locate and analyze the defective link, and an optimized acquisition instruction is generated for the defective link. Based on the optimized acquisition instructions, update the historical inspection case library used for task decomposition or the defect feature knowledge base used for multimodal fusion analysis; Based on the updated historical inspection case library or defect feature knowledge base, the steps from generating the inspection task context to generating the initial inspection report are re-executed until the initial inspection report passes the defect detection rate and false alarm rate evaluation, and the final cable laying status inspection report and risk warning information are output.

2. The intelligent UAV inspection method for cable laying status according to claim 1, characterized in that, The flight control commands and the configuration parameters of the multi-source sensors are used for environmental perception and task analysis to generate an inspection task context that includes terrain features, cable spatial locations, and a list of facilities to be inspected, including: Based on the geographic information of the cable corridor, obtain the original point cloud data of the cable corridor area; The original point cloud data is subjected to ground point filtering and classification processing to separate ground point cloud from non-ground point cloud; Using the ground point cloud, a digital elevation model of the cable corridor is generated through an interpolation algorithm, and the digital elevation model represents the terrain features; From the non-ground point cloud, point cloud clusters belonging to the cable body are identified and extracted using a density-based spatial clustering algorithm; A three-dimensional spatial curve is fitted to the point cloud cluster of the cable body to calculate a continuous spatial position coordinate sequence of the cable in three-dimensional space. The spatial position of the cable is characterized by the continuous spatial position coordinate sequence. Based on the preset inspection target, match the set of facility types associated with the geographical information of the cable corridor from the preset facility type template library; By combining the spatial location of the cable with the set of facility types, spatial proximity analysis is used to locate the spatial anchor points of the facilities to be inspected in the scene composed of the digital elevation model and the non-ground point cloud, thus forming the list of facilities to be inspected. The digital elevation model, the continuous spatial location coordinate sequence, and the list of facilities to be inspected are integrated and associated, and encapsulated into the inspection task context.

3. The intelligent UAV inspection method for cable laying status according to claim 2, characterized in that, From the non-ground point cloud, point cloud clusters belonging to the cable body are identified and extracted using a density-based spatial clustering algorithm, including: The non-ground point cloud is subjected to voxel mesh downsampling to reduce the point cloud density while maintaining spatial structure features, thereby generating a downsampled non-ground point cloud. In the downsampled non-ground point cloud, the average distance between several nearest neighbor points is calculated for each point; Based on the statistical distribution of the average distance of all points, determine the distance threshold and the minimum number of points threshold; Using each unvisited point in the downsampled non-ground point cloud as a seed point, find all points whose spatial distance from the seed point is less than the distance threshold; If the number of points found is greater than or equal to the minimum number of points threshold, these points are marked as a temporary point cloud cluster, and the seed point is marked as visited. Using each unvisited point in the temporary point cloud cluster as a new seed point, the search and marking process is iteratively executed until the temporary point cloud cluster no longer expands. At this point, the temporary point cloud cluster is marked as a final point cloud cluster. Traverse all unvisited points and repeat the process from selecting seed points to marking final point cloud clusters until all points have been visited, resulting in multiple final point cloud clusters. From the multiple final point cloud clusters, select the final point cloud clusters whose geometric morphological features conform to the cable morphological features. The geometric morphological features include the ratio of the main direction length of the point cloud cluster to the cross-sectional diameter and the linearity of the point cloud cluster. The final point cloud clusters that conform to the cable morphological characteristics are defined as point cloud clusters belonging to the cable body.

4. The intelligent UAV inspection method for cable laying status according to claim 1, characterized in that, Based on the inspection task context, the preset inspection target is decomposed into multiple data acquisition sub-tasks with spatial order and sensor scheduling relationship through an adaptive planning algorithm, forming a structured data acquisition scheme, including: Based on the cable spatial location contained in the inspection task context, an initial flight path covering the cable spatial location is generated through a path planning algorithm; Based on the terrain features and list of facilities to be inspected contained in the inspection task context, collision detection and field of view analysis are performed on the initial flight trajectory, the initial flight trajectory is adjusted to avoid terrain obstacles and ensure that each facility to be inspected is within the effective observation range of the preset sensor, and an optimized flight trajectory is generated. Based on the preset inspection target and the type of each facility in the list of facilities to be inspected, the data acquisition mode set required for status detection of each facility is determined by querying the preset sensor-task mapping rule base. The data acquisition mode set includes one or more of visible light imaging, infrared thermal imaging, lidar scanning and acoustic vibration signal acquisition. The optimized flight path is discretized into ordered path points, and each path point is associated with one or more facilities in the list of facilities to be inspected that are located in a preset neighborhood of that path point. For each track point and its associated facility to be inspected, based on the data acquisition mode set, multiple sensor action commands to be executed at that track point are generated. The sensor action commands include sensor type, acquisition parameters, acquisition duration and gimbal pointing angle. Based on the order of waypoints on the optimized flight path, the sensor action commands corresponding to all waypoints are sorted and combined to form the multiple acquisition sub-tasks with spatial order and sensor scheduling relationship. The multiple acquisition sub-tasks, the optimized flight path, and the data acquisition mode set are structurally encapsulated to generate the structured acquisition scheme.

5. The intelligent UAV inspection method for cable laying status according to claim 4, characterized in that, And for each acquisition subtask, identify its required data acquisition mode and positioning method, including: For each acquisition subtask in the structured acquisition scheme, extract the sensor action command corresponding to that acquisition subtask; Based on the sensor type contained in the sensor action command, the data acquisition mode required to perform the acquisition subtask is determined from a preset mode-capability mapping table. The data acquisition mode includes at least one of three-dimensional geometric perception mode, visible light texture perception mode, infrared radiation perception mode, and acoustic vibration physical perception mode. Based on the spatial coordinates of the track point corresponding to the acquisition sub-task in the optimized flight path, and the terrain features contained in the context of the inspection task, the positioning environment category of the track point is determined. The positioning environment category includes open space positioning environment and satellite denial positioning environment. If the positioning environment category is an open space positioning environment, then the positioning method of the acquisition sub-task is identified as a fusion positioning method of global navigation satellite system positioning and inertial measurement unit positioning; If the positioning environment category is a satellite denied positioning environment, then the positioning method of the acquisition sub-task is identified as a fusion positioning method of visual inertial odometry positioning and ultra-wideband anchor point positioning. The identified data acquisition mode and the positioning method are associated as metadata and written into the corresponding acquisition subtask.

6. The intelligent UAV inspection method for cable laying status according to claim 1, characterized in that, Simultaneously acquire 3D point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data of the cable corridor, including: Upon arrival at each waypoint, execute the data collection sub-task associated with that waypoint; When the acquisition subtask includes a lidar scanning command, the environment around the current track point is scanned according to the preset scanning frequency and angle range to obtain the original lidar point cloud frame. The original lidar point cloud frames obtained from all track points are then timestamped and their coordinate systems are transformed, and the three-dimensional point cloud data covering the cable corridor is generated. When the acquisition subtask includes a visible light imaging instruction, one or more visible light images are acquired at the gimbal pointing angle specified by the acquisition subtask, and the acquired visible light images are arranged into the high-definition image sequence according to the track point order and timestamp. When the acquisition subtask includes an infrared thermal imaging command, a single frame or multiple frames of infrared thermal images are acquired at the gimbal pointing angle specified by the acquisition subtask, and the acquired infrared thermal images are arranged into infrared temperature field data according to the track point order and timestamp. When the acquisition subtask includes an acoustic and vibration signal acquisition instruction, the original audio signal and the original vibration signal are acquired within the acquisition duration specified by the acquisition subtask, and the original audio signal and the original vibration signal are subjected to noise reduction and filtering preprocessing to generate the acoustic and vibration signal data. During the execution of the data acquisition sub-task at each track point, the real-time positioning data and attitude data output by the positioning sensor corresponding to the positioning method identified by the data acquisition sub-task are recorded synchronously. The real-time positioning data and attitude data are used as spatiotemporal references and are synchronized and spatially correlated with the three-dimensional point cloud data, the high-definition image sequence, the infrared temperature field data, and the acoustic vibration signal data, respectively.

7. The intelligent UAV inspection method for cable laying status according to claim 1, characterized in that, The three-dimensional point cloud data, high-definition image sequences, infrared temperature field data, and acoustic vibration signal data are subjected to spatiotemporal registration and multimodal fusion analysis to generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis, and mechanical condition classification, including: Based on the real-time positioning and attitude data associated with the three-dimensional point cloud data, the high-definition image sequence, the infrared temperature field data, and the acoustic vibration signal data, each frame of the high-definition image sequence, each frame of the infrared temperature field data, and each segment of the acoustic vibration signal data are uniformly registered to the world coordinate system defined by the three-dimensional point cloud data through coordinate transformation and time interpolation algorithms. The mapping relationship between image pixels, heat map pixels, signal segments, and three-dimensional point cloud spatial points is established to complete the spatiotemporal registration. From the spatiotemporally registered 3D point cloud data, extract cable point cloud segments that characterize the cable body; The point cloud segment of the cable is fitted with a three-dimensional spatial curve to obtain the catenary equation parameters of the cable. Based on the catenary equation parameters and the digital elevation model contained in the inspection task context, the vertical distance between the lowest point of the cable sag and the terrain surface below is calculated, and the cable sag safety distance value is generated. Along the cable route, the safe distance values ​​of multiple sampling points are calculated at preset intervals to form a cable sag safe distance distribution curve. The cable sag safety assessment is composed of the cable sag safe distance values ​​and the cable sag safe distance distribution curve. From the spatiotemporally registered high-definition image sequence, the surface appearance features of the cable and its accessories are identified by a target detection algorithm. These surface appearance features include insulator dirt, metal component corrosion, foreign objects hanging on the conductors, and signs of loose connectors. From the spatiotemporally registered infrared temperature field data, the local temperature distribution corresponding to the spatial location of the surface appearance features is extracted, and the local temperature rise, temperature gradient and hot spot features are calculated. The surface appearance features are associated with the corresponding local temperature distribution. Matching and reasoning are performed using a pre-set appearance-temperature association rule base to determine whether there is a temperature anomaly pattern induced or accompanied by the appearance anomaly. A latent defect diagnosis conclusion is generated, which includes the defect type, location, and risk level. From the spatiotemporally registered acoustic and vibration signal data, the Mel frequency cepstral coefficient features of the audio signal and the spectral features of the vibration signal are extracted. The Mel frequency cepstral coefficient features and the spectral features are input into a pre-trained acoustic vibration classification model. The acoustic vibration classification model outputs mechanical state classification labels for cables and their accessories. The mechanical state classification labels include normal tightness, bolt looseness, corona discharge, and eddy-induced vibration. The cable sag safety assessment, the latent defect diagnosis conclusion, and the mechanical condition classification label are integrated and packaged into a structured initial inspection report.

8. The intelligent UAV inspection method for cable laying status according to claim 1, characterized in that, The initial inspection report is evaluated for defect detection rate and false alarm rate, including: The latent defect diagnosis conclusions and mechanical condition classification labels contained in the initial inspection report are combined and defined as the defect set automatically identified by the algorithm. From the preset cable defect standard library, retrieve the set of known standard defects that match the cable corridor geographic information and preset inspection target corresponding to the current inspection task. The set of known standard defects includes defect type, standard location and standard severity. The set of defects automatically identified by the algorithm is matched in terms of spatial location and type consistency with the set of known standard defects; The number of defects in the defect set automatically identified by the algorithm that match the known standard defect set in both spatial location and defect type is calculated, and then divided by the total number of defects in the known standard defect set to obtain the defect detection rate. The number of defects in the defect set automatically identified by the algorithm that cannot be matched with the known standard defect set in terms of spatial location or defect type is calculated and divided by the total number of defects in the defect set automatically identified by the algorithm to obtain the defect false alarm rate. The defect detection rate value is compared with a preset detection rate threshold, and the defect false alarm rate value is compared with a preset false alarm rate threshold. If the defect detection rate is lower than the detection rate threshold, or the false alarm rate is higher than the false alarm rate threshold, then the initial inspection report is determined to have failed the defect detection rate and false alarm rate assessment. If the defect detection rate is not lower than the detection rate threshold and the false alarm rate is not higher than the false alarm rate threshold, then the initial inspection report is determined to have passed the defect detection rate and false alarm rate assessment.

9. The intelligent UAV inspection method for cable laying status according to claim 1, characterized in that, Then, the execution chain of the structured acquisition scheme is traced back to locate and analyze the defective link, and optimized acquisition instructions are generated for the defective link, including: Based on the comparison results between the defect set automatically identified by the algorithm and the known standard defect set, missed defects and false alarm defects are identified. The missed defect refers to a defect that exists in the known standard defect set but is not identified in the defect set automatically identified by the algorithm; The false alarm defect refers to a defect that exists in the defect set automatically identified by the algorithm but cannot be matched with the known standard defect set; For each of the aforementioned missed defects, based on the standard location of the missed defect recorded in the known standard defect set, the nearest track point to the standard location is found in the optimized flight track of the structured acquisition scheme; In the data acquisition sub-task associated with this track point, analyze whether the data acquisition mode specified by the sensor action command covers the sensing capability required to identify the missed defect; If not covered, the analysis defect corresponding to the missed detection defect is determined to be a missing sensor configuration step; If it has been covered, the quality of the raw data collected by the track point will be further checked. If the raw data has problems such as blurriness, occlusion or low signal-to-noise ratio, the analysis defect corresponding to the missed defect will be determined to be the data quality deficiency link. For each false alarm defect, based on the identification location of the false alarm defect recorded in the defect set automatically identified by the algorithm, the nearest track point to the identification location is found in the optimized flight track of the structured acquisition scheme; In the data acquisition subtask associated with the waypoint, analyze whether the data acquisition parameters specified by the sensor action command are within a range that is prone to interference, or whether there are known sources of interference in the environment where the waypoint is located. If there are improper parameters or environmental interference, the analysis defect corresponding to the false alarm defect is determined to be the interference introduction stage; The analysis defect links corresponding to all the missed detection defects and the analysis defect links corresponding to all the false alarm defects are summarized to form an analysis defect link set; Based on the type of each analytical defect in the set of analytical defect links, a corresponding optimization operation is matched from a pre-set optimization strategy library, including: Add or adjust the data acquisition mode to address missing components in the sensor configuration; Adjust flight paths, gimbal angles, or sensor parameters to improve data quality in areas with insufficient data quality. To mitigate interference, adjust data acquisition parameters or add filtering preprocessing steps. All matched optimization operations are converted into optimized acquisition instructions that can be parsed and executed by the structured acquisition scheme.

10. A UAV intelligent inspection system for cable laying status, characterized in that, The system applicable to the method of any one of claims 1 to 9 comprises: The mission command receiving module is used to receive flight control commands containing cable corridor geographic information and preset inspection targets, as well as associated multi-source sensor configuration parameters. The environmental perception and analysis module is used to perform environmental perception and task analysis on the flight control commands and the configuration parameters of the multi-source sensors, and generate an inspection task context that includes terrain features, cable spatial location and a list of facilities to be inspected. The task planning and scheme generation module is used to decompose the preset inspection target into multiple acquisition sub-tasks with spatial order and sensor scheduling relationship based on the inspection task context and through an adaptive planning algorithm, forming a structured acquisition scheme, and identifying the required data acquisition mode and positioning method for each acquisition sub-task. The data synchronization acquisition module is used to execute each acquisition sub-task in sequence according to the structured acquisition scheme, and synchronously acquire the three-dimensional point cloud data, high-definition image sequence, infrared temperature field data and acoustic vibration signal data of the cable corridor. The multimodal data analysis module is used to perform spatiotemporal registration and multimodal fusion analysis on the three-dimensional point cloud data, high-definition image sequences, infrared temperature field data and acoustic vibration signal data, and generate an initial inspection report that includes cable sag safety assessment, latent defect diagnosis and mechanical condition classification. The report quality assessment module is used to assess the defect detection rate and false alarm rate of the initial inspection report. The link backtracking and optimization instruction generation module is used to backtrack the execution link of the structured acquisition scheme to locate and analyze the defective link when the initial inspection report fails to pass the defect detection rate and false alarm rate assessment, and generate an optimized acquisition instruction for the analyzed defective link. The knowledge base update module is used to update the historical inspection case library for task decomposition or the defect feature knowledge base for multimodal fusion analysis according to the optimized collection instructions. The iterative execution and output module is used to re-execute the steps from generating the inspection task context to generating the initial inspection report based on the updated historical inspection case library or defect feature knowledge base, until the initial inspection report passes the defect detection rate and false alarm rate evaluation, and outputs the final cable laying status inspection report and risk warning information.

Images