An unmanned aerial vehicle inspection method, device and system

Abstract

The application discloses a kind of unmanned aerial vehicle inspection method, device and system, it is related to unmanned aerial vehicle inspection technical field, comprising: when unmanned aerial vehicle inspection platform reaches any inspection waypoint of three-dimensional inspection route, according to the data acquisition strategy of inspection equipment Multi-angle synchronous acquisition of multimodal original inspection data;In any angle, after the data preprocessing of original inspection data is carried out and the corresponding target inspection data is constructed, the single-angle defect detection result is determined by detection model;According to the image data clustering of single-angle defect detection result to target inspection data forms defect event cluster;Determine the abnormal risk level of inspection equipment, combine single-angle defect detection result, the cluster label of defect event cluster and image abstract construct inspection abstract report;Based on the uploading mode of abnormal risk level and network transmission performance determined data is uploaded to cloud management platform;To next inspection waypoint executes inspection task until inspection is completed.Based on the above scheme, the reliability of unmanned aerial vehicle inspection is improved.

Description

Technical Field

[0001] This invention relates to the field of drone inspection technology, and in particular to a drone inspection method, apparatus and system. Background Technology

[0002] In recent years, drones have been widely used in inspection tasks due to their advantages such as maneuverability, comprehensive field of view, operational safety, and controllable cost. However, existing drone inspection methods often directly upload large amounts of collected image and sensor data to cloud services for analysis and processing. In remote mountainous areas and suburbs where network signals are weak and bandwidth is insufficient, slow uploads and interruptions are common problems, making it impossible to achieve real-time transmission and rapid reporting of critical anomalies. Furthermore, the same equipment defect may be repeatedly identified and trigger multiple independent alarm messages during multiple inspections, resulting in a large amount of redundant alarm information. High-risk faults are easily buried, severely reducing the efficiency of operation and maintenance and the reliability of alarms. Therefore, it is urgent to improve the reliability of drone inspections. Summary of the Invention

[0003] This invention provides a method, apparatus, and system for drone inspection, which improves the reliability of drone inspection.

[0004] The first aspect of this invention provides a method for unmanned aerial vehicle (UAV) inspection, comprising:

[0005] The inspection route is planned according to the inspection task, and the inspection task is carried out along the three-dimensional inspection route by the UAV inspection platform.

[0006] When the UAV inspection platform reaches any inspection waypoint of the three-dimensional inspection route, it synchronously collects multi-modal raw inspection data from multiple angles according to the data acquisition strategy associated with the inspection equipment.

[0007] At any of the aforementioned angles, after spatiotemporal alignment and data cleaning of the original inspection data to construct the corresponding target inspection data, the single-angle defect detection result is determined by target detection and instance segmentation through the detection model.

[0008] Based on the single-angle defect detection results, the image data of the target inspection data is clustered by defect category to form defect event clusters;

[0009] The abnormal risk level of the inspection equipment is determined based on the target inspection data and the single-angle defect detection results, and an inspection summary report of the inspection equipment is constructed by combining the single-angle defect detection results, the cluster labels of the defect event clusters and the image summary.

[0010] Based on the abnormal risk level and network transmission performance, the upload mode is determined, and the target inspection data and inspection summary report are uploaded to the cloud management platform.

[0011] If the three-dimensional inspection route is not completed, the UAV inspection platform will proceed along the three-dimensional inspection route to the next inspection point to perform the inspection task until the inspection is completed.

[0012] Optionally, determining the abnormal risk level of the inspection equipment based on the target inspection data and the single-angle defect detection result includes:

[0013] For any of the aforementioned angles, the abnormal value of the infrared temperature difference of the equipment is determined based on the infrared thermal image of the equipment in the target inspection data, and the abnormal values ​​of the environmental temperature and humidity and the abnormal amplitude of the environmental vibration are determined based on the environmental temperature and humidity data and the environmental vibration data in the target inspection data.

[0014] The corresponding single-angle equipment anomaly score is determined by using the constant value of the infrared temperature difference of the equipment associated with each angle, the abnormal value of the ambient temperature and humidity, the abnormal amplitude of the ambient vibration, the confidence level of the single-angle defect detection and the area of ​​the single-angle equipment contour region in the single-angle defect detection results.

[0015] The overall anomaly score of the equipment is calculated by combining the anomaly scores of each single-angle device, and the anomaly risk level of the inspection equipment is determined by combining the equipment importance level and historical defect change trend of the inspection equipment.

[0016] Optionally, the process of determining the image summary of the defect event cluster includes:

[0017] Calculate a comprehensive image score for the RGB image and infrared thermal image of the device associated with any of the defect event clusters; wherein the comprehensive image score includes a weighted sum of image sharpness, the proportion of the area of ​​the device outline region in the image at a single angle, the maximum infrared temperature difference of the device, and the integrity of the viewing angle;

[0018] The device's RGB image and infrared thermal image with the highest overall image score are used as the device's RGB image summary and infrared thermal image summary for the defect event cluster.

[0019] Optionally, the process of determining the upload mode includes:

[0020] When the abnormal risk level is high, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be the parallel upload of the full inspection summary report and the target inspection data. If the network transmission performance is less than the performance threshold, the upload mode is determined to be the priority upload of the full inspection summary report and the local caching of the target inspection data until the network transmission performance is greater than or equal to the performance threshold before uploading.

[0021] When the abnormal risk level is medium, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be the full inspection summary report and the target inspection data being uploaded in a queue. If the network transmission performance is less than the performance threshold, the upload mode is determined to be the lightweight inspection summary report being uploaded first, and the target inspection data being cached locally until the network transmission performance is greater than or equal to the performance threshold before being uploaded.

[0022] When the abnormal risk level is low, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be local caching of the compressed inspection summary report and the target inspection data until the inspection task is completed and then uploaded in parallel. If the network transmission performance is less than the performance threshold, the upload mode is determined to be local caching of the compressed inspection summary report and the target inspection data until the inspection task is completed, with the compressed inspection summary report uploaded first, and the target inspection data uploaded when the network transmission performance is greater than or equal to the performance threshold.

[0023] Optionally, the step of planning the three-dimensional inspection route according to the inspection task includes:

[0024] Based on the equipment inspection information and UAV flight performance information associated with the inspection task, construct the route constraints and route cost function;

[0025] Under the constraints of the route constraints, the route cost function is optimized and solved to determine the three-dimensional inspection route associated with the inspection task planning.

[0026] The route constraints include total route length constraints, estimated route energy consumption constraints, and route flight altitude constraints. The route cost function includes total route length cost, route flight altitude cost, estimated route energy consumption cost, and route flight risk cost.

[0027] Optionally, it also includes:

[0028] The cloud management platform verifies the received target inspection data and inspection summary report, issues real-time warnings, summarizes the inspection summary report, and updates the data collection strategy and detection model.

[0029] A second aspect of the present invention provides a drone inspection device, comprising:

[0030] The route generation module is used to plan the three-dimensional inspection route according to the inspection task, and to execute the inspection task along the three-dimensional inspection route through the UAV inspection platform.

[0031] The data acquisition module is used to synchronously acquire multi-modal raw inspection data from multiple angles according to the data acquisition strategy associated with the inspection equipment when the UAV inspection platform reaches any inspection waypoint of the three-dimensional inspection route.

[0032] Anomaly identification module is used to perform spatiotemporal alignment and data cleaning on the original inspection data at any angle to construct the corresponding target inspection data, and then use the detection model to perform target detection and instance segmentation to determine the single-angle defect detection result.

[0033] The event aggregation module is used to cluster the image data of the target inspection data according to the single-angle defect detection results to form defect event clusters.

[0034] The report generation module is used to determine the abnormal risk level of the inspection equipment based on the target inspection data and the single-angle defect detection results, and to construct an inspection summary report of the inspection equipment by combining the single-angle defect detection results, the cluster labels of the defect event clusters and the image summary.

[0035] The data reporting module is used to upload target inspection data and inspection summary reports to the cloud management platform based on the upload mode determined by the abnormal risk level and network transmission performance.

[0036] The inspection control module is used to, if the three-dimensional inspection route has not been completed, use the UAV inspection platform to reach the next inspection waypoint along the three-dimensional inspection route to perform the inspection task until the inspection is completed.

[0037] Optionally, it also includes a decision management module for:

[0038] The cloud management platform verifies the received target inspection data and inspection summary reports, issues real-time warnings, summarizes inspection summary reports, and updates data collection strategies and detection models.

[0039] The third aspect of this invention provides a drone inspection system, comprising a cloud management platform and a drone inspection platform connected by communication;

[0040] The cloud-based management platform is used to plan the assigned 3D inspection routes based on the inspection tasks.

[0041] The UAV inspection platform is used for: performing inspection tasks along the three-dimensional inspection route; upon reaching any inspection waypoint on the three-dimensional inspection route, simultaneously collecting multi-modal raw inspection data from multiple angles according to the data acquisition strategy associated with the inspection equipment; at any angle, performing spatiotemporal alignment and data cleaning on the raw inspection data to construct corresponding target inspection data, and then using a detection model to perform target detection and instance segmentation to determine the single-angle defect detection result; clustering the image data of the target inspection data according to the single-angle defect detection results to form defect event clusters; determining the abnormal risk level of the inspection equipment based on the target inspection data and the single-angle defect detection results, and constructing an inspection summary report of the inspection equipment by combining the single-angle defect detection results, the cluster labels of the defect event clusters, and the image summary; uploading the target inspection data and the inspection summary report to the cloud management platform based on the abnormal risk level and the upload mode determined by the network transmission performance; if the three-dimensional inspection route is not completed, proceeding along the three-dimensional inspection route to the next inspection waypoint to perform the inspection task until the inspection is completed.

[0042] Optionally, it also includes an operation and maintenance terminal that communicates with the cloud management platform;

[0043] The cloud management platform is also used to review the received target inspection data and inspection summary reports, issue real-time warnings, summarize inspection summary reports, and update data collection strategies and detection models.

[0044] The operation and maintenance terminal is used to visually display inspection summary reports and receive real-time alarms.

[0045] As can be seen from the above technical solutions, the present invention has the following advantages:

[0046] The above-mentioned solution of the present invention provides a drone inspection method, which realizes high-quality data collection of inspection equipment through a set data acquisition strategy, performs abnormal event-level multi-graph aggregation and summary generation of inspection equipment according to defect category to reduce redundant alarms, and realizes abnormal graded alarm and priority reporting according to the abnormal risk level and network transmission performance. Overall, it helps to make inspection safer, more efficient, lighter and more traceable, and improves the reliability of drone inspection. Attached Figure Description

[0047] 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.

[0048] Figure 1This is a flowchart of the steps of an unmanned aerial vehicle (UAV) inspection method provided in Embodiment 1 of the present invention;

[0049] Figure 2 This is a flowchart of the drone inspection framework provided in Embodiment 1 of the present invention;

[0050] Figure 3 This is a flowchart of the route planning and task-driven mechanism provided in Embodiment 1 of the present invention;

[0051] Figure 4 This is a structural block diagram of a drone inspection device provided in Embodiment 2 of the present invention;

[0052] Figure 5 This is a structural block diagram of a drone inspection device provided in Embodiment 3 of the present invention;

[0053] Figure 6 This is a schematic diagram of the structure of the UAV inspection platform provided in Embodiment 4 of the present invention. Detailed Implementation

[0054] This invention provides a method, apparatus, and system for drone inspection, which improves the reliability of drone inspection.

[0055] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0056] Please see Figures 1 to 3 The UAV inspection method provided in Embodiment 1 of this invention can be applied to scenarios such as communication base stations, substations, industrial parks, dikes, and ancient buildings, where automated inspection is conducted using UAVs. The method includes:

[0057] Step 101: Plan the three-dimensional inspection route according to the inspection task, and execute the inspection task along the three-dimensional inspection route through the UAV inspection platform.

[0058] It should be noted that, in response to the received inspection task, the system can retrieve equipment inspection information stored in the operation and maintenance database by parsing the inspection task. This equipment inspection information refers to information reflecting the structure and environment of the inspected equipment, such as, but not limited to, inspection waypoint coordinates, equipment dimensions, distribution of surrounding obstacles, and no-fly zone boundaries. In this embodiment, a drone inspection platform, primarily equipped with a drone, is used to execute the inspection task. Simultaneously, it acquires drone flight performance information related to the drone inspection platform for the task to be performed. This drone flight performance information refers to a set of quantitative parameters characterizing the drone's own flight capabilities, such as, but not limited to, remaining battery power, endurance limit, safe flight altitude (e.g., minimum and maximum flight altitude), maximum flight speed, and payload. Combining the equipment inspection information and the drone flight performance information, a three-dimensional inspection route adapted to the inspection task can be generated. Safe obstacle avoidance buffer zones can also be set around the inspection equipment and near obstacles to ensure flight operation safety. The drone inspection platform then proceeds to each inspection waypoint according to the three-dimensional inspection route to carry out the inspection work.

[0059] In one specific implementation of this method, the three-dimensional inspection route is planned according to the inspection task, including:

[0060] Based on the equipment inspection information and UAV flight performance information associated with the inspection task, construct the route constraints and route cost function;

[0061] The route cost function is optimized and solved under the constraints of the route constraint condition to determine the three-dimensional inspection route associated with the inspection task planning.

[0062] The route constraints include total route length constraints, estimated route energy consumption constraints, and route flight altitude constraints. The route cost function includes total route length cost, route flight altitude cost, estimated route energy consumption cost, and route flight risk cost.

[0063] It should be noted that, in one feasible implementation, this embodiment generates a 3D inspection route through task-driven constraint modeling and multi-objective optimization methods;

[0064] The route cost function, constructed as the optimization objective, can be specifically expressed as:

[0065] ;

[0066] In the formula, Indicates the three-dimensional inspection route. Indicates the cost of the flight route. This represents the total cost of the route length. This represents the flight altitude cost (or flight altitude constraint) for the flight route. This indicates the estimated energy cost of the flight route. This indicates the cost of flight route risks (including but not limited to constraints such as obstacles and no-fly zones). This represents the weighting coefficients for different cost items (used to balance factors such as route length, flight altitude, energy consumption, and flight safety). It is understood that the specific implementation formulas for each cost can be found in existing technologies and will not be elaborated here. The aim is to satisfy the constraints of relevant factors on route planning and to simultaneously optimize in the direction of maximization or minimization.

[0067] The constructed route constraints include total route length constraints, estimated route energy consumption constraints, and route flight altitude constraints, which can be specifically expressed as follows:

[0068] ;

[0069] In the formula, Indicates the total length of the route. This indicates the estimated energy consumption of the flight route. Indicates the minimum flight altitude for the route. This indicates the maximum flight altitude along the flight path.

[0070] It is understood that in some implementations, other algorithmic processes can also be used to achieve route planning. For example, heuristic search algorithms, graph search algorithms, or sampling-based path planning methods can be used to generate candidate inspection paths; integer programming, constraint programming, or other optimization algorithms can be used to determine the constraint satisfaction and select the optimal solution for the route; reinforcement learning and deep reinforcement learning methods can be used to achieve policy learning and dynamic adaptive optimization of the inspection route; the composition of the objective function or the constraints can be adjusted according to actual needs, such as adjusting the weights or expanding the structure of constraints on route length, flight altitude, energy consumption, and obstacle avoidance risk; and online updating and dynamic replanning of the inspection route can be achieved by combining real-time obstacle changes and remaining battery status. For details, please refer to existing inspection route planning modeling and path optimization mechanisms. This embodiment does not impose any limitations on this.

[0071] Step 102: When the UAV inspection platform reaches any inspection waypoint on the three-dimensional inspection route, it synchronously collects multi-modal raw inspection data from multiple angles according to the data acquisition strategy associated with the inspection equipment.

[0072] It should be noted that the inspection waypoint is the specific location where the inspection equipment is inspected during the inspection mission. After the UAV inspection platform takes off from the starting point of the flight path, it can obtain the UAV's attitude and dynamic obstacle avoidance in real time through RTK-GPS and IMU during the cruise flight. That is, it continuously perceives the surrounding environment during the flight and judges in real time whether obstacles such as trees, buildings, and other UAVs are detected. If an obstacle is detected, it immediately performs local dynamic adjustments to the flight path to avoid the obstacle and continue to the next inspection waypoint to continue the mission. If no obstacle is detected, it further judges whether it has reached the preset inspection waypoint. It can be understood that the specific obstacle judgment and local flight path adjustment and other dynamic obstacle avoidance strategies of the UAV can be referred to in the existing technology.

[0073] If it is determined that the UAV inspection platform has not reached the preset inspection waypoint on the 3D inspection route, the UAV inspection platform is controlled to continue cruising until it reaches the inspection waypoint. In a preferred implementation of this embodiment, when constructing the 3D inspection route, a data acquisition strategy is also customized according to the inspection equipment corresponding to each inspection waypoint. The data acquisition strategy refers to the set of acquisition execution requirements that match the type of inspection equipment, including but not limited to UAV acquisition pose, gimbal parameters (such as angles), and sensor acquisition parameters (such as camera shooting parameters and acquisition modes). When the UAV inspection platform reaches the inspection waypoint and the flight pose and gimbal angle meet the data acquisition strategy associated with the inspection equipment, multimodal synchronous data acquisition is triggered to obtain the original inspection data. Each set of inspection data corresponds to a precise spatial position and flight state to achieve deep collaboration between flight control and data acquisition, and the inspection equipment is acquired from multiple angles to obtain original inspection data from multiple angles.

[0074] Step 103: At any angle, after performing spatiotemporal alignment and data cleaning on the original inspection data to construct the corresponding target inspection data, target detection and instance segmentation are performed through the detection model to determine the single-angle defect detection result.

[0075] It should be noted that, in order to enhance data correlation and usability and improve positioning, the raw inspection data collected from each angle is time-aligned and spatially aligned before data cleaning, thereby obtaining the corresponding target inspection data. This helps to improve the accuracy of subsequent defect identification and reduce bandwidth consumption for invalid data transmission. For example, in one feasible approach, time-space alignment can be achieved by adding a unified timestamp and spatial coordinate mapping. Data cleaning aims to construct high-quality inspection data, and for example, it can include, but is not limited to, data processing such as image denoising, distortion correction, outlier filtering, duplicate data removal, and data augmentation. This embodiment does not limit this.

[0076] After determining the target inspection data, the defect detection input features are constructed based on the target inspection data and input into the detection model used for equipment defect detection to perform target detection and instance segmentation, and then output single-angle defect detection results. The single-angle defect detection results refer to the defect detection results obtained by inputting the model constructed based on multimodal inspection data from a single acquisition angle, including but not limited to single-angle defect category, single-angle detection box coordinates, single-angle defect detection confidence, and single-angle equipment contour area.

[0077] In one specific implementation of this embodiment, the raw inspection data / target inspection data includes the equipment RGB image, the equipment infrared thermal image, the ambient temperature and humidity data, and the ambient vibration data.

[0078] In a more specific implementation of this embodiment, after feature encoding and linear mapping of the equipment RGB image, equipment infrared thermal image, environmental temperature and humidity data, environmental vibration data, and UAV-collected pose data in the target inspection data, feature fusion (such as stitching, weighted fusion, or cross-modal fusion based on attention mechanism) is performed to construct defect detection input features, so as to automatically identify various defects such as equipment hot spots, cracks, corrosion, loose screws, component tilting, abnormal vibration, and temperature and humidity exceeding thresholds through the detection model. The specific implementation of the detection model can refer to the prior art, such as the YOLO series model, and this embodiment does not limit it.

[0079] Step 104: Based on the defect detection results of each single angle, cluster the image data of the target inspection data according to the defect category to form defect event clusters.

[0080] It should be noted that, in order to avoid the problem of repeated reporting of anomalies and the proliferation of redundant alarms that interfere with operation and maintenance decisions, this embodiment clusters the acquisition results of the same inspection equipment and the same anomaly event from multiple angles and multiple modes. That is, the single-angle defect category is used as the cluster label, the single-angle defect category is extracted from each single-angle defect detection result, and the image data of the target inspection data belonging to each single-angle defect detection category is clustered according to the single-angle defect detection category to form a defect event cluster, which is also merged into a single alarm event.

[0081] Step 105: Determine the abnormal risk level of the inspection equipment based on the target inspection data and the single-angle defect detection results, and construct an inspection summary report of the inspection equipment by combining the single-angle defect detection results, the cluster labels of the defect event clusters and the image summary.

[0082] It should be noted that the anomaly risk level reflects the severity of the current defect anomaly of the inspection equipment. In order to provide differentiated warnings for different defect detection results, this embodiment determines the corresponding anomaly risk level by combining the target inspection data of the inspection equipment with the single-angle defect detection results. Meanwhile, the inspection summary report refers to the condensed and structured presentation of the results collected and analyzed by the inspection equipment, and the image summary refers to the key image that can representatively reflect the defect category. In this embodiment, the inspection summary report of the inspection equipment can be constructed based on the single-angle defect detection results, the cluster labels of the defect event clusters, and the image summary.

[0083] In one specific implementation of this method, the abnormal risk level of the inspection equipment is determined based on the target inspection data and the single-angle defect detection results, including:

[0084] For any angle, determine the abnormal value of infrared temperature difference of the equipment based on the infrared thermal image of the equipment in the target inspection data, and determine the abnormal values ​​of environmental temperature and humidity and the abnormal amplitude of environmental vibration based on the environmental temperature and humidity data and environmental vibration data in the target inspection data.

[0085] The corresponding single-angle equipment anomaly score is determined by using the constant value of the infrared temperature difference of the equipment at each angle, the abnormal value of the ambient temperature and humidity, the abnormal amplitude of the ambient vibration, the confidence level of the single-angle defect detection and the area of ​​the single-angle equipment contour region in the single-angle defect detection results.

[0086] The overall anomaly score of the equipment is calculated by combining the anomaly scores of each single-angle device, and the anomaly risk level of the inspected equipment is determined by combining the equipment importance level and historical defect change trend of the inspected equipment.

[0087] It should be noted that the process of determining the level of abnormal risk can be expressed as follows:

[0088] ;

[0089] In the formula, Indicates the level of abnormal risk. This indicates the overall anomaly score of the equipment. Indicates the importance level of the equipment. Indicates the trend of historical defects. This represents a model for determining the level of abnormal risk.

[0090] The process for determining the overall equipment anomaly score includes:

[0091] ;

[0092] In the formula, Indicates the total number of angles collected. Indicates the first A comprehensive anomaly score for single-angle equipment from each angle. This indicates the constant value of the infrared temperature difference of the equipment. This represents the area of ​​the outline region of a single-angle device. Indicates the abnormal amplitude of environmental vibration. Indicates the confidence level of single-angle defect detection. This indicates abnormal values ​​for ambient temperature and humidity. The weight coefficients represent different anomaly scoring items. It can be understood that the abnormal values ​​of the equipment infrared temperature difference, the abnormal amplitude of environmental vibration, and the abnormal values ​​of environmental temperature and humidity represent the degree of abnormality that deviates from the normal working range of the inspection equipment. The results can be judged based on the normalized deviation or threshold between the equipment and the corresponding normal values. This embodiment does not impose any restrictions on this.

[0093] In one specific embodiment of this implementation, the process of determining the image summary of the defect event cluster includes:

[0094] For any cluster of defect events, calculate a comprehensive image score for the associated RGB image and infrared thermal image of the equipment. The comprehensive image score includes a weighted sum of image sharpness, the proportion of the equipment outline area in the image at a single angle, the maximum infrared temperature difference of the equipment, and the integrity of the viewing angle.

[0095] The device's RGB image and infrared thermal image with the highest overall image score are used as the device's RGB image summary and infrared thermal image summary for the defect event cluster.

[0096] It should be noted that in this embodiment, corresponding image summaries are selected from the image data belonging to each defect event cluster based on the comprehensive image score, thereby retaining the key images and key thermal imaging data that most intuitively reflect the equipment defect status and have the most operational and maintenance reference value. Here, image clarity refers to an indicator value that measures the degree of detail resolution of an image, which can be determined according to an image clarity evaluation algorithm; the image proportion of the equipment outline area refers to the proportion of the pixel area of ​​the inspected equipment segmented from the instance to the entire image, used to reflect the degree of anomaly significance; the maximum equipment infrared temperature difference refers to the difference between the highest and lowest equipment temperatures in the equipment infrared thermal imaging image; and visual integrity refers to an indicator value that evaluates the integrity of the equipment display, for example, it can be... The score is determined by a weighted sum of the device outline area area as a percentage of the total area and the shooting angle score. The device outline area area percentage refers to the ratio of the device outline area (i.e., the pixel area of ​​the inspection device segmented from the instance) to the normal, complete area of ​​the device. The shooting angle score evaluates the quality of the shooting angle relative to the surface being photographed, which can be determined by combining the drone's position and the gimbal angle. The shooting angle score is relatively high when shooting from the front and relatively low when shooting from the side or with obstruction. Based on the comprehensive image score, the most representative image summary can be selected according to the principles of highest clarity, most obvious abnormal areas, most significant infrared temperature difference, and most complete viewpoint, which helps to improve the information expression ability of multi-image summaries.

[0097] Step 106: Based on the upload mode determined by the abnormal risk level and network transmission performance, upload the target inspection data and inspection summary report to the cloud management platform.

[0098] It should be noted that the cloud management platform is usually deployed on a ground station server or cloud server. During the inspection, the inspection-related content obtained by the drone inspection platform will be uploaded to the cloud management platform for structured storage and archiving, in-depth data mining, and scheduling warnings. The upload mode is designed to define the way and timing of data upload.

[0099] This embodiment sets up three risk levels for anomalies: high, medium, and low, ranked from highest to lowest (the specific level division can be adjusted adaptively). Differentiated uploading based on anomaly level allows for priority handling of high-risk faults and reasonable scheduling of low-risk information. Simultaneously, real-time monitoring of network transmission performance enables adaptive uploading based on communication status, significantly reducing communication pressure while ensuring the real-time availability of critical information. This helps avoid the problems of high bandwidth consumption and overwhelming critical alarms caused by indiscriminate reporting of all anomalies. In one specific implementation, the process of determining the uploading mode includes:

[0100] When the abnormal risk level is high, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be the parallel upload of the full inspection summary report and the target inspection data. If the network transmission performance is less than the performance threshold, the upload mode is determined to be the priority upload of the full inspection summary report and the local cache of the target inspection data until the network transmission performance is greater than or equal to the performance threshold before uploading.

[0101] When the abnormal risk level is medium, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be the full inspection summary report and the target inspection data being uploaded in a queue. If the network transmission performance is less than the performance threshold, the upload mode is determined to be the lightweight inspection summary report being uploaded first, and the target inspection data being cached locally until the network transmission performance is greater than or equal to the performance threshold before being uploaded.

[0102] When the abnormal risk level is low, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be local caching of the compressed inspection summary report and the target inspection data until the inspection task is completed and then uploaded in parallel. If the network transmission performance is less than the performance threshold, the upload mode is determined to be local caching of the compressed inspection summary report and the target inspection data until the inspection task is completed, with the compressed inspection summary report uploaded first, and the target inspection data uploaded when the network transmission performance is greater than or equal to the performance threshold.

[0103] It is understandable that the full inspection summary report, lightweight inspection summary report, and compressed inspection summary report are designed to represent different inspection summary reports with varying data volumes, from large to small. For example, the full inspection summary report can be the inspection summary report directly generated in step 105, retaining relatively complete inspection evidence, such as equipment number, inspection time, anomaly type (i.e., cluster label), anomaly risk level, comprehensive equipment anomaly score, equipment RGB image summary, equipment infrared thermal image summary, single-angle detection box coordinates associated with the equipment RGB image summary, sensor data change information (such as change curves of environmental temperature and humidity data and environmental vibration data), and automatically generated text results. The full inspection summary report is used for remote fault diagnosis and rapid decision-making. The lightweight inspection summary report, while ensuring maintenance personnel can identify the type and location of anomalies, simplifies the data from the full inspection summary report, retaining only core information such as device number, inspection time, anomaly type, anomaly risk level, device RGB image summary, the coordinates of the single-angle detection frame associated with the device RGB image summary, and textual conclusions, thus reducing data transmission pressure. The compressed summary further compresses the data, retaining only the device number, inspection time, anomaly type, textual conclusions, and the compressed device RGB image summary for subsequent batch archiving and trend analysis. Specific report content can be adaptively adjusted as needed.

[0104] In one feasible approach, network transmission performance can be evaluated by signal strength and bandwidth status, and judged by comparison with corresponding performance thresholds. For example, network transmission performance less than the performance threshold indicates weak network signal and insufficient bandwidth, while network transmission performance greater than or equal to the performance threshold indicates strong network signal and sufficient bandwidth.

[0105] Step 107: If the three-dimensional inspection route has not been completed, the UAV inspection platform will proceed along the three-dimensional inspection route to the next inspection point to perform the inspection task until the inspection is completed.

[0106] It should be noted that after a single waypoint inspection is completed, it is determined whether all waypoints along the three-dimensional inspection route have been inspected. If not, the UAV inspection platform is controlled to move to the next waypoint to continue the task. If the inspection is completed, the inspection task ends.

[0107] In a preferred implementation of this embodiment, an inspection log chain can also be constructed, which, for example, includes time-series recorded information such as the collection trigger time, UAV pose, data number, single-angle defect detection result, abnormal risk level, and upload status, so as to support the playback and review of the inspection process by time, device, and abnormality type.

[0108] In one specific embodiment of this implementation, it also includes:

[0109] The cloud management platform reviews the received target inspection data and inspection summary reports, issues real-time warnings, summarizes inspection summary reports, and updates data collection strategies and detection models.

[0110] It should be noted that the cloud management platform allows for long-term storage and management of target inspection data, inspection summary reports, and inspection log chains. Upon receiving target inspection data and inspection summary reports, the platform can review and perform secondary verification of anomaly identification results. It also combines historical inspection data from the equipment for trend analysis and comparison to confirm the authenticity and severity of defects, mitigating potential misjudgments at the edge and improving alarm accuracy and reliability. After review, a complete structured inspection report is generated based on the review results, and alarm information is pushed to maintenance personnel according to the anomaly risk level. It supports multiple push methods such as SMS, APP notifications, and pop-ups on the operation and maintenance end to ensure that operation and maintenance personnel can obtain fault information in a timely manner and carry out handling work. At the same time, it can further perform historical data updates and inspection strategy optimization. Among them, the inspection strategy optimization can provide feedback optimization for route planning strategies, data collection strategies, detection model updates, alarm classification strategies, upload strategies, and image summary generation strategies. In addition, it can also replay the entire inspection process based on the inspection log chain to achieve process reconstruction and verification. This forms a collaborative working mechanism of "rapid edge judgment, cloud-based fine review, model synchronization, and closed-loop verification".

[0111] In this embodiment of the invention, high-quality data collection of inspection equipment is achieved through a set data acquisition strategy. The inspection equipment is subjected to abnormal event-level multi-graph aggregation and summary generation according to the defect category to reduce redundant alarms. The abnormal risk level and network transmission performance are used to realize abnormal graded alarms and priority reporting. Overall, it helps to make inspection safer, more efficient, lighter and more traceable, and improves the reliability of UAV inspection.

[0112] Please see Figure 4 The second embodiment of the present invention provides a drone inspection device, comprising:

[0113] The route generation module 401 is used to plan the three-dimensional inspection route according to the inspection task, and to execute the inspection task along the three-dimensional inspection route through the UAV inspection platform.

[0114] The data acquisition module 402 is used to synchronously acquire multi-modal raw inspection data from multiple angles according to the data acquisition strategy associated with the inspection equipment when the UAV inspection platform reaches any inspection waypoint of the three-dimensional inspection route.

[0115] The anomaly identification module 403 is used to perform spatiotemporal alignment and data cleaning on the original inspection data at any angle to construct the corresponding target inspection data, and then use the detection model to perform target detection and instance segmentation to determine the single-angle defect detection result.

[0116] The event aggregation module 404 is used to cluster the image data of the target inspection data according to the defect detection results of each single angle, forming a defect event cluster;

[0117] The report generation module 405 is used to determine the abnormal risk level of the inspection equipment based on the target inspection data and the single-angle defect detection results, and to construct an inspection summary report of the inspection equipment by combining the single-angle defect detection results, the cluster labels of the defect event clusters and the image summary.

[0118] The data reporting module 406 is used to upload target inspection data and inspection summary reports to the cloud management platform based on the upload mode determined by the anomaly risk level and network transmission performance.

[0119] The inspection control module 407 is used to move along the three-dimensional inspection route to the next inspection waypoint to perform the inspection task if the three-dimensional inspection route has not been completed, until the inspection is completed.

[0120] In one specific embodiment of this example, a decision management module is further included, used for:

[0121] The cloud management platform reviews the received target inspection data and inspection summary reports, issues real-time warnings, summarizes inspection summary reports, and updates data collection strategies and detection models.

[0122] Please see Figure 5 and Figure 6 The third embodiment of the present invention provides a drone inspection system, comprising:

[0123] This includes a cloud-based management platform for communication connectivity and a drone inspection platform;

[0124] The cloud management platform 501 is used to: plan the assigned 3D inspection route according to the inspection task;

[0125] The UAV inspection platform 502 is used for: performing inspection tasks along a 3D inspection route; upon reaching any inspection waypoint along the 3D inspection route, simultaneously collecting multi-modal raw inspection data from multiple angles according to the data acquisition strategy associated with the inspection equipment; at any angle, performing spatiotemporal alignment and data cleaning on the raw inspection data to construct corresponding target inspection data, and then using a detection model to perform target detection and instance segmentation to determine the single-angle defect detection result; clustering the image data of the target inspection data according to the single-angle defect detection results to form defect event clusters; determining the abnormal risk level of the inspection equipment based on the target inspection data and the single-angle defect detection results, and constructing an inspection summary report of the inspection equipment by combining the single-angle defect detection results, the cluster labels of the defect event clusters, and the image summary; uploading the target inspection data and the inspection summary report to the cloud management platform based on the upload mode determined by the abnormal risk level and network transmission performance; if the 3D inspection route is not completed, proceeding along the 3D inspection route to the next inspection waypoint to perform the inspection task until the inspection is completed.

[0126] In one specific embodiment of this implementation, it also includes an operation and maintenance terminal 503 that is communicatively connected to the cloud management platform;

[0127] The cloud management platform is also used to review the received target inspection data and inspection summary reports, issue real-time warnings, summarize inspection summary reports, and update data collection strategies and detection models.

[0128] The operation and maintenance terminal is used to visually display inspection summary reports and receive real-time alarms.

[0129] It should be noted that the cloud management platform is the core of the entire system's scheduling and management. Deployed on ground station servers or cloud servers, it is used to realize global task management, route planning, data processing, and strategy optimization. Specifically, it can be configured as follows:

[0130] Inspection task management module: can be used to access the operation and maintenance database to obtain equipment inspection information, receive inspection task instructions issued by users, support batch import, editing and priority configuration of multiple inspection tasks, and realize dynamic allocation and parallel scheduling of multiple drone tasks;

[0131] Flight path planning module: can be used to plan the three-dimensional inspection route based on the equipment inspection information and UAV flight performance information according to the inspection task; at the same time, it can set up obstacle avoidance buffer zones for each inspection point in combination with the distribution of surrounding obstacles, bind the flight path to the inspection point, and preset data collection strategies such as shooting angle and acquisition frequency;

[0132] Anomaly analysis module: It can be used to receive data uploaded by drone inspections, combine it with historical inspection data for review and analysis, verify the accuracy of edge anomaly identification results, and avoid misjudgment and missed judgment;

[0133] Report generation module: It can be used to automatically summarize and generate standardized inspection reports based on anomaly analysis results and multi-figure summary files, and supports structured export and visualization.

[0134] Historical database: can be used to store the raw data, processing results, summary files, alarm records and inspection reports of all inspection tasks, build health status files of inspection equipment, and support the query, comparison and trend analysis of historical data;

[0135] Strategy optimization module: It can be used to optimize strategies such as flight path planning, data acquisition, alarm classification, upload, and image summary generation based on historical inspection data, alarm statistics, and cloud review feedback. It also supports sending model update commands to the UAV inspection platform to achieve iterative optimization of system performance.

[0136] The drone inspection platform is deployed on the drone itself. It uses the Robot Operating System (ROS) as its core to achieve flight control, data acquisition, edge computing, and local control. It serves as the execution platform for inspection tasks. Specifically, it can be configured as follows:

[0137] Flight control management module (including GPS / IMU): As the core of the UAV's flight controller, it achieves high-precision positioning and attitude perception based on GPS / IMU, receives the three-dimensional inspection route sent from the cloud, combines real-time obstacle information to realize dynamic route adjustment, and outputs flight attitude information.

[0138] Task execution module (flight path execution): Based on the ROS flight control interface node, it parses the inspection flight path and data acquisition strategy issued by the cloud and controls the UAV to fly along the planned flight path; when the UAV reaches the preset inspection waypoint, it links with the flight control management module through a high-precision pose trigger to automatically trigger the acquisition actions of the acquisition devices such as RGB camera, infrared camera and environmental sensor according to the current pose, so as to realize the linkage control of flight attitude, gimbal angle and acquisition parameters; in a specific implementation, the high-precision pose trigger is integrated on both sides of the fuselage;

[0139] Sensor Module: Includes various heterogeneous sensors to achieve multi-dimensional data acquisition; specifically: an RGB camera for acquiring visible light image data and recording the appearance of the equipment; an infrared camera for acquiring thermal imaging data to detect abnormally heated components; and environmental multi-parameter sensors, including but not limited to temperature and humidity sensors for acquiring temperature and humidity data of the environment where the inspection equipment is located, and vibration sensors for monitoring the vibration of the environment where the inspection equipment is located (to help determine structural stability); the above-mentioned sensing data can be continuously published through the ROS Topic mechanism to provide multimodal input to the system; in a specific implementation, the dual light sensor (RGB / IR camera) can be mounted on the top of the head, and the environmental multi-parameter sensors can be distributed on both sides of the body;

[0140] Edge computing modules (such as embedded GPUs): provide high-performance computing capabilities for local preprocessing and preliminary analysis of data, reducing backhaul bandwidth consumption. Specifically:

[0141] Data preprocessing: Perform spatiotemporal alignment and data cleaning to filter out low-quality data;

[0142] Preliminary detection: Target detection and instance segmentation are performed based on the detection model to identify defects and anomalies;

[0143] Summary generation: Aggregate multiple images of the same device or the same abnormal event, filter keyframes with high clarity and obvious abnormal features, generate multi-image summary files, and sort them according to the anomaly priority;

[0144] Tiered Alarms: Alarms are tiered based on the determined level of abnormal risk, triggering different upload modes;

[0145] Data storage: Used for local caching of inspection data, summary reports, inspection logs and model version information, etc. It supports temporary storage of data under communication-restricted conditions, and completes the retransmission after the network is restored.

[0146] Preferably, the drone in the drone inspection platform can adopt a multi-rotor platform architecture. This multi-rotor power configuration provides lift and maneuverability to the entire aircraft through symmetrically distributed propellers, supporting various flight modes required for inspection, such as hovering, fixed-point shooting, and flight path following, which can meet the requirements of inspection operations for stability, maneuverability, and safety. At the same time, the fuselage can also be equipped with auxiliary hardware such as microphones and antennas to support voice interaction and signal transmission.

[0147] The operation and maintenance terminal provides users with a visual interactive interface for inspection results, supporting multiple formats such as PC and mobile APP. Specifically, it can be used to: view inspection reports and multi-image summary files to intuitively understand the health status of inspected equipment; receive real-time alarm notifications, supporting multiple alarm methods such as SMS, APP push, and operation and maintenance terminal pop-ups, facilitating timely response by operation and maintenance personnel to high-risk anomalies.

[0148] The communication connection between the cloud management platform and the drone inspection platform, as well as between the cloud management platform and the operation and maintenance terminal, supports multiple communication methods such as 5G, LTE, and Wi-Fi to achieve data transmission.

[0149] Data upload: The data obtained from the edge drone inspection platform is uploaded in various modes, such as full upload, summary-first upload, or local caching, based on network transmission performance conditions such as communication bandwidth status. Data that was interrupted is also re-uploaded after the network is restored.

[0150] Task distribution: The cloud-generated 3D inspection route, data acquisition strategy, model update instructions and control instructions are distributed to the UAV inspection platform, supporting real-time response and breakpoint resume.

[0151] Data delivery: Transmitting inspection reports, alarms, etc. from the cloud to the operation and maintenance terminal.

[0152] By adopting a three-level collaborative architecture of "cloud-edge-device", the system realizes the closed-loop control of the entire process, including automated execution of inspection tasks, multimodal data acquisition and processing, anomaly identification and hierarchical alarm, and automatic summary generation and reporting. All parts work together to achieve efficient, reliable and low-redundancy drone inspection.

[0153] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0154] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, apparatuses, or modules, and may be electrical, mechanical, or other forms.

[0155] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0156] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0157] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for unmanned aerial vehicle (UAV) inspection, characterized in that, include: The inspection route is planned according to the inspection task, and the inspection task is carried out along the three-dimensional inspection route by the UAV inspection platform. When the UAV inspection platform reaches any inspection waypoint of the three-dimensional inspection route, it synchronously collects multi-modal raw inspection data from multiple angles according to the data acquisition strategy associated with the inspection equipment. At any of the aforementioned angles, after spatiotemporal alignment and data cleaning of the original inspection data to construct the corresponding target inspection data, the single-angle defect detection result is determined by target detection and instance segmentation through the detection model. Based on the single-angle defect detection results, the image data of the target inspection data is clustered by defect category to form defect event clusters; The abnormal risk level of the inspection equipment is determined based on the target inspection data and the single-angle defect detection results, and an inspection summary report of the inspection equipment is constructed by combining the single-angle defect detection results, the cluster labels of the defect event clusters and the image summary. Based on the abnormal risk level and network transmission performance, the upload mode is determined, and the target inspection data and inspection summary report are uploaded to the cloud management platform. If the three-dimensional inspection route is not completed, the UAV inspection platform will proceed along the three-dimensional inspection route to the next inspection point to perform the inspection task until the inspection is completed.

2. The UAV inspection method according to claim 1, characterized in that, The step of determining the abnormal risk level of the inspection equipment based on the target inspection data and the single-angle defect detection results includes: For any of the aforementioned angles, the abnormal value of the infrared temperature difference of the equipment is determined based on the infrared thermal image of the equipment in the target inspection data, and the abnormal values ​​of the environmental temperature and humidity and the abnormal amplitude of the environmental vibration are determined based on the environmental temperature and humidity data and the environmental vibration data in the target inspection data. The corresponding single-angle equipment anomaly score is determined by using the constant value of the infrared temperature difference of the equipment associated with each angle, the abnormal value of the ambient temperature and humidity, the abnormal amplitude of the ambient vibration, the confidence level of the single-angle defect detection and the area of ​​the single-angle equipment contour region in the single-angle defect detection results. The overall anomaly score of the equipment is calculated by combining the anomaly scores of each single-angle device, and the anomaly risk level of the inspection equipment is determined by combining the equipment importance level and historical defect change trend of the inspection equipment.

3. The UAV inspection method according to claim 1, characterized in that, The process of determining the image summary of the defect event cluster includes: Calculate a comprehensive image score for the RGB image and infrared thermal image of the device associated with any of the defect event clusters; wherein the comprehensive image score includes a weighted sum of image sharpness, the proportion of the area of ​​the device outline region in the image at a single angle, the maximum infrared temperature difference of the device, and the integrity of the viewing angle; The device's RGB image and infrared thermal image with the highest overall image score are used as the device's RGB image summary and infrared thermal image summary for the defect event cluster.

4. The UAV inspection method according to claim 1, characterized in that, The process of determining the upload mode includes: When the abnormal risk level is high, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be the parallel upload of the full inspection summary report and the target inspection data. If the network transmission performance is less than the performance threshold, the upload mode is determined to be the priority upload of the full inspection summary report and the local caching of the target inspection data until the network transmission performance is greater than or equal to the performance threshold before uploading. When the abnormal risk level is medium, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be the full inspection summary report and the target inspection data being uploaded in a queue. If the network transmission performance is less than the performance threshold, the upload mode is determined to be the lightweight inspection summary report being uploaded first, and the target inspection data being cached locally until the network transmission performance is greater than or equal to the performance threshold before being uploaded. When the abnormal risk level is low, if the network transmission performance is greater than or equal to the performance threshold, the upload mode is determined to be local caching of the compressed inspection summary report and the target inspection data until the inspection task is completed and then uploaded in parallel. If the network transmission performance is less than the performance threshold, the upload mode is determined to be local caching of the compressed inspection summary report and the target inspection data until the inspection task is completed, with the compressed inspection summary report uploaded first, and the target inspection data uploaded when the network transmission performance is greater than or equal to the performance threshold.

5. The UAV inspection method according to claim 1, characterized in that, The three-dimensional inspection route planned according to the inspection task includes: Based on the equipment inspection information and UAV flight performance information associated with the inspection task, construct the route constraints and route cost function; Under the constraints of the route constraints, the route cost function is optimized and solved to determine the three-dimensional inspection route associated with the inspection task planning. The route constraints include total route length constraints, estimated route energy consumption constraints, and route flight altitude constraints. The route cost function includes total route length cost, route flight altitude cost, estimated route energy consumption cost, and route flight risk cost.

6. The UAV inspection method according to claim 1, characterized in that, Also includes: The cloud management platform verifies the received target inspection data and inspection summary reports, issues real-time warnings, summarizes inspection summary reports, and updates data collection strategies and detection models.

7. A drone inspection device, characterized in that, include: The route generation module is used to plan the three-dimensional inspection route according to the inspection task, and to execute the inspection task along the three-dimensional inspection route through the UAV inspection platform. The data acquisition module is used to synchronously acquire multi-modal raw inspection data from multiple angles according to the data acquisition strategy associated with the inspection equipment when the UAV inspection platform reaches any inspection waypoint of the three-dimensional inspection route. Anomaly identification module is used to perform spatiotemporal alignment and data cleaning on the original inspection data at any angle to construct the corresponding target inspection data, and then use the detection model to perform target detection and instance segmentation to determine the single-angle defect detection result. The event aggregation module is used to cluster the image data of the target inspection data according to the single-angle defect detection results to form defect event clusters. The report generation module is used to determine the abnormal risk level of the inspection equipment based on the target inspection data and the single-angle defect detection results, and to construct an inspection summary report of the inspection equipment by combining the single-angle defect detection results, the cluster labels of the defect event clusters and the image summary. The data reporting module is used to upload target inspection data and inspection summary reports to the cloud management platform based on the upload mode determined by the abnormal risk level and network transmission performance. The inspection control module is used to, if the three-dimensional inspection route has not been completed, use the UAV inspection platform to reach the next inspection waypoint along the three-dimensional inspection route to perform the inspection task until the inspection is completed.

8. The UAV inspection device according to claim 7, characterized in that, It also includes a decision management module, used for: The cloud management platform verifies the received target inspection data and inspection summary reports, issues real-time warnings, summarizes inspection summary reports, and updates data collection strategies and detection models.

9. A drone inspection system, characterized in that, This includes a cloud-based management platform for communication connectivity and a drone inspection platform; The cloud-based management platform is used to plan the assigned 3D inspection routes based on the inspection tasks. The drone inspection platform is used to perform inspection tasks along the three-dimensional inspection route. When the inspection point of the three-dimensional inspection route is reached, multi-modal raw inspection data is collected synchronously from multiple angles according to the data acquisition strategy associated with the inspection equipment. At any of the aforementioned angles, after spatiotemporal alignment and data cleaning of the original inspection data to construct corresponding target inspection data, target detection and instance segmentation are performed using a detection model to determine the single-angle defect detection result; according to each of the single-angle defect detection results, the image data of the target inspection data is clustered into defect categories to form defect event clusters. The abnormal risk level of the inspection equipment is determined based on the target inspection data and the single-angle defect detection results, and an inspection summary report of the inspection equipment is constructed by combining the single-angle defect detection results, the cluster labels of the defect event clusters and the image summary. Based on the abnormal risk level and network transmission performance, the upload mode is determined, and the target inspection data and inspection summary report are uploaded to the cloud management platform. If the three-dimensional inspection route is not completed, the inspection task will be carried out along the three-dimensional inspection route to the next inspection waypoint until the inspection is completed.

10. The UAV inspection system according to claim 9, characterized in that, It also includes operation and maintenance terminals that communicate with the cloud management platform; The cloud management platform is also used to review the received target inspection data and inspection summary reports, issue real-time warnings, summarize inspection summary reports, and update data collection strategies and detection models. The operation and maintenance terminal is used to visually display inspection summary reports and receive real-time alarms.

Images