An unmanned aerial vehicle inspection method based on edge computing and an inspection system thereof
By constructing a two-tier architecture of UAV-borne edge nodes and field edge computing units, localized processing and collaborative computing of UAV inspection data are achieved, solving the problems of high data latency and poor security in traditional UAV inspection, improving the real-time performance and accuracy of inspection, and adapting to the needs of multiple scenarios.
Patent Information
- Application Number
- CN202610675462.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-15
- Publication Date
- 2026-11-06
AI Technical Summary
Existing drone inspection technology relies on public network transmission, resulting in high data latency and severe packet loss, making it impossible to achieve real-time identification and alarms. Furthermore, the onboard processing model has low accuracy and poor security, making it difficult to meet the high precision, real-time performance, and security requirements of industrial inspection.
A two-level edge architecture of UAV-borne edge nodes and field edge computing units is adopted to perform localized processing of inspection data. Through real-time preprocessing, lightweight coding and adaptive task offloading, localized data analysis and collaborative computing are realized to generate real-time inspection alarms.
It reduces data transmission latency, improves the accuracy and security of defect identification, meets the real-time and security requirements of industrial inspection, adapts to multiple scenarios, and enhances the system's versatility and adaptability.
Smart Images
Figure CN123047584A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of drone inspection technology, and in particular to a drone inspection method and system based on edge computing. Background Technology
[0002] With the improvement of industrial intelligence, drone inspection has been widely used in various fields such as power, oil and gas, and photovoltaics due to its advantages such as high operating efficiency, wide coverage, and no need for human close contact with high-risk areas. At present, most existing drone inspection technologies adopt the traditional architecture of "airborne data acquisition + cloud processing". This architecture has many technical defects and cannot meet the high precision, real-time performance, and high security requirements of industrial inspection.
[0003] On the one hand, traditional inspection architectures heavily rely on public network communication to transmit massive amounts of inspection data. However, drone inspection scenarios are mostly in areas with low public network coverage and unstable links, such as mountains, deserts, and remote factories. This results in high latency and severe packet loss in the return of inspection data, making it impossible to achieve real-time defect identification and alarms, which can easily lead to the expansion of hidden dangers. On the other hand, the centralized cloud processing mode requires all raw inspection data to be uploaded to the cloud, which not only increases the transmission power consumption of drones and shortens flight time, but also poses security risks of leakage and tampering of sensitive inspection data, making it difficult to meet the data compliance requirements of key industries.
[0004] In addition, some improvement solutions attempt to deploy lightweight processing models on the airborne end. However, due to the limitations of the load, computing power and power consumption of the UAV's airborne equipment, the lightweight models often suffer from low defect recognition accuracy and high false negative rate for small targets. Furthermore, the airborne end and field equipment lack an effective collaboration mechanism, resulting in low utilization of computing resources. They cannot simultaneously meet the requirements of real-time inspection, accuracy and long flight time, and are difficult to adapt to complex and diverse industrial inspection scenarios. Summary of the Invention
[0005] In view of this, the present invention aims to provide an edge computing-based UAV inspection method and system to solve or alleviate the technical problems existing in the prior art, and at least provide a beneficial option.
[0006] To address the aforementioned technical problems, this application adopts the following technical solution: providing a drone inspection method based on edge computing, comprising: Acquire inspection data collected by the UAV's onboard edge nodes; The inspection data is preprocessed in real time based on the UAV's onboard edge node. The UAV's onboard edge node generates a task unloading decision based on a preset inspection scenario and onboard computing power load; Based on the task offloading decision, some complex inspection and analysis tasks are diverted to the nearest on-site edge computing unit. Equipment defect detection results are obtained through collaborative computing between the UAV-borne edge node and the field edge computing unit. Real-time inspection alarm information is generated based on the equipment defect detection results.
[0007] As a further preferred embodiment of this technical solution: the acquisition of inspection data collected by the UAV's onboard edge node includes: Visible light image data is acquired using a multispectral camera mounted on a drone; Temperature field data of the equipment is collected using an infrared thermal imager mounted on a drone; Surface point cloud data is collected using a lidar device mounted on a drone; The airborne edge nodes perform timestamp alignment on the multi-source data to form a unified inspection data frame.
[0008] As a further preferred embodiment of this technical solution, the real-time preprocessing of the inspection data based on the UAV's onboard edge node includes: Noise filtering and key area cropping are performed on the visible light images and infrared images in the inspection data. Preliminary feature extraction is performed on the preprocessed image based on a pre-defined defect feature library; The extracted preliminary feature data is lightweight encoded to generate feature data that can be transmitted quickly.
[0009] As a further preferred embodiment of this technical solution: the step of generating task unloading decisions by the UAV's onboard edge node based on a preset inspection scenario and onboard computing power load includes: Real-time acquisition of the drone's remaining battery power, onboard edge node computing power utilization, and communication quality with the on-site edge computing unit; Based on the preset unloading strategy, determine whether the current airborne edge node meets the computing power requirements of the complex defect analysis task; If the conditions are not met, the complex defect analysis task will be marked as an unloadable task, and an unload decision instruction will be generated.
[0010] As a further preferred embodiment of this technical solution: the step of diverting some complex inspection and analysis tasks to the nearest on-site edge computing unit based on the task offloading decision includes: Based on the unloading decision instruction, the lightweight encoded feature data and the unloadable task request are sent to the field edge computing unit. If the communication quality does not meet the transmission conditions, the offloadable task will be temporarily stored in the local cache of the airborne edge node and will be split again after communication is restored.
[0011] As a further preferred embodiment of this technical solution: obtaining equipment defect detection results through collaborative computing between the UAV-borne edge node and the on-site edge computing unit includes: The on-site edge computing unit receives the feature data of the offloadable task and lightweight coding. The feature data is analyzed in depth based on a preset high-precision defect detection model to obtain defect detection results. The defect detection results are then transmitted back to the UAV's onboard edge node. The final equipment defect detection result is obtained by fusing the local preliminary analysis results with the returned defect detection results from the UAV's onboard edge node.
[0012] As a further preferred embodiment of this technical solution, the generation of real-time inspection alarm information based on the equipment defect detection results includes: The equipment defect detection results are classified into risk levels according to the preset defect level judgment rules; The graded defect information is associated and bound with the device ID and location coordinates; Generate structured alarm data containing defect type, risk level, and location information; The structured alarm data is stored locally and pushed to the on-site operation and maintenance terminal.
[0013] As a further preferred embodiment of this technical solution, the method further includes: An inspection data cache area is set in the field edge computing unit; Receive and store lightweight feature data and defect detection results transmitted back by the drone; After the drone returns to base, the cached data is synchronized to the local server in batches to generate a complete inspection report.
[0014] As a further preferred embodiment of this technical solution: the preset inspection scenarios include at least one of power line inspection, oil and gas pipeline inspection, and photovoltaic power station inspection. The airborne edge node automatically adjusts the preprocessing parameters, defect detection model thresholds, and task unloading trigger conditions according to different inspection scenarios.
[0015] To solve the above-mentioned technical problems, another technical solution adopted in this application is: an edge computing-based UAV inspection system, applied to the method described above, including UAV onboard edge nodes, on-site edge computing units, local servers, and on-site operation and maintenance terminals; The UAV-borne edge node is mounted on the UAV and is used to collect inspection data, perform real-time preprocessing and generate task unloading decisions, and interact with the on-site edge computing unit. The on-site edge computing unit is used to receive offloadable tasks and feature data, perform in-depth analysis, and provide feedback on defect detection results. The local server is used to synchronize cached data and generate a complete inspection report; The on-site maintenance terminal is used to receive structured alarm data and enable real-time viewing of potential hazards during inspections.
[0016] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions: 1. It solves the pain point of traditional drone inspection relying on public network transmission. By constructing a two-level edge architecture of drone-borne edge nodes and field edge computing units, it realizes local processing of inspection data, eliminating the need to transmit all data back to the cloud. This effectively reduces data transmission latency, avoids inspection interruptions caused by public network instability, enables real-time defect identification and alarms, and improves inspection efficiency.
[0017] 2. It balances the contradiction between airborne computing power and defect identification accuracy. Through real-time preprocessing of inspection data, lightweight coding and task adaptive offloading mechanism, it adapts to the computing power and power consumption constraints of airborne equipment, and leverages the computing power advantage of the on-site edge computing unit to achieve in-depth analysis of complex defects, ensuring high accuracy of defect identification, while reducing the power consumption of UAVs and ensuring the needs of long-endurance inspection.
[0018] 3. Improved security and compliance of inspection data by adopting local encrypted storage and domain-free control mode. The original inspection data and analysis results only circulate on the edge intranet, eliminating the risk of data leakage and tampering caused by cloud uploads, and meeting the data security compliance requirements of key industries.
[0019] 4. Adaptable to inspection needs in multiple scenarios: By pre-setting multi-scenario task templates and automatically adjusting parameters and model thresholds, it can be flexibly applied to different inspection scenarios such as power lines, oil and gas pipelines, and photovoltaic power stations, improving the versatility and adaptability of the method; at the same time, through multi-source data fusion and collaborative computing, it further improves the accuracy of defect detection and positioning precision.
[0020] 5. The system has a simple and highly scalable architecture, with clear division of labor and efficient collaboration among its components. It can cache, synchronize, and generate complete reports of inspection data, enabling on-site maintenance personnel to quickly respond to potential problems, reduce maintenance costs, and promote the intelligent and autonomous upgrade of drone inspections.
[0021] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the modules of the system of the present invention. Detailed Implementation
[0024] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.
[0025] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.
[0026] Figure 1 This is a flowchart illustrating a drone inspection method based on edge computing according to an embodiment of the present invention. It should be noted that if substantially the same result is achieved, the method of this application is not necessarily identical. Figure 1 The illustrated process sequence is limited. For example... Figure 1 As shown: A drone inspection method based on edge computing includes the following steps: S100: Acquire inspection data collected by the UAV's onboard edge node; Based on the above settings, inspection data can be collected synchronously by multi-source sensors mounted on the drone, enabling multi-dimensional and comprehensive equipment status monitoring. This avoids missed defects caused by single data collection. At the same time, the airborne edge node performs timestamp alignment and format standardization processing on the collected multi-source data to ensure the accuracy of subsequent preprocessing and analysis. The types of inspection data collected can be flexibly adjusted according to preset inspection scenarios to adapt to the differentiated needs of different industrial inspections. Furthermore, the sensor working status can be monitored in real time during the data collection process to avoid invalid data collection due to sensor failure.
[0027] In practical applications, for example, a drone is equipped with three sensors: a multispectral camera, an infrared thermal imager, and a lidar. The multispectral camera is used to collect visible light image data of the equipment surface (resolution no less than 1080P, frame rate 15-20FPS), the infrared thermal imager is used to collect temperature field data of the equipment (temperature measurement range -20℃~150℃, temperature measurement accuracy ±0.5℃), and the lidar is used to collect point cloud data of the equipment surface (point cloud density no less than 100 points / cm²). The airborne edge node has a built-in time synchronization module, which uses the NTP time synchronization algorithm to align the timestamps of the data collected by the three sensors, with the alignment error controlled within 50ms. The alignment formula is as follows: ; in, For the aligned standard timestamp, Let be the original timestamp of the data collected by the i-th sensor. The time deviation between the i-th sensor and the airborne edge node (calculated in real time by the NTP algorithm); the aligned data is standardized and packaged in the format of "timestamp-sensor type-data content" to form a unified inspection data frame, which is convenient for subsequent preprocessing steps.
[0028] S200. Perform real-time preprocessing of the inspection data based on the UAV's onboard edge node; Based on the above settings, preprocessing parameters can be flexibly adjusted according to the defect characteristics of the preset inspection scenario to achieve targeted optimization of inspection data, reduce interference from invalid data, and improve the efficiency and accuracy of subsequent defect identification. The preprocessing process adopts a lightweight algorithm to adapt to the computing power and power consumption constraints of airborne edge nodes, avoiding inference delays caused by excessive computing power consumption during preprocessing. At the same time, the preprocessed feature data is lightweight encoded to reduce the data transmission bandwidth requirements and facilitate rapid transmission during subsequent task unloading. During the preprocessing process, data quality can be judged in real time, and invalid data with excessive noise or ambiguity can be marked and removed to ensure the validity of the data input to the defect detection model.
[0029] In practical applications, such as power line inspection scenarios, the preprocessing steps specifically include three sub-steps: 1. Noise Removal and Key Region Cropping: An adaptive median filtering algorithm is used to remove noise from visible light and infrared images. The algorithm formula is as follows: ; in, The grayscale value of the pixel after filtering. This represents the original pixel grayscale value. The minimum grayscale value within the filtering window. The maximum grayscale value within the filtering window. The value is the median gray level within the filtering window; the size of the filtering window is adaptively adjusted according to the image noise intensity (range 3×3~7×7); key area cropping uses the YOLO lightweight model to locate the power line area, and crops out irrelevant background areas such as the sky and trees. After cropping, the image size is compressed to 640×480 pixels, reducing the computational power requirements for subsequent processing.
[0030] 2. Preliminary Feature Extraction: Based on a pre-set power line defect feature database (including defects such as broken conductor strands, damaged insulators, and corroded hardware), the SIFT feature extraction algorithm is used to extract key feature points from the preprocessed image. The feature extraction formula is as follows: ; in, For scale-space eigenvalues, It is a Gaussian function. The image is cropped, and * represents the convolution operation. The extracted feature points include information in three dimensions: position, scale, and orientation, forming a feature vector set.
[0031] 3. Lightweight Encoding: The PCA (Principal Component Analysis) algorithm is used to encode the feature vector set in a dimensionality-reducing manner, thereby reducing the data dimensionality and generating feature data that can be transmitted quickly. The encoding formula is as follows: ; in, For the encoded lightweight feature data, For the original feature vector set, The PCA dimensionality reduction matrix (obtained in advance from training samples) reduces the dimension of the feature data to 1 / 5 of the original dimension after encoding, reduces the transmission bandwidth requirement by more than 60%, and retains more than 95% of the effective feature information.
[0032] S300: The UAV's onboard edge node generates a task unloading decision based on a preset inspection scenario and onboard computing power load. Based on the above settings, multiple task offloading strategies can be preset to adapt to different inspection scenarios, different computing loads and communication quality, so as to realize adaptive adjustment of task offloading and avoid computing power waste or inference delay caused by fixed offloading strategies. In the process of generating task offloading decisions, three core factors are comprehensively considered: the remaining power of the UAV, the onboard computing power utilization rate, and the communication quality with the field edge computing unit. A multi-objective decision algorithm is adopted to achieve the optimal offloading decision with "low latency, low power consumption and high precision". At the same time, decision thresholds are set. When each parameter reaches the threshold, task offloading is automatically triggered without manual intervention, thereby improving the autonomy of inspection.
[0033] In practical applications, for example, the Analytic Hierarchy Process (AHP) can be used to construct a task offloading decision model. The specific steps are as follows: 1. Determine decision-making indicators: Select the remaining battery power of the drone ( ), computing power utilization rate of airborne edge nodes ( ), communication quality ( The three decision indicators are: communication quality is measured by signal-to-noise ratio (SNR), and computing power utilization rate is the ratio of current computing power to maximum computing power.
[0034] 2. Determining Indicator Weights: The weights of each indicator are calculated using the Analytic Hierarchy Process (AHP), a judgment matrix is constructed, and after consistency testing, the weight vector is obtained. ,in (Remaining battery power) (Computing power utilization rate) (Communication quality).
[0035] 3. Indicator Standardization: Linear normalization is used to standardize each indicator, eliminating the influence of dimensions. The standardization formula is: Regarding remaining battery power (the higher the better): ; Regarding computing power utilization (lower is better): ; For communication quality (higher is better): ; in, , These are the maximum and minimum remaining battery power (ranging from 20% to 100%). , These are the maximum and minimum values of computing power utilization (ranging from 0% to 100%). , These are the maximum and minimum values for communication quality (ranging from 0dB to 30dB).
[0036] 4. Decision value calculation and judgment: Calculate the decision value. Preset decision threshold ;when When it is determined that the current airborne edge node cannot meet the computing power requirements of the complex defect analysis task, the complex defect reasoning task is marked as an unloadable task, and an unloading decision instruction is generated; when When it is determined that the airborne edge node can independently complete the complex defect analysis task, the task unloading is not performed.
[0037] S400. Based on the task unloading decision, some complex inspection and analysis tasks are diverted to the nearest on-site edge computing unit. Based on the above settings, priority sorting rules for on-site edge computing units can be preset. The optimal on-site edge computing unit can be selected for task distribution based on the distance between the UAV and each unit, communication quality, and load conditions. This avoids analysis delays caused by tasks being distributed to edge units with excessive load or poor communication. An encrypted transmission protocol is used during task distribution to ensure the secure transmission of feature data and task instructions. A transmission timeout mechanism is also set; when a transmission timeout occurs or communication is interrupted, unloadable tasks are temporarily stored in the local cache of the airborne edge node and re-distributed after communication is restored, preventing task loss. Furthermore, complex inspection and analysis tasks can be broken down into multiple sub-tasks based on task complexity and distributed to different on-site edge computing units for parallel processing, improving analysis efficiency.
[0038] In practical applications, for example, the task offloading process specifically includes: 1. Selection of on-site edge computing units: A weighted summation method is used to prioritize available on-site edge computing units in the surrounding area. Ranking criteria include the distance between the UAV and the edge unit. ), communication quality ( ), edge unit load rate ( The priority calculation formula is as follows: ; in, , , These are the weighting coefficients. The preset maximum communication distance is 5km. This is a standardized value for communication quality. The current load rate of the field edge computing unit is set (range 0%~100%); the field edge computing unit with the highest priority is selected as the task offloading target.
[0039] 2. Task Offloading and Transmission: Based on the offloading decision command, lightweight encoded feature data and offloadable task requests (including task type, processing requirements, time limits, etc.) are sent to the target edge computing unit via the MQTT encrypted transmission protocol. During transmission, the data is encrypted using the AES-128 encryption algorithm. The encryption formula is as follows: ; in, For encrypted feature data, The encryption key is a pre-agreed encryption key between the airborne edge nodes and the field edge computing units. This refers to the feature data after lightweight encoding.
[0040] 3. Anomaly Handling: Set the transmission timeout threshold to 3 seconds. If no confirmation signal is received from the edge computing unit within 3 seconds, a transmission timeout is determined. At this point, check the current communication quality. If the communication quality... The unloadable tasks will be temporarily stored in the local cache of the onboard edge node (cache capacity not less than 16GB), and the communication quality will be checked every 5 seconds. If the communication quality is poor, the traffic splitting will be re-executed; If the task is not successfully resent, the task diversion request will be resent, with a maximum of 3 retries. If the retry fails, the edge computing unit at that location is deemed unavailable, and the next highest priority edge unit will be selected for diversion.
[0041] S500. Through the collaborative calculation of the UAV-borne edge node and the field edge computing unit, the equipment defect detection result is obtained; Based on the above settings, a multimodal data fusion algorithm can be constructed to integrate the local preliminary analysis results of the airborne edge nodes with the in-depth analysis results of the field edge computing unit, thereby improving the accuracy and reliability of defect detection and avoiding missed or false detections caused by a single analysis result. The field edge computing unit deploys a high-precision defect detection model and uses optimization methods such as model quantization and pruning to improve the model inference efficiency. At the same time, the defect detection model can be updated online according to changes in the inspection scenario to adapt to the identification needs of different defect types. During the collaborative computing process, the analysis progress of both parties is synchronized in real time to ensure the consistency and timeliness of the calculation results, and the credibility of the analysis results is verified, eliminating invalid results with credibility below the threshold.
[0042] In practical applications, for example, the collaborative computing process specifically includes: 1. In-depth analysis of the field edge computing unit: After receiving the encrypted feature data and the offloadable task request, the field edge computing unit decrypts the feature data using the AES-128 decryption algorithm to obtain lightweight feature data. Deploy a high-precision defect detection model based on an improved YOLOv8, perform in-depth analysis of feature data, and use TensorRT to accelerate and optimize the model inference process, improving inference speed by more than 50%. The loss function formula for the defect detection model is: ; in, The loss is categorical (using the cross-entropy loss function). The bounding box regression loss is calculated using the CIoU loss function. The target confidence loss is calculated using the cross-entropy loss function; through model inference, information such as defect type, defect location, and confidence level is obtained to form preliminary defect detection results. .
[0043] 2. Preliminary Local Analysis at Airborne Edge Nodes: While offloading tasks, the airborne edge nodes employ a lightweight defect detection model (based on YOLOv8-nano) to perform preliminary analysis on the preprocessed feature data, obtaining preliminary local analysis results. It includes the initially identified defect type, location, and confidence level.
[0044] 3. Results Integration: The DS evidence theory was used to analyze the results. and To improve the reliability of defect detection results, fusion is performed. The fusion formula is as follows: ; in, To assess the credibility of defect A after fusion, ; The conflict coefficient, To determine the confidence level of defect A1 in the analysis results of the field edge computing unit, The confidence level of defect A2 in the airborne edge node analysis results is set; the preset confidence threshold is 0.7. If the defect is deemed valid, it will be included in the final equipment defect detection results. ;when If the defect is identified as suspicious, it is marked and fed back to the on-site maintenance terminal for further manual confirmation.
[0045] 4. Result Feedback: The on-site edge computing unit will provide preliminary defect detection results. After encryption, the data is transmitted back to the airborne edge node. The airborne edge node then performs result fusion to obtain the final equipment defect detection result. It includes information such as defect type, defect location (calibrated by GPS positioning, with a positioning accuracy of ±1m), confidence level, and defect size.
[0046] S600. Generate real-time inspection alarm information based on the equipment defect detection results; Based on the above settings, differentiated defect level judgment rules can be preset according to the severity of defects in different inspection scenarios to achieve graded alarms for defect risks, making it easier for on-site maintenance personnel to prioritize the handling of high-risk hazards. The alarm information adopts a structured format, including core defect information and location information, which can be directly pushed to the on-site maintenance terminal, while being stored locally with encryption to ensure the traceability of alarm information. In addition, an alarm linkage mechanism can be set up so that when a high-risk defect is detected, an on-site audible and visual alarm is automatically triggered, the drone inspection mission is paused, and the flight path is adjusted to the defect area for secondary shooting to obtain clearer defect data, assisting maintenance personnel in judging the severity of the hazard.
[0047] In practical applications, for example, the alarm information generation and push process specifically includes: 1. Defect Level Determination: Based on preset defect level determination rules, and considering defect type, size, confidence level, and inspection scenario, defects are classified into three levels: Level 1 (high risk), Level 2 (medium risk), and Level 3 (low risk). The determination formula is as follows: ; in, The defect level score is calculated (range 0-10). Defect type weights (high-risk defects T=10, medium-risk defects T=6, low-risk defects T=3). This is the standardized value for the defect size (the larger the defect size, the larger the S value, ranging from 0 to 1). For the credibility of the defect, , , For weighting coefficients; when When, it is judged as a level one defect; when When, it is judged as a level two defect; when At that time, it was determined to be a level three defect.
[0048] 2. Alarm Information Generation: Generate structured alarm data, including the following core information: UAV ID, inspection scenario, defect type, defect level, and defect location (GPS coordinates: latitude). ,longitude The alarm information includes defect size, confidence level, detection time, and handling recommendations. Alarm information is encapsulated in JSON format for easy parsing by on-site maintenance terminals. An example of this encapsulation is shown below. {"drone_id":"UAV-001","inspection_scene":"Power line inspection","defect_type":"Conductor strand breakage","defect_grade":"Level 1","position":{"latitude":39.9042,"longitude":116.4074},"defect_size":"5mm×20mm","confidence":0.92,"detection_time":"2026-04-30 10:30:25","processing_suggestion":"Immediately stop the machine for inspection and repair, and replace the conductor"}
[0049] 3. Alarm Information Push and Storage: Structured alarm data is pushed to the on-site maintenance terminal (mobile APP, tablet terminal) using encrypted transmission, with a push delay of less than 1 second; at the same time, the alarm information, along with the corresponding inspection data and defect detection results, is stored locally on the airborne edge node and the on-site edge computing unit using the AES-256 encryption algorithm, with the storage path being "Inspection Date / Inspection Scene / UAV ID / Alarm Information", which facilitates subsequent traceability and review; for Level 1 defects, an additional on-site audible and visual alarm is triggered (alarm volume ≥80dB, light flashing frequency 1 time / second) to remind on-site maintenance personnel to handle the emergency.
[0050] Figure 2 This is a functional module diagram of an edge computing-based drone inspection system according to an embodiment of this application, as shown below. Figure 2 As shown, an edge computing-based drone inspection system includes an onboard edge node of the drone, a field edge computing unit, a local server, and a field operation and maintenance terminal. The UAV-borne edge node is mounted on the UAV and is used to collect inspection data, perform real-time preprocessing and generate task unloading decisions, and interact with the on-site edge computing unit. The on-site edge computing unit is used to receive offloadable tasks and feature data, perform in-depth analysis, and provide feedback on defect detection results. The local server is used to synchronize cached data and generate a complete inspection report; The on-site maintenance terminal is used to receive structured alarm data and enable real-time viewing of potential hazards during inspections.
[0051] For other details regarding the implementation techniques of each module in the above embodiments, please refer to the description of an edge computing-based UAV inspection method and its inspection system in the above embodiments, which will not be repeated here.
[0052] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system-type embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
[0053] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.
Claims
1. A drone inspection method based on edge computing, characterized in that, include: Acquire inspection data collected by the UAV's onboard edge nodes; The inspection data is preprocessed in real time based on the UAV's onboard edge node. The UAV's onboard edge node generates a task unloading decision based on a preset inspection scenario and onboard computing power load; Based on the task offloading decision, some complex inspection and analysis tasks are diverted to the nearest on-site edge computing unit. Equipment defect detection results are obtained through collaborative computing between the UAV-borne edge node and the field edge computing unit. Real-time inspection alarm information is generated based on the equipment defect detection results.
2. The UAV inspection method based on edge computing according to claim 1, characterized in that: The acquisition of inspection data collected by the UAV's onboard edge nodes includes: Visible light image data is acquired using a multispectral camera mounted on a drone; Temperature field data of the equipment is collected using an infrared thermal imager mounted on a drone; Surface point cloud data is collected using a lidar device mounted on a drone; The airborne edge nodes perform timestamp alignment on the multi-source data to form a unified inspection data frame.
3. The UAV inspection method based on edge computing according to claim 1, characterized in that: The real-time preprocessing of the inspection data based on the UAV's onboard edge node includes: Noise filtering and key area cropping are performed on the visible light images and infrared images in the inspection data. Preliminary feature extraction is performed on the preprocessed image based on a pre-defined defect feature library; The extracted preliminary feature data is lightweight encoded to generate feature data that can be transmitted quickly.
4. The UAV inspection method based on edge computing according to claim 1, characterized in that: The process of generating task offloading decisions by the UAV's onboard edge node based on a preset inspection scenario and onboard computing power load includes: Real-time acquisition of the drone's remaining battery power, onboard edge node computing power utilization, and communication quality with the on-site edge computing unit; Based on the preset unloading strategy, determine whether the current airborne edge node meets the computing power requirements of the complex defect analysis task; If the conditions are not met, the complex defect analysis task will be marked as an unloadable task, and an unload decision instruction will be generated.
5. The UAV inspection method based on edge computing according to claim 4, characterized in that: The step of offloading some complex inspection and analysis tasks to the nearest on-site edge computing unit based on the task offloading decision includes: Based on the unloading decision instruction, the lightweight encoded feature data and the unloadable task request are sent to the field edge computing unit. If the communication quality does not meet the transmission conditions, the offloadable task will be temporarily stored in the local cache of the airborne edge node and will be split again after communication is restored.
6. The UAV inspection method based on edge computing according to claim 1, characterized in that: The process of obtaining equipment defect detection results through collaborative computing between the UAV-borne edge node and the on-site edge computing unit includes: The on-site edge computing unit receives the feature data of the offloadable task and lightweight coding. The feature data is analyzed in depth based on a preset high-precision defect detection model to obtain defect detection results. The defect detection results are then transmitted back to the UAV's onboard edge node. The final equipment defect detection result is obtained by fusing the local preliminary analysis results with the returned defect detection results from the UAV's onboard edge node.
7. The UAV inspection method based on edge computing according to claim 1, characterized in that: The generation of real-time inspection alarm information based on the equipment defect detection results includes: The equipment defect detection results are classified into risk levels according to the preset defect level judgment rules; The graded defect information is associated and bound with the device ID and location coordinates; Generate structured alarm data containing defect type, risk level, and location information; The structured alarm data is stored locally and pushed to the on-site operation and maintenance terminal.
8. The UAV inspection method based on edge computing according to claim 1, characterized in that, The method further includes: An inspection data cache area is set in the field edge computing unit; Receive and store lightweight feature data and defect detection results transmitted back by the drone; After the drone returns to base, the cached data is synchronized to the local server in batches to generate a complete inspection report.
9. The UAV inspection method based on edge computing according to claim 1, characterized in that, The preset inspection scenarios include at least one of power line inspection, oil and gas pipeline inspection, and photovoltaic power station inspection. The airborne edge node automatically adjusts the preprocessing parameters, defect detection model thresholds, and task unloading trigger conditions according to different inspection scenarios.
10. An edge computing-based unmanned aerial vehicle (UAV) inspection system, applied to the method described in any one of claims 1-9, characterized in that, This includes UAV-borne edge nodes, on-site edge computing units, local servers, and on-site maintenance terminals; The UAV-borne edge node is mounted on the UAV and is used to collect inspection data, perform real-time preprocessing and generate task unloading decisions, and interact with the on-site edge computing unit. The on-site edge computing unit is used to receive offloadable tasks and feature data, perform in-depth analysis, and provide feedback on defect detection results. The local server is used to synchronize cached data and generate a complete inspection report; The on-site maintenance terminal is used to receive structured alarm data and enable real-time viewing of potential hazards during inspections.