Unmanned aerial vehicle control system and method for water conservancy inspection

Abstract

The invention discloses an unmanned aerial vehicle control system and method for water conservancy inspection, and relates to the technical field of water conservancy inspection, the unmanned aerial vehicle control system comprises an unmanned aerial vehicle main body, the unmanned aerial vehicle main body carries a multi-source sensor module, an edge calculation module, a self-adaptive navigation module and an obstacle avoidance module, and the method further comprises the steps of S1 system self-inspection, S2 environment modeling, S3 path planning and S4 self-adaptive inspection. S5, real-time processing; S6, abnormal feedback; and S7, data fusion. According to the invention, through cooperative work of the source sensor module, the edge calculation module, the adaptive navigation module and the obstacle avoidance module, the effect of efficient and accurate inspection is achieved; through optimal design of a multispectral camera, a temperature and humidity sensor, a self-adaptive navigation module and an obstacle avoidance module, the effect of adapting to routing inspection in a complex environment is achieved; through the steps of S1, system self-inspection, S2, environment modeling, S3, path planning and S4, adaptive inspection, the effects of scientifically planning the inspection process and improving the normalization and effectiveness of the inspection work are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy inspection, and in particular to a drone control system and method for water conservancy inspection. Background Art

[0002] As important infrastructure for safeguarding people's livelihoods and promoting economic development, the safe monitoring of the operating status of water conservancy facilities is of vital importance. Water conservancy projects such as dams, sluices, and water pipelines are subjected to long-term water erosion, seepage pressure, and environmental erosion, and are prone to safety hazards such as cracks, leakage, and structural deformation. Once a failure occurs, it may cause serious consequences such as flooding and waste of water resources, threatening the safety of life and property and affecting social stability. Traditional water conservancy inspections mainly rely on manual inspections and fixed-point sensor monitoring. Manual inspections require inspectors to conduct in-depth inspections of complex and dangerous water conservancy sites and detect hidden hazards through visual inspections and simple tool measurements. This has problems such as low efficiency, strong subjectivity, and difficulty in discovering hidden defects. Although fixed-point sensor monitoring can achieve real-time data collection in some areas, it has limitations such as limited monitoring range, high equipment maintenance costs, and inability to fully cover water conservancy facilities. Drone technology, with its strong maneuverability, wide field of view, and flexible deployment, is gradually being applied to the field of water conservancy inspections.

[0003] The existing drone-based water conservancy inspection technology has a single sensor configuration, which makes it difficult to comprehensively collect multi-dimensional information about facilities; data processing relies on cloud transmission, which has problems such as high latency and poor real-time performance; the navigation and obstacle avoidance function is not adaptable enough in complex water conservancy environments and cannot ensure the safe and stable flight of drones; defect detection and risk assessment lack intelligent means, making it difficult to accurately identify hidden dangers and assess their degree of harm; it is necessary to design a drone control system and method for water conservancy inspection to solve the above-mentioned problems. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: a drone control system for water conservancy inspection, comprising a drone body equipped with a multi-source sensor module, an edge computing module, an adaptive navigation module, and an obstacle avoidance module; The multi-source sensor module includes a high-precision laser radar, a multispectral camera and a temperature and humidity sensor. The laser radar has a scanning frequency of ≥20 Hz, a point cloud density of ≥100 points / ㎡, and a multispectral camera has a resolution of ≥20 million pixels. The edge computing module is internally provided with a lightweight neural network model for real-time processing of data from multi-source sensor modules and generating surface defect detection results for water conservancy facilities; The adaptive navigation module plans the inspection route in real time based on the dynamic window algorithm, and combines the preset three-dimensional model of the water conservancy facilities with the real-time environmental point cloud data to control the flight altitude of the drone body to 10-50 meters and the horizontal speed to 0.5-5m / s; The obstacle avoidance module integrates a UWB positioning unit and a millimeter-wave radar, with a detection range of 0.1-30 meters and a response time of ≤50ms.

[0005] Preferably, the multispectral camera includes visible light band and near infrared band, with a spectral resolution of ≤5nm and a frame rate of ≥30fps; The temperature and humidity sensor has a measurement accuracy of ±0.5°C and ±2%RH, and a sampling interval of ≤1s.

[0006] Preferably, the adaptive navigation module automatically switches to a wall-flight mode in a narrow waterway area, maintaining a distance of ≥0.5 meters between the main body of the drone and obstacles; The obstacle avoidance module enables the millimeter-wave radar-led obstacle avoidance strategy in an environment with visibility less than 10 meters and generates a three-dimensional polar coordinate system obstacle avoidance path.

[0007] A method for controlling a drone for water conservancy inspection is applied to the aforementioned drone control system for water conservancy inspection, and includes the following control steps: S1. System self-test: Initialize the multi-source sensor module, verify the lidar point cloud alignment error is less than 2cm, and the multispectral camera white balance parameters; S2. Environmental modeling: Obtain point cloud data of water conservancy facilities through LiDAR, segment the ground point cloud using the RANSAC algorithm, and construct a 3D semantic map of the inspection area; S3, Path Planning: Generate the global optimal path based on the dynamic window algorithm, and automatically increase the number of circling scans ≥ 3 times in the leakage risk area; S4, Adaptive inspection: Control the drone to fly in a zigzag trajectory. The flight altitude H and the camera field of view angle θ satisfy the relationship H = W / (2tan(θ / 2)), where W is the minimum width of the crack to be detected, W ≥ 1 cm. S5, real-time processing: The edge computing module uses the YOLOv5 lightweight model to detect crack targets and output location coordinates and crack width estimation with an accuracy error of ≤0.1mm; S6, Abnormal feedback: When a leak is detected, the drone is triggered to hover and the 5x digital zoom function of the multispectral camera is activated to capture local high-definition images with a resolution of ≥50 million pixels; S7. Data fusion: Associate defect data with the BIM model of water conservancy facilities to generate an inspection report including risk level assessment, which is uploaded to the control center via the LoRa wireless module.

[0008] Preferably, in step S2, voxel filtering and downsampling are used for point cloud data processing, the voxel size is set to 0.05m×0.05m×0.05m, and the density of the feature point cloud is retained to be ≥80 points / ㎡; At the same time, the octree structure is used to organize and manage the point cloud data after voxel filtering, and the efficient access of point cloud data is achieved through hierarchical storage and fast retrieval mechanism; A curvature-constrained smoothing algorithm is introduced to optimize the downsampled point cloud. While maintaining the surface geometric characteristics of the water conservancy facilities, noise points are effectively removed, making the constructed three-dimensional semantic map more accurate and providing a reliable data basis for subsequent inspection route planning and defect detection.

[0009] Preferably, in step S4, the flight speed v and the image acquisition frame rate f satisfy the relationship v≤0.6f*d, where d is the single-frame coverage distance when the image overlap rate is ≥80%; Furthermore, considering the impact of different lighting conditions and the surface materials of water conservancy facilities on image acquisition quality, the system has a built-in adaptive lighting adjustment module. When the ambient light intensity falls below a preset threshold, it automatically increases the sensitivity of the multispectral camera and simultaneously adjusts the parameter relationship between the image acquisition frame rate and flight speed to ensure clear and complete image data can be obtained even in low-light environments. At the same time, according to the reflective characteristics of the surface of water conservancy facilities, the camera exposure time is dynamically adjusted to avoid the loss of image information due to overexposure or underexposure, thereby improving the quality of inspection images and detection accuracy.

[0010] Preferably, in step S5, the crack width estimation adopts a sub-pixel edge detection algorithm combined with the laser radar point cloud depth data to achieve three-dimensional crack reconstruction; At the same time, deep learning-assisted edge refinement technology is introduced. By training a convolutional neural network model, sub-pixel edge detection results are optimized twice, automatically identifying and correcting misjudgments and missing parts in edge detection. By utilizing the spatial topological relationship of 3D point cloud data, a three-dimensional geometric model of the crack is constructed, the width of the crack is accurately measured, and the 3D parameters such as the depth and direction of the crack are precisely analyzed.

[0011] Preferably, in step S7, the risk level assessment is based on an expert system, the input parameters include the crack length L, width W, and the angle α between the crack strike and the hydraulic gradient direction, and the output risk score Score=0.3L+0.5W+0.2|sinα|; Environmental data collected by temperature and humidity sensors is incorporated into the assessment system. A mathematical model of environmental humidity and crack development rate is established based on historical data. The humidity parameter is introduced as a correction factor into the risk scoring formula, enabling the risk score to reflect the impact of environmental factors on the safety of water conservancy facilities in real time. The expert system also has a dynamic update function for risk levels. When inspections find that crack parameters have changed, the risk score is automatically recalculated and a risk change trend report is generated.

[0012] In summary, the present invention provides a drone control system and method for water conservancy inspection, which has the following beneficial effects: 1. Through the coordinated work of the multi-source sensor module, edge computing module, adaptive navigation module and obstacle avoidance module, efficient and accurate inspection results are achieved. The multi-source sensor module integrates high-precision lidar, multispectral camera and temperature and humidity sensor, which can comprehensively collect the three-dimensional structure, surface features and environmental data of water conservancy facilities; the edge computing module uses a lightweight neural network model to process data in real time to achieve rapid defect detection; the adaptive navigation module plans the optimal path based on the dynamic window algorithm, and intelligently adjusts the flight parameters based on the three-dimensional model and point cloud data; the obstacle avoidance module relies on the UWB positioning unit and millimeter wave radar to ensure the safe flight of the drone in complex environments, significantly improving the efficiency and accuracy of water conservancy inspections.

[0013] 2. Through the optimized design of a multispectral camera, temperature and humidity sensor, adaptive navigation module, and obstacle avoidance module, the drone achieves adaptability to complex inspection environments. The multispectral camera covers visible and near-infrared wavelengths, with high spectral resolution and frame rate ensuring image quality. The high-precision temperature and humidity sensor monitors environmental changes in real time, providing data for risk assessment. The adaptive navigation module automatically switches to wall-following flight mode in narrow waterways, and the obstacle avoidance module utilizes a millimeter-wave radar-dominated strategy in low-visibility environments. The combination of these two enables the drone to stably and safely complete inspection missions in a variety of complex environments, effectively expanding its application scenarios.

[0014] 3. Through the S1 system self-check, S2 environmental modeling, S3 path planning and S4 adaptive inspection steps, the scientific planning of the inspection process is achieved, and the standardization and effectiveness of the inspection work are improved. The S1 system self-check ensures the accuracy of the sensor's initial state; S2 environmental modeling uses lidar and RANSAC algorithm to build a high-precision three-dimensional semantic map; S3 path planning generates the global optimal path based on the dynamic window algorithm and focuses on scanning risk areas; S4 adaptive inspection uses the relationship between Z-shaped trajectory and specific parameters to ensure comprehensive inspection coverage without omissions.

[0015] 4. Through the S5 real-time processing, S6 abnormal feedback and S7 data fusion steps, the effect of intelligent analysis and decision-making is achieved. The entire chain from detection to analysis and decision-making is intelligent, providing a reliable basis for the maintenance of water conservancy facilities. The S5 real-time processing step uses the YOLOv5 lightweight model and sub-pixel edge detection algorithm to achieve high-precision crack identification and three-dimensional reconstruction; the S6 abnormal feedback step quickly responds to obtain high-definition images when leakage is found; the S7 data fusion step associates defect data with the BIM model, and the risk level assessment combines multiple parameters and environmental data, and the score is dynamically updated through the expert system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the process architecture of a drone control system and method for water conservancy inspection in the present invention. DETAILED DESCRIPTION

[0017] The following is combined with Figure 1 , further details of this application are given.

[0018] Example: The present invention provides a technical solution: a drone control system for water conservancy inspection, comprising a drone body, the drone body being equipped with a multi-source sensor module, an edge computing module, an adaptive navigation module, and an obstacle avoidance module. The multi-source sensor module can comprehensively perceive information about water conservancy facilities and the surrounding environment, providing a rich data basis for subsequent data analysis and decision-making; the edge computing module realizes real-time data processing, reduces data transmission delay, and improves the timeliness of defect detection; the adaptive navigation module ensures that the drone can fly efficiently and safely in complex water conservancy environments; and the obstacle avoidance module ensures that the drone can effectively avoid obstacles during flight, thereby improving flight safety. The multi-source sensor module includes a high-precision lidar, a multispectral camera, and a temperature and humidity sensor. The lidar has a scanning frequency of ≥20Hz, a point cloud density of ≥100 points / ㎡, and a multispectral camera with a resolution of ≥20 megapixels. The high-precision lidar can quickly and accurately obtain three-dimensional point cloud data of water conservancy facilities, providing high-precision spatial information for environmental modeling and defect detection. The high-resolution multispectral camera can clearly capture surface images of water conservancy facilities, facilitating the detection of subtle defects. The temperature and humidity sensor monitors ambient temperature and humidity in real time, providing data support for analyzing the environmental impact of water conservancy facilities. A lightweight neural network model is built into the edge computing module to process data from multi-source sensor modules in real time and generate surface defect detection results for water conservancy facilities. This lightweight neural network model runs within the edge computing module, significantly reducing the time it takes to transmit data to the cloud. This allows for rapid detection and analysis of surface defects in water conservancy facilities, enabling timely identification of potential safety hazards. The adaptive navigation module plans the inspection route in real time based on a dynamic window algorithm. Combining a preset 3D model of water conservancy facilities with real-time environmental point cloud data, it controls the drone's flight altitude to 10-50 meters and a horizontal speed of 0.5-5m / s. The dynamic window algorithm, combined with the 3D model and real-time point cloud data, enables the drone to flexibly plan the optimal inspection route based on the actual environment, while ensuring inspection quality and improving inspection efficiency. A reasonable flight altitude and speed range facilitates comprehensive and detailed inspections of water conservancy facilities. The obstacle avoidance module integrates a UWB positioning unit and a millimeter-wave radar, with a detection range of 0.1-30 meters and a response time of ≤50ms. The integration of the UWB positioning unit and the millimeter-wave radar enables the drone to have all-round, high-precision obstacle detection capabilities. The short response time allows the drone to make obstacle avoidance actions in a timely manner, effectively avoiding collision accidents.

[0019] The multispectral camera covers visible light and near-infrared bands, with a spectral resolution of ≤5nm and a frame rate of ≥30fps. The combination of visible light and near-infrared bands, as well as high spectral resolution and frame rate, can obtain richer surface information of water conservancy facilities, helping to identify facilities of different materials and conditions, and improving the accuracy and comprehensiveness of defect detection. The temperature and humidity sensor has a measurement accuracy of ±0.5°C and ±2%RH, with a sampling interval of ≤1s. The high-precision temperature and humidity measurement and short sampling interval can accurately capture subtle changes in ambient temperature and humidity, providing reliable data for evaluating the performance of water conservancy facilities under different environmental conditions.

[0020] The adaptive navigation module automatically switches to wall-flight mode in narrow waterways, maintaining a distance of ≥0.5 meters between the drone and obstacles. This allows the drone to fly safely in complex and narrow spaces, maintain a safe distance from obstacles, and avoid collisions. It also enables closer and more detailed inspections of narrow areas of water conservancy facilities. The obstacle avoidance module uses a millimeter-wave radar-led obstacle avoidance strategy when visibility is less than 10 meters, and generates a three-dimensional polar coordinate obstacle avoidance path. In low-visibility environments, the millimeter-wave radar-led obstacle avoidance strategy can fully utilize its advantage of being unaffected by light. The three-dimensional polar coordinate obstacle avoidance path can more accurately plan the drone's obstacle avoidance actions, ensuring flight safety in harsh environments.

[0021] See also Figure 1 A method for controlling a drone for water conservancy inspection is applied to the above-mentioned drone control system for water conservancy inspection, and includes the following control steps: S1. System self-test: Initialize the multi-source sensor module, verify the LiDAR point cloud alignment error is less than 2cm, and the multispectral camera white balance parameters, to ensure that each sensor is in the best working condition. The small LiDAR point cloud alignment error and accurate multispectral camera white balance parameters can ensure the accuracy and reliability of subsequent collected data, laying a good foundation for the entire inspection work; S2. Environmental Modeling: LiDAR is used to acquire point cloud data of water conservancy facilities. The RANSAC algorithm is used to segment the ground point cloud and construct a 3D semantic map of the inspection area. This 3D semantic map, constructed using LiDAR point cloud data and the RANSAC algorithm, can intuitively and accurately present the spatial structure of water conservancy facilities and their surroundings, providing precise geographic information reference for UAV path planning and defect location. S3. Path Planning: Generates a globally optimal path based on a dynamic window algorithm. Automatically increases the number of circling scans by 3 or more times in leakage risk areas. This globally optimal path generated by the dynamic window algorithm improves inspection efficiency. Increasing the number of circling scans in leakage risk areas allows for more detailed inspections, increases the probability of discovering potential leakage hazards, and ensures the safety of water conservancy facilities. S4. Adaptive Inspection: The drone is controlled to fly in a zigzag trajectory. The flight altitude H and the camera's field of view angle θ satisfy the relationship H = W / (2tan(θ / 2)), where W is the minimum width of the crack to be detected, W ≥ 1cm. The zigzag trajectory flight, combined with the specific relationship between the flight altitude and the camera's field of view, ensures that the drone conducts a comprehensive and complete inspection of the surface of water conservancy facilities, improving the detection coverage of defects such as cracks. S5. Real-time processing: The edge computing module uses the YOLOv5 lightweight model to detect crack targets and output location coordinates and crack width estimates with an accuracy error of ≤0.1mm. The YOLOv5 lightweight model, under the real-time processing of the edge computing module, can quickly and accurately detect crack targets and provide high-precision location coordinates and crack width estimates, providing key data for assessing the extent of damage to water conservancy facilities. S6. Abnormal feedback: When a leak is detected, the drone is triggered to hover and the 5x digital zoom function of the multispectral camera is activated to collect local high-definition images with a resolution of ≥50 million pixels. The rapid response operation when a leak is detected can obtain high-resolution local images of the leak point, which helps to further analyze the cause and severity of the leak and provide detailed basis for subsequent repairs; S7. Data Fusion: Defect data is linked to the BIM model of water conservancy facilities to generate inspection reports including risk level assessments. These reports are uploaded to the control center via the LoRa wireless module. Data fusion generates inspection reports with risk assessments, allowing managers to fully understand the status of water conservancy facilities. Data uploaded via the LoRa wireless module can be transmitted remotely, facilitating timely decision-making and scheduling of maintenance work.

[0022] In step S2, voxel filtering and downsampling are used for point cloud data processing. The voxel size is set to 0.05m×0.05m×0.05m, and the feature point cloud density is retained to be ≥80 points / ㎡. Voxel filtering and downsampling reduce the data volume while retaining key features, ensuring the subsequent efficient processing and analysis of point cloud data and meeting the system's requirements for data accuracy and processing efficiency. At the same time, the octree structure is used to organize and manage the point cloud data after voxel filtering. Through layered storage and fast retrieval mechanisms, efficient access to point cloud data is achieved. The octree structure organizes and manages point cloud data, which can significantly improve data storage and retrieval efficiency, facilitate rapid acquisition of required data in subsequent operations, and improve the overall operation speed of the system. A curvature-constrained smoothing algorithm is also introduced to optimize the downsampled point cloud. While maintaining the surface geometric features of the water conservancy facilities, noise points are effectively removed, making the constructed three-dimensional semantic map more accurate, providing a reliable data basis for subsequent inspection path planning and defect detection. The curvature-constrained smoothing algorithm optimizes point cloud data, removes noise while retaining geometric features, improves the quality of the three-dimensional semantic map, provides more accurate data for path planning and defect detection, and enhances the system's detection and analysis capabilities.

[0023] In step S4, the flight speed v and the image acquisition frame rate f satisfy the relationship v≤0.6f*d, where d is the single-frame coverage distance when the image overlap rate is ≥80%. This ensures that the image acquisition frame rate matches the flight speed at different flight speeds and the image overlap rate is guaranteed, thereby obtaining comprehensive and continuous surface images of the water conservancy facilities, which is beneficial for subsequent image analysis and defect detection. In addition, considering the impact of different lighting conditions and the surface materials of water conservancy facilities on image acquisition quality, the system has a built-in adaptive lighting adjustment module. When the ambient light intensity is lower than the preset threshold, it automatically increases the sensitivity of the multispectral camera and synchronously adjusts the parameter relationship between the image acquisition frame rate and the flight speed to ensure that clear and complete image data can be obtained even in low-light environments. The adaptive lighting adjustment module can automatically adjust parameters according to the ambient light and the surface material of the facility to ensure that high-quality images can be collected under various lighting conditions, improving the system's adaptability to complex environments and the quality of inspection images. At the same time, according to the reflective characteristics of the surface of water conservancy facilities, the exposure time of the camera is dynamically adjusted to avoid the loss of image information due to overexposure or underexposure, thereby improving the quality of inspection images and detection accuracy. Dynamic adjustment of exposure time can effectively deal with the reflective problem on the surface of water conservancy facilities, prevent image information loss, further improve the quality of inspection images, and enhance the accuracy of defect detection.

[0024] In step S5, the crack width is estimated using a sub-pixel edge detection algorithm combined with the LiDAR point cloud depth data to achieve 3D crack reconstruction. The sub-pixel edge detection algorithm combined with the LiDAR point cloud depth data can accurately measure the crack width and achieve 3D reconstruction, providing detailed 3D information for comprehensive and in-depth analysis of crack morphology and development trends. At the same time, deep learning-assisted edge refinement technology is introduced. By training a convolutional neural network model, sub-pixel edge detection results are optimized twice, automatically identifying and correcting misjudgments and missing parts in edge detection. Deep learning-assisted edge refinement technology further optimizes edge detection results, improves the accuracy and reliability of crack detection, reduces misjudgments and missed detections, and improves the system's crack detection accuracy. By utilizing the spatial topological relationship of three-dimensional point cloud data, a three-dimensional geometric model of the crack is constructed, the width of the crack is accurately measured, and three-dimensional parameters such as the depth and direction of the crack are precisely analyzed. The three-dimensional geometric model constructed based on the spatial topological relationship of three-dimensional point cloud data can comprehensively and accurately analyze the various three-dimensional parameters of the crack, providing more scientific and comprehensive data support for the safety assessment of water conservancy facilities.

[0025] In step S7, the risk level assessment is based on an expert system. The input parameters include the fracture length L, width W, and the angle α between the fracture strike and the hydraulic gradient direction. The output risk score Score = 0.3L + 0.5W + 0.2|sinα|. The expert system combines key parameters to perform risk level assessment, providing a scientific and quantitative basis for water conservancy facility risk assessment, helping managers quickly understand the facility risk status. Incorporate environmental data collected by temperature and humidity sensors into the assessment system, establish a mathematical model of environmental humidity and crack development rate based on historical data, and introduce humidity parameters as a correction factor into the risk scoring formula, so that the risk score can reflect the impact of environmental factors on the safety of water conservancy facilities in real time. Incorporating environmental data and establishing a mathematical model can make the risk score more comprehensive and accurate in reflecting the actual risk situation of water conservancy facilities, taking into account the impact of environmental factors on facility safety, and improving the scientific nature and practicality of risk assessment; The expert system also has a dynamic update function for risk levels. When inspections find changes in crack parameters, the risk score is automatically recalculated and a risk change trend report is generated. The dynamic update function for risk levels can track changes in water conservancy facility risks in a timely manner. The generated trend report helps managers understand the development trend of facility risks and formulate maintenance and management strategies in advance.

[0026] The examples of this specific embodiment are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Identical components are represented by the same reference numerals. Therefore, any equivalent changes made based on the structure, shape, and principle of this application should be included in the scope of protection of this application.

Claims

1. A drone control system for water conservancy inspection, comprising a drone body, characterized in that: The drone body is equipped with a multi-source sensor module, an edge computing module, an adaptive navigation module and an obstacle avoidance module; The multi-source sensor module includes a high-precision laser radar, a multispectral camera and a temperature and humidity sensor. The laser radar has a scanning frequency of ≥20 Hz, a point cloud density of ≥100 points / ㎡, and a multispectral camera has a resolution of ≥20 million pixels. The edge computing module is internally provided with a lightweight neural network model for real-time processing of data from multi-source sensor modules and generating surface defect detection results for water conservancy facilities; The adaptive navigation module plans the inspection route in real time based on the dynamic window algorithm, and combines the preset three-dimensional model of the water conservancy facilities with the real-time environmental point cloud data to control the flight altitude of the drone body to 10-50 meters and the horizontal speed to 0.5-5m / s; The obstacle avoidance module integrates a UWB positioning unit and a millimeter-wave radar, with a detection range of 0.1-30 meters and a response time of ≤50ms.

2. The UAV control system for water conservancy inspection according to claim 1 is characterized in that: The multispectral camera includes visible light band and near infrared band, with a spectral resolution of ≤5nm and a frame rate of ≥30fps; The temperature and humidity sensor has a measurement accuracy of ±0.5°C and ±2%RH, and a sampling interval of ≤1s.

3. The UAV control system for water conservancy inspection according to claim 1 is characterized in that: The adaptive navigation module automatically switches to wall-flight mode in narrow waterways, maintaining a distance of ≥0.5 meters between the drone and obstacles. The obstacle avoidance module enables the millimeter-wave radar-led obstacle avoidance strategy in an environment with visibility less than 10 meters and generates a three-dimensional polar coordinate system obstacle avoidance path.

4. A method for controlling a drone for water conservancy inspection, applied to the drone control system for water conservancy inspection according to any one of claims 1 to 3, characterized in that: The control steps include: S1. System self-test: Initialize the multi-source sensor module, verify the lidar point cloud alignment error is less than 2cm, and the multispectral camera white balance parameters; S2. Environmental modeling: Obtain point cloud data of water conservancy facilities through LiDAR, segment the ground point cloud using the RANSAC algorithm, and construct a 3D semantic map of the inspection area; S3, Path Planning: Generate the global optimal path based on the dynamic window algorithm, and automatically increase the number of circling scans ≥ 3 times in the leakage risk area; S4, Adaptive inspection: Control the drone to fly in a zigzag trajectory. The flight altitude H and the camera field of view angle θ satisfy the relationship H = W / (2tan(θ / 2)), where W is the minimum width of the crack to be detected, W ≥ 1 cm. S5, real-time processing: The edge computing module uses the YOLOv5 lightweight model to detect crack targets and output location coordinates and crack width estimation with an accuracy error of ≤0.1mm; S6, Abnormal feedback: When a leak is detected, the drone is triggered to hover and the 5x digital zoom function of the multispectral camera is activated to capture local high-definition images with a resolution of ≥50 million pixels; S7. Data fusion: Associate defect data with the BIM model of water conservancy facilities to generate an inspection report including risk level assessment, which is uploaded to the control center via the LoRa wireless module.

5. The method for controlling a drone for water conservancy inspection according to claim 4, characterized in that: In step S2, the point cloud data is processed by voxel filtering and downsampling, the voxel size is set to 0.05m×0.05m×0.05m, and the feature point cloud density is retained to be ≥80 points / ㎡; At the same time, the octree structure is used to organize and manage the point cloud data after voxel filtering, and the efficient access of point cloud data is achieved through hierarchical storage and fast retrieval mechanism; A curvature-constrained smoothing algorithm is introduced to optimize the downsampled point cloud. While maintaining the surface geometric characteristics of the water conservancy facilities, noise points are effectively removed, making the constructed three-dimensional semantic map more accurate and providing a reliable data basis for subsequent inspection route planning and defect detection.

6. The method for controlling a drone for water conservancy inspection according to claim 4, characterized in that: In step S4, the flight speed v and the image acquisition frame rate f satisfy the relationship v≤0.6f*d, where d is the single-frame coverage distance when the image overlap rate is ≥80%; Furthermore, considering the impact of different lighting conditions and the surface materials of water conservancy facilities on image acquisition quality, the system has a built-in adaptive lighting adjustment module. When the ambient light intensity falls below a preset threshold, it automatically increases the sensitivity of the multispectral camera and simultaneously adjusts the parameter relationship between the image acquisition frame rate and flight speed to ensure clear and complete image data can be obtained even in low-light environments. At the same time, according to the reflective characteristics of the surface of water conservancy facilities, the camera exposure time is dynamically adjusted to avoid the loss of image information due to overexposure or underexposure, thereby improving the quality of inspection images and detection accuracy.

7. The method for controlling a drone for water conservancy inspection according to claim 4, characterized in that: In step S5, the crack width estimation adopts a sub-pixel edge detection algorithm combined with the laser radar point cloud depth data to achieve three-dimensional crack reconstruction; At the same time, deep learning-assisted edge refinement technology is introduced. By training a convolutional neural network model, sub-pixel edge detection results are optimized twice, automatically identifying and correcting misjudgments and missing parts in edge detection. By utilizing the spatial topological relationship of 3D point cloud data, a three-dimensional geometric model of the crack is constructed, the width of the crack is accurately measured, and the 3D parameters such as the depth and direction of the crack are precisely analyzed.

8. The method for controlling a drone for water conservancy inspection according to claim 4, characterized in that: In step S7, the risk level assessment is based on an expert system, with input parameters including fracture length L, width W, and angle α between the fracture strike and the hydraulic gradient direction, and an output risk score Score = 0.3L + 0.5W + 0.2|sinα|; Environmental data collected by temperature and humidity sensors is incorporated into the assessment system. A mathematical model of environmental humidity and crack development rate is established based on historical data. The humidity parameter is introduced as a correction factor into the risk scoring formula, enabling the risk score to reflect the impact of environmental factors on the safety of water conservancy facilities in real time. The expert system also has a dynamic update function for risk levels. When inspections find that crack parameters have changed, the risk score is automatically recalculated and a risk change trend report is generated.

Images