Unmanned aerial vehicle cluster water conservancy monitoring system for in-situ self-calibration of water body

Abstract

This invention discloses a UAV swarm water conservancy engineering monitoring system for in-situ self-calibration of water bodies, relating to water conservancy engineering monitoring technology. The system includes: a UAV swarm and multimodal sensing devices mounted on each UAV; edge computing nodes for real-time preprocessing and in-situ self-calibration of multi-source monitoring data; and a cloud-based data processing center configured with an engineering adaptation module, a task scheduling module, and a data analysis module. The engineering adaptation module matches corresponding monitoring strategies based on the target water conservancy project's engineering type. The task scheduling module dynamically generates cruise paths and payload configuration instructions based on the monitoring strategies and distributes them to the UAV swarm. The data analysis module incorporates a water conservancy engineering deformation interpretation model and a water environment inversion model to calculate the dam deformation rate index and water pollution probability, and dynamically adjusts the cruise path based on the calculation results. This invention improves the monitoring accuracy, early warning capability, and environmental adaptability of water conservancy projects.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering monitoring technology, specifically to a UAV swarm water conservancy engineering monitoring system for in-situ self-calibration of water bodies. Background Technology

[0002] With the continuous expansion of the existing scale of water conservancy projects, the routine safety monitoring of water conservancy facilities such as dams and dikes faces enormous challenges. Traditional monitoring methods mainly rely on a combination of manual inspections and fixed-point sensors. Although drone technology has been introduced, the following technical bottlenecks still urgently need to be addressed.

[0003] First, data acquisition quality is severely constrained by the environment and lacks an in-situ calibration mechanism. In inspections of high dams or long-distance dikes, single-unit drones are often used for aerial photography. However, water conservancy sites are characterized by high humidity, high fog, and strong absorption. The multispectral and infrared sensors carried by drones are susceptible to data drift due to water vapor scattering and changes in solar irradiance, or color distortion caused by lens contamination with water vapor. Existing ground station systems lack real-time closed-loop evaluation and physical calibration mechanisms for image quality. Often, invalid data due to sensor drift is only discovered after data is transmitted back, making it difficult to meet the requirements for high-precision identification of dam cracks or seepage.

[0004] Secondly, the monitoring methods are outdated and lack dynamic risk assessment capabilities. Current water conservancy monitoring mostly involves periodic static measurements, such as periodically comparing historical elevation data to determine dam deformation. This method cannot accurately capture minute accelerated deformations within the intervals between measurements. Furthermore, in water pollution monitoring, traditional methods rely on manual sampling and testing, which is time-consuming. Existing spectral inversion technology does not consider real-time correction during flight, making it impossible to accurately and in real-time invert water quality deterioration trends using spectral data, thus hindering the response to sudden algal blooms or piping incidents.

[0005] Finally, the system suffers from low intelligence and poor adaptability to complex scenarios. Existing drone monitoring systems typically employ a single transmission link, making them difficult to operate in remote reservoir areas where communication is limited. Furthermore, for enclosed spaces such as corridors and gate wells, alarms are usually triggered by a single type of gas sensor, which is susceptible to cross-interference leading to false alarms. Moreover, the lack of effective edge computing capabilities for real-time quantitative assessment of on-site risks poses a threat to the safety of inspection personnel.

[0006] Therefore, there is an urgent need for a water conservancy engineering monitoring system based on UAV swarms. By introducing intelligent algorithms and collaborative architecture, especially to solve the problems of sensor data drift and real-time self-calibration in complex field environments, the system can improve monitoring accuracy, early warning capabilities, and environmental adaptability. Summary of the Invention

[0007] This invention addresses the technical challenges in water conservancy project monitoring, such as complex environments, sensor susceptibility to water vapor interference leading to data drift, and high real-time requirements. It provides a UAV swarm water conservancy project monitoring system with in-situ self-calibration for water bodies, which improves the monitoring accuracy, early warning capabilities, and environmental adaptability of water conservancy projects.

[0008] This invention provides a UAV swarm water conservancy engineering monitoring system for in-situ self-calibration of water bodies, the system comprising:

[0009] The drone swarm and the multimodal sensing device mounted on each drone, wherein the drone swarm includes routine inspection drones, surveying and monitoring drones, water quality inspection drones and blind spot drones in confined spaces, and the multimodal sensing device is used to collect multi-source monitoring data;

[0010] Edge computing nodes, deployed on the airborne end of the UAV cluster or on a ground station near the site, are used to perform real-time preprocessing and in-situ self-calibration of the multi-source monitoring data output by the multimodal sensing devices;

[0011] The cloud-based data processing center is equipped with an engineering adaptation module, a task scheduling module, and a data analysis module. The engineering adaptation module matches the corresponding monitoring strategy based on the engineering type of the target water conservancy project. The task scheduling module dynamically generates cruise paths and payload configuration instructions based on the monitoring strategies and sends them to the UAV cluster. The data analysis module incorporates a water conservancy project deformation interpretation model and a water environment inversion model, used to calculate the dam deformation rate index and water pollution probability based on calibrated monitoring data, and dynamically adjust the cruise path based on the calculation results.

[0012] In some embodiments, the multimodal sensing device includes: a visible light camera and an infrared thermal imaging camera mounted on the daily inspection drone, a lidar sensing device mounted on the mapping and monitoring drone, a multispectral sensing device mounted on the water quality inspection drone, and a gas sensor mounted on the confined space blind spot filling drone.

[0013] In some embodiments, the edge computing node has a built-in in-situ self-calibration module, used to control the UAV to hover above a still water body within the monitoring area before performing a monitoring task, collect the multispectral reflectance and surface temperature of the water body, and calculate the correction gain coefficient based on the theoretical reflectance calculated from a priori library of water body optical properties.

[0014] = / ;

[0015] in, Indicates the correction gain coefficient. This represents the multispectral reflectance of the collected water surface. This represents the theoretical reflectivity.

[0016] In some embodiments, the edge computing node is further configured to perform Gaussian filtering-based noise reduction processing on the visible light image output by the visible light camera on the daily inspection drone, and after using the correction gain coefficient to perform gain compensation on the image grayscale value, calculate the image sharpness score:

[0017] = ;

[0018] in, Rate the image sharpness. To correct the grayscale values ​​of the pixels, gradient, × For image resolution; when When the image sharpness score is below a preset threshold, the drone is triggered to adjust its focus or flight altitude until the image sharpness score is greater than or equal to the preset threshold.

[0019] The edge computing node also uses the water surface temperature collected by the in-situ self-calibration module as a reference to calculate the relative temperature difference on the dam surface:

[0020] = - ;

[0021] in, For the first monitoring via sensors on the drone Temperature value of each pixel The water surface temperature is collected by the in-situ self-calibration module; when When the temperature exceeds the preset value, it is marked as a suspected leakage area, and the data analysis module is triggered to increase the cyclic sampling frequency of that area.

[0022] In some embodiments, the data analysis module includes a water environment inversion model, wherein the water environment inversion model is configured to perform water quality assessment based on optical monitoring data output by the multispectral sensing device on the water quality inspection drone, the water quality assessment including:

[0023] The original light reflectance in the light monitoring data is corrected based on the correction gain coefficient:

[0024] = , = ;

[0025] in, This represents the original infrared reflectivity. This is the corrected infrared reflectivity; This represents the original infrared reflectivity. This is the corrected infrared reflectivity;

[0026] Normalized difference vegetation index was calculated based on the corrected infrared and near-infrared reflectance to retrieve chlorophyll concentration:

[0027] NDVI= ;

[0028] NDVI is the Normalized Difference Vegetation Index.

[0029] Based on the normalized difference vegetation index and combined with the turbidity inversion algorithm, the probability of water pollution is calculated:

[0030] = ;

[0031] in, For the probability of water pollution, The turbidity parameters are calculated based on the blue-green light band. , , For weight parameters, It is the Sigmoid activation function;

[0032] When the probability of water pollution is greater than a preset threshold, it is determined that the water area is at risk of algal bloom or oil film pollution.

[0033] In some embodiments, the data analysis module is configured to: perform a dam deformation analysis process based on the radar data output by the lidar monitoring device.

[0034] Acquire point cloud data collected by the lidar monitoring equipment in the current period and historical baseline point cloud data;

[0035] Iterative nearest point (ICP) registration is performed on point cloud data and historical baseline point cloud data to calculate the 3D displacement of corresponding points;

[0036] Calculate the dam deformation rate index based on the aforementioned three-dimensional displacement:

[0037] = ;

[0038] in, This is the dam deformation rate index. and These are the coordinates of the base period and the current period, respectively. For time intervals;

[0039] When the dam deformation rate index exceeds a preset safety threshold, a deformation warning is issued.

[0040] In some embodiments, the confined space gap-filling drone is a micro multi-rotor drone, and the gas sensors it carries include at least a CO sensor and... sensor;

[0041] The data analysis module has a built-in gas diffusion early warning model for calculating the comprehensive gas hazard index in enclosed spaces.

[0042] = ;

[0043] in, To calculate the comprehensive gas hazard index, and To monitor concentration in real time, and This is the occupational exposure limit; when the comprehensive gas hazard index is greater than the preset hazard index, personnel are prohibited from entering and a drone is triggered for retesting.

[0044] 8. The system according to claim 1, wherein the task scheduling module is configured to dynamically adjust the cruise path using an adaptive artificial intelligence (AI) algorithm, the cost function of which... Defined as:

[0045] = + + ;

[0046] in, The actual path cost from the origin to node n. For heuristic functions, Let n be the risk weight value of node n. As a risk adjustment factor, the task scheduling module prioritizes... In areas with high values, hover calibration nodes that pass through stagnant water bodies are forcibly inserted into the path planning.

[0047] In some embodiments, when the engineering adaptation module determines that the target project is a dam project, the following monitoring strategy is executed:

[0048] The task scheduling module prioritizes scheduling the daily inspection drone to conduct close-range patrols of the dam body, calculates the low-temperature seepage zone, and uses in-situ self-calibrated infrared data to calculate the low-temperature seepage zone.

[0049] At preset intervals, the surveying and monitoring drone is scheduled to collect three-dimensional point cloud data and calculate the dam deformation rate index.

[0050] During the flood season or under conditions of sudden changes in reservoir water level, the narrow space gap-filling drone is triggered to enter the corridor and gate well, calculate the comprehensive gas hazard index, and transmit the collected wide-angle images back to the cloud for crack and defect identification.

[0051] In some embodiments, when the engineering adaptation module determines that the target project is a river embankment project, the following monitoring strategy is executed:

[0052] The task scheduling module schedules the daily inspection drone to conduct long-distance strip patrols along the dike line, uses the edge computing node to identify dike cracks and animal burrows in real time, and dynamically adjusts the patrol density for dangerous sections.

[0053] The water quality inspection drone is dispatched to perform spectral scanning of the river section. Before each scan begins, in-situ self-calibration is performed using adjacent water bodies, and the probability of water pollution is calculated. Under high water level conditions, the infrared thermal imaging camera of the daily inspection drone is linked to conduct initial screening of piping and seepage at the embankment toe and structures penetrating the embankment based on temperature difference.

[0054] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0055] (1. Based on the in-situ self-calibration mechanism of water medium, the static water body is used as a natural radiation calibration source. The gain coefficient is dynamically calculated to compensate for multispectral and infrared data in real time, which effectively overcomes the distortion problems caused by sensor data drift and lens contamination in high humidity and high fog environments.

[0056] (2) By introducing the water surface temperature as a reference blackbody, the emissivity parameters of the infrared thermal imaging camera are dynamically corrected, which solves the problem of interference from changes in atmospheric transmittance in traditional temperature measurement. Combined with the local window temperature difference algorithm, water surface reflection and environmental radiation noise are significantly eliminated, and the false alarm rate is greatly reduced.

[0057] (3) By forcibly embedding the in-situ self-calibration node into the UAV cruise path planning, an integrated edge computing architecture of "flight-calibration-monitoring" was constructed. This not only eliminated the measurement errors accumulated in long-distance inspections, but also reduced the bandwidth occupation of the original invalid data backhaul, enabling the system to maintain high-precision and low-latency stable operation in remote water conservancy hubs with limited communication. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a structural block diagram of a drone swarm water conservancy engineering monitoring system for in-situ self-calibration of water bodies, according to an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are only for explaining this invention and are not intended to limit this invention.

[0061] This invention constructs a drone swarm monitoring system based on edge-cloud collaborative computing. By introducing an in-situ self-calibration mechanism for water bodies, it utilizes the large-area static water bodies inherent in water conservancy projects as natural radiation calibration sources. Through specific algorithm models (such as clarity evaluation, deformation rate calculation, gas risk index, and spectral gain correction), it transforms the traditional "manual map reading" into "intelligent arithmetic," fundamentally solving the problem of monitoring accuracy attenuation in complex field environments.

[0062] On the one hand, the present invention provides a UAV swarm water conservancy engineering monitoring system for in-situ self-calibration of water bodies. Figure 1 This is a structural block diagram of a drone swarm water conservancy engineering monitoring system for in-situ self-calibration in water bodies, according to an embodiment of the present invention. (Reference) Figure 1 The monitoring system includes: a cluster of drones and multimodal sensing devices mounted on each drone, edge computing nodes, and a cloud data processing center.

[0063] The drone swarm comprises heterogeneous drones of various types and multimodal sensing devices mounted on each drone. These multimodal sensors are used to collect multi-source monitoring data for water conservancy projects. The drone swarm also includes routine inspection drones, surveying and monitoring drones, water quality inspection drones, and drones for filling gaps in confined spaces. The multimodal sensors are used to collect multi-source monitoring data.

[0064] Edge computing nodes are deployed on the airborne terminals of each drone in the drone swarm or on ground stations near the site to perform real-time preprocessing and in-situ self-calibration of multi-source monitoring data.

[0065] The cloud-based data processing center is equipped with an engineering adaptation module, a task scheduling module, and a data analysis module. The engineering adaptation module is used to match the corresponding monitoring strategy according to the engineering type of the target water conservancy project. The task scheduling module dynamically generates the cruise path and payload configuration instructions based on the monitoring strategy and sends them to the UAV cluster. The data analysis module has a built-in water conservancy project deformation interpretation model and water environment inversion model. It is used to calculate the dam deformation rate index and water pollution probability based on the calibrated monitoring data, and dynamically adjust the cruise path based on the calculation results.

[0066] In some embodiments, the multimodal sensing device includes a visible light camera, an infrared thermal imaging camera, a lidar sensor, a multispectral sensor, and a gas sensor.

[0067] Edge computing nodes have a built-in in-situ self-calibration module. Before performing a monitoring task, the drone is controlled to hover above a still body of water within the monitoring area to collect multispectral reflectance data from the water surface. The theoretical reflectivity was calculated based on a priori library of optical properties of water. Obtain the correction gain coefficient = / .

[0068] Edge computing nodes are also used for Gaussian filtering-based noise reduction of visible light images, and utilize... After gain compensation of the image grayscale values, an image sharpness score is calculated. :

[0069] = ; (1)

[0070] in, The corrected coordinates are ( The grayscale value of the pixel at position ) Its gradient, × For image resolution; when Below the preset threshold At that time, the drone is triggered to adjust its focus or flight altitude until... ≥ .

[0071] In some embodiments, the drone swarm includes four groups of patrol drones that are all communicatively connected to an edge computing node and a cloud data processing center, and each patrol drone is configured differently according to a monitoring strategy. The in-situ self-calibration module dynamically calibrates the corresponding drone's sensors using the stagnant water body it passes through before each cross-regional patrol, thereby achieving collaborative monitoring of all elements of the water conservancy project.

[0072] In some embodiments, the routine inspection drone is configured as a cluster host, and its mounted visible light camera is a zoom visible light camera, which together with an infrared thermal imaging camera forms a dual-gimbal monitoring device.

[0073] Edge computing nodes use the water surface temperature collected by the in-situ self-calibration module as a reference benchmark to dynamically correct the emissivity parameters of the infrared thermal imaging camera, and perform temperature matrix analysis on the thermal infrared image. The relative temperature difference on the dam surface is calculated using the following formula. To identify low-temperature leakage zones:

[0074] = - ; (2)

[0075] in, It is the first The temperature value of each pixel (obtained by sensors on the drone). It is the average background temperature within the current monitoring window (the water surface temperature collected by the in-situ self-calibration module). When the temperature is greater than 2℃, it is marked as a suspected leakage area, and the data analysis module is triggered to increase the cyclic sampling frequency of the area.

[0076] In some embodiments, the surveying and monitoring drone is equipped with a three-in-one lidar monitoring device, and the data analysis module is configured to perform the following dam deformation analysis process: acquiring lidar point cloud data for the current period. Cloud data with historical benchmarks ;right and Perform ICP (Iterative Closest Point) registration and calculate the 3D displacement D of corresponding points. ; Calculate the dam deformation rate index based on displacement :

[0077] = ; (3)

[0078] in, and These are the coordinates of the base period and the current period, respectively. For time intervals, when Exceeding the preset security threshold At that time, the system issues a deformation warning.

[0079] In some embodiments, the water quality inspection drone is equipped with multispectral monitoring equipment, and the water environment inversion model in the data analysis module is configured to execute the following water quality assessment algorithm: calling the correction gain coefficient output by the edge computing node. Correct the original reflectivity:

[0080] = , = ; (4)

[0081] The Normalized Difference Vegetation Index (NDVI) is calculated based on the corrected reflectance to retrieve chlorophyll concentration.

[0082] NDVI= ; (5)

[0083] in, For near-infrared reflectivity, Infrared reflectance; combined with turbidity inversion algorithm, water pollution probability is calculated. :

[0084] = ; (6)

[0085] in, It is a turbidity parameter calculated based on the blue-green light band. , , The preset weight parameters, It is the Sigmoid activation function; when If the value is greater than 0.85, the water area is considered to be at risk of algal bloom or oil film pollution.

[0086] In some embodiments, the confined space blind spot filling drone is a micro multi-rotor drone with a wheelbase of less than 20 cm, and its mounted gas sensors include at least a CO sensor and Sensors. The data analysis module has a built-in gas diffusion early warning model for calculating the comprehensive gas hazard index in enclosed spaces. :

[0087] = ; (7)

[0088] in, and To monitor concentration in real time, and It is the occupational exposure limit, when When the value is greater than 1, the system prohibits personnel from entering and triggers the drone to conduct a retest.

[0089] In some embodiments, the task scheduling module is configured to dynamically adjust the cruise path using an adaptive AI (artificial intelligence) algorithm, the cost function of which... Defined as:

[0090] = + + ; (8)

[0091] in, The actual path cost from the origin to node n. For heuristic functions, Let n be the risk weight value of node n. As a risk adjustment factor, the task scheduling module takes priority. In areas with high values, hover calibration nodes that pass through stagnant water bodies are forcibly inserted into the path planning.

[0092] In some embodiments, when the engineering adaptation module determines that the target project is a dam project, the monitoring system is configured to execute the following monitoring strategy: the task scheduling module prioritizes scheduling routine inspection drones to conduct close-range patrols of the dam body, calculates the low-temperature seepage zone, and uses in-situ self-calibrated infrared data to calculate the low-temperature seepage zone; at preset intervals, a surveying and monitoring drone is scheduled to collect three-dimensional point cloud data and calculate the dam body deformation rate index. During the flood season or under conditions of sudden changes in reservoir water level, a blind spot-filling drone is triggered to enter the corridor and gate well to calculate the comprehensive gas hazard index. The collected wide-angle images are then transmitted back to the cloud for crack and defect identification.

[0093] In some embodiments, when the engineering adaptation module determines that the target project is a river embankment project, the monitoring system is configured to execute the following monitoring strategy: the task scheduling module schedules daily inspection drones to conduct long-distance strip-shaped patrols along the embankment line, uses edge computing nodes to identify cracks and animal burrows in the embankment in real time, and dynamically adjusts the patrol density for dangerous sections; the task scheduling water quality inspection drones perform spectral scanning of the river cross-section, performs in-situ self-calibration using nearby water bodies before each scan begins, and then calculates the probability of water pollution. Furthermore, under high water level conditions, the infrared thermal imaging camera of the daily inspection drone is used to conduct initial screening for piping and seepage at the toe of the dike and the structures penetrating the dike based on temperature differences.

[0094] Compared with the prior art, the present invention has the following beneficial effects.

[0095] (1. Based on the in-situ self-calibration mechanism of water medium, the static water body is used as a natural radiation calibration source. The gain coefficient is dynamically calculated to compensate for multispectral and infrared data in real time. This effectively overcomes the distortion problems caused by sensor data drift and lens contamination in high humidity and high fog environments, and improves the long-term accuracy of water quality inversion and leakage identification by more than 15%.

[0096] (2) By introducing the water surface temperature as a reference blackbody, the emissivity parameters of the infrared thermal imaging camera are dynamically corrected, which solves the problem of interference from changes in atmospheric transmittance in traditional temperature measurement. Combined with the local window temperature difference algorithm, the reflection of water surface and environmental radiation noise are significantly eliminated, which increases the confidence of identifying the low temperature zone of dam seepage to more than 95% and greatly reduces the false alarm rate.

[0097] (3) By forcibly embedding the in-situ self-calibration node into the UAV cruise path planning, an integrated edge computing architecture of "flight-calibration-monitoring" was constructed. This not only eliminated the measurement errors accumulated in long-distance inspections, but also reduced the bandwidth occupation of the original invalid data backhaul, enabling the system to maintain high-precision and low-latency stable operation in remote water conservancy hubs with limited communication.

[0098] The technical solution of the present invention will be verified and explained through specific examples below.

[0099] Example 1: Refined Deformation and Leakage Monitoring for High Dam Projects

[0100] Taking a concrete gravity dam as an example, this scenario presents technical challenges such as the dam's towering height, monotonous surface texture, enclosed corridor environment, and sensor drift caused by high humidity.

[0101] Step 1: Anti-interference image acquisition and preprocessing based on in-situ self-calibration

[0102] To address the technical challenges of blurred images due to high wind speeds in high dam environments and color distortion caused by water vapor in the lens, the system has activated routine inspection drones.

[0103] First, the edge computing nodes execute an in-situ self-calibration algorithm: controlling the drone to hover above the calm water surface of the reservoir and collecting multispectral reflectance data from the water surface. Theoretical reflectivity is calculated based on a priori library of optical properties of water. Obtain the correction gain coefficient = / Subsequently, using Pixel-by-pixel gain compensation is performed on the raw images acquired by the visible light camera. Based on this, an image sharpness evaluation algorithm is executed in real time, and a sharpness score is calculated based on equation (1). If caused by airflow disturbance The system triggers the drone to adjust its gimbal damping or change its flight attitude for adaptive compensation; if the compensation is still insufficient, the drone is instructed to adjust its focal length or lower its flight altitude to re-collect data. This closed-loop control mechanism of "calibrate first, then evaluate, then compensate" ensures the validity of the data input to the cloud from the source.

[0104] Step 2: Leakage Identification Based on Temperature Difference Analysis and Emissivity Correction

[0105] To address the technical challenges of concealed seepage points in the dam body and the susceptibility of infrared thermal imaging to interference from stray light from the water surface, the system utilizes an infrared thermal imaging camera to acquire a temperature matrix. Specifically, this step uses the water surface temperature obtained in step one as a reference blackbody to dynamically correct the emissivity parameter ϵ of the infrared thermal imaging camera, eliminating temperature measurement errors caused by changes in atmospheric transmittance.

[0106] The data analysis module calculates the relative temperature difference based on equation (2). Considering the natural temperature difference interference between the sunny and shady sides of the dam, this system introduces the concept of a dynamic background temperature field (i.e., (This is the local window mean). When When the temperature is greater than 2℃, it is identified as a suspected low-temperature leakage zone, and the drone is triggered to reduce its speed to conduct staring monitoring of the area. Instantaneous interference such as water surface reflection is eliminated through continuous frame analysis.

[0107] Step 3: Millimeter-level deformation interpretation

[0108] To address the technical challenges of time-consuming large-scale point cloud data processing and the difficulty in providing real-time early warnings, surveying and monitoring drones collect current point cloud data. The cloud-based data analysis module retrieves historical benchmark points. ICP registration was performed. The three-dimensional displacement of the corresponding points was calculated, and then the deformation rate index of the dam body was obtained based on equation (3). By With safety threshold The comparison enabled the transformation from "static elevation measurement" to "dynamic rate early warning".

[0109] Step 4: Safety Assessment of Enclosed Spaces

[0110] To address the issues of uneven gas diffusion within corridors and the susceptibility of single-type gas sensors to false alarms, a blind spot-filling drone equipped with CO and [other gas sensors] is being developed for confined spaces. Sensors enter the corridor, and the comprehensive gas hazard index is calculated based on equation (7). Equation (7) significantly reduces the false alarm rate and improves the safety of inspections by weighted fusion of the concentration data of the two gases.

[0111] Example 2: Comprehensive Survey of Hidden Dangers in River Embankments Over Long Distances

[0112] This embodiment addresses the challenges of long and wide river embankments, sudden emergencies, and variable atmospheric transmittance along the riverbanks. It focuses on resolving issues such as low inspection efficiency, delayed pollution detection, and distorted spectral data.

[0113] Step 1: Path dynamic planning based on risk weights and mandatory calibration nodes

[0114] Traditional patrol routes are fixed and cannot target dangerous sections for focused inspection. The task scheduling module of this invention employs an improved adaptive AI algorithm, the cost function of which is shown in equation (8). Unlike existing technologies, this invention forcibly inserts hovering nodes passing through still water bodies (such as puddles or ponds) during path planning as trigger conditions for in-situ self-calibration. During long-distance flights, the UAV automatically calibrates at regular intervals to ensure data consistency throughout the entire flight. This algorithm prioritizes traversing high-risk areas, improving inspection efficiency by more than 30%.

[0115] Step 2: Gain-corrected multispectral inversion of the water environment

[0116] To address the shortcomings of traditional water quality monitoring, which relies on manual sampling, suffers from poor timeliness, and is susceptible to light interference, water quality inspection drones perform spectral scanning. The system first retrieves the correction gain coefficient obtained from the most recent hovering calibration. Based on equation (4), the original red and near-infrared reflectances are corrected. Then, based on the corrected reflectances, the normalized difference vegetation index (NDVI) is calculated using equation (5). Subsequently, the turbidity parameter is combined with the normalized difference vegetation index (NDVI). Using the Sigmoid activation function, the probability of water pollution is calculated using equation (6). Because of the introduction After calibration, this model effectively eliminates interference from changes in solar irradiance and lens fouling, and can smoothly reflect the trend of water quality deterioration. When When the value is greater than 0.85, the system can not only determine pollution, but also preliminarily distinguish between algal blooms and oil film pollution.

[0117] Step 3: Collaborative screening for piping vulnerabilities

[0118] Under high water levels, the system coordinates routine inspection drones and water quality monitoring drones. Emissivity-corrected infrared thermal imaging cameras scan the levee toe to locate low-temperature anomalies caused by seepage; simultaneously, visible light cameras detect surface cracks. When a crack and a low-temperature zone overlap geographically, the system significantly increases the confidence level, classifying it as a suspected piping channel, achieving a comprehensive, three-dimensional investigation integrating air, ground, and air surveillance.

[0119] Example 3: Edge-Cloud Collaborative Computing and In-Situ Self-Calibration Mechanism

[0120] As the underlying support for the aforementioned embodiments, this embodiment elaborates on the system's computing architecture to solve the problems of bandwidth congestion and sensor drift caused by massive data backhaul.

[0121] The system employs a hierarchical computing strategy. At the edge (UAV / Ground station): lightweight, high real-time algorithms are executed. This includes: in-situ self-calibration calculation: calculating the correction gain coefficient. Image sharpness evaluation: Calculation of scores Gas index calculation: Calculation Edge nodes perform preliminary filtering and correction on the data, only including corrected feature values ​​(such as...). Maximum value, after calibration Instead of transmitting the original video stream back, this greatly saves bandwidth. Heavyweight, computationally intensive algorithms, such as point cloud ICP registration and deformation analysis, are executed in the cloud.

[0122] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A water body oriented unmanned aerial vehicle cluster water conservancy project monitoring system for in-situ self-calibration, characterized in that, The system includes: The drone swarm and the multimodal sensing device mounted on each drone, wherein the drone swarm includes routine inspection drones, surveying and monitoring drones, water quality inspection drones and blind spot drones in confined spaces, and the multimodal sensing device is used to collect multi-source monitoring data; Edge computing nodes, deployed on the airborne end of the UAV cluster or on a ground station near the site, are used to perform real-time preprocessing and in-situ self-calibration of the multi-source monitoring data output by the multimodal sensing devices; The cloud-based data processing center is equipped with an engineering adaptation module, a task scheduling module, and a data analysis module. The engineering adaptation module matches the corresponding monitoring strategy based on the engineering type of the target water conservancy project. The task scheduling module dynamically generates cruise paths and payload configuration instructions based on the monitoring strategies and sends them to the UAV cluster. The data analysis module incorporates a water conservancy project deformation interpretation model and a water environment inversion model, used to calculate the dam deformation rate index and water pollution probability based on calibrated monitoring data, and dynamically adjust the cruise path based on the calculation results.

2. The system of claim 1, wherein, The multimodal sensing devices include: a visible light camera and an infrared thermal imaging camera mounted on the daily inspection drone, a lidar sensing device mounted on the surveying and monitoring drone, a multispectral sensing device mounted on the water quality inspection drone, and a gas sensor mounted on the confined space blind spot filling drone.

3. The system of claim 1 or 2, wherein, The edge computing node has a built-in in-situ self-calibration module, which is used to control the UAV to hover above the still water body in the monitoring area before performing the monitoring task, collect the multispectral reflectance and surface temperature of the water body, and calculate the correction gain coefficient based on the theoretical reflectance calculated from the prior library of optical properties of the water body. = / ； wherein, denotes a correction gain coefficient, denotes the acquired multi-spectral reflectance of the water surface, denotes the theoretical reflectance.

4. The system according to claim 3, characterized in that, The edge computing node is also used to perform Gaussian filtering-based noise reduction processing on the visible light images output by the visible light camera on the daily inspection drone, and to perform gain compensation on the image grayscale values ​​using the correction gain coefficient, and then calculate the image sharpness score: = ； in, Rate the image sharpness. To correct the grayscale values ​​of the pixels, gradient, × For image resolution; when When the image sharpness score is below a preset threshold, the drone is triggered to adjust its focus or flight altitude until the image sharpness score is greater than or equal to the preset threshold. The edge computing node also uses the water surface temperature collected by the in-situ self-calibration module as a reference to calculate the relative temperature difference on the dam surface: = - ； in, For the first monitoring via sensors on the drone Temperature value of each pixel The water surface temperature is collected by the in-situ self-calibration module; when When the temperature exceeds the preset value, it is marked as a suspected leakage area, and the data analysis module is triggered to increase the cyclic sampling frequency of that area.

5. The system according to claim 3, characterized in that, The data analysis module is equipped with a water environment inversion model, which is configured to perform water quality assessment based on the light monitoring data output by the multispectral sensing device on the water quality inspection drone. The water quality assessment includes: The original light reflectance in the light monitoring data is corrected based on the correction gain coefficient: = ， = ； in, This represents the original infrared reflectivity. This is the corrected infrared reflectivity; This represents the original infrared reflectivity. This is the corrected infrared reflectivity; Normalized difference vegetation index was calculated based on the corrected infrared and near-infrared reflectance to retrieve chlorophyll concentration: NDVI= ; NDVI is the Normalized Difference Vegetation Index. Based on the normalized difference vegetation index and combined with the turbidity inversion algorithm, the probability of water pollution is calculated: = ； in, For the probability of water pollution, The turbidity parameters are calculated based on the blue-green light band. , , For weight parameters, It is the Sigmoid activation function; When the probability of water pollution is greater than a preset threshold, it is determined that the water area is at risk of algal bloom or oil film pollution.

6. The system according to claim 3, characterized in that, The data analysis module is configured to perform a dam deformation analysis process based on the radar data output by the lidar monitoring device. Acquire point cloud data collected by the lidar monitoring equipment in the current period and historical baseline point cloud data; Iterative nearest point (ICP) registration is performed on point cloud data and historical baseline point cloud data to calculate the 3D displacement of corresponding points; Calculate the dam deformation rate index based on the aforementioned three-dimensional displacement: = ； in, This is the dam deformation rate index. and These are the coordinates of the base period and the current period, respectively. For time intervals; When the dam deformation rate index exceeds a preset safety threshold, a deformation warning is issued.

7. The system according to claim 3, characterized in that, The confined space gap-filling drone is a micro multi-rotor drone, and its onboard gas sensors include at least a CO sensor and... sensor; The data analysis module has a built-in gas diffusion early warning model for calculating the comprehensive gas hazard index in enclosed spaces. = ； in, To calculate the comprehensive gas hazard index, and To monitor concentration in real time, and This is the occupational exposure limit; when the comprehensive gas hazard index is greater than the preset hazard index, personnel are prohibited from entering and a drone is triggered for retesting.

8. The system according to claim 1, characterized in that, The task scheduling module is configured to dynamically adjust the cruise path using an adaptive artificial intelligence (AI) algorithm, the cost function of which is... Defined as: = + + ； in, The actual path cost from the origin to node n. For heuristic functions, Let n be the risk weight value of node n. As a risk adjustment factor, the task scheduling module prioritizes... In areas with high values, hover calibration nodes that pass through stagnant water bodies are forcibly inserted into the path planning.

9. The system according to claim 1, characterized in that, When the engineering adaptation module determines that the target project is a dam project, the following monitoring strategy is executed: The task scheduling module prioritizes scheduling the daily inspection drone to conduct close-range patrols of the dam body, calculates the low-temperature seepage zone, and uses in-situ self-calibrated infrared data to calculate the low-temperature seepage zone. At preset intervals, the surveying and monitoring drone is scheduled to collect three-dimensional point cloud data and calculate the dam deformation rate index. During the flood season or under conditions of sudden changes in reservoir water level, the narrow space gap-filling drone is triggered to enter the corridor and gate well, calculate the comprehensive gas hazard index, and transmit the collected wide-angle images back to the cloud for crack and defect identification.

10. The system according to claim 1, characterized in that, When the engineering adaptation module determines that the target project is a river embankment project, the following monitoring strategy is executed: The task scheduling module schedules the daily inspection drone to conduct long-distance strip patrols along the dike line, uses the edge computing node to identify dike cracks and animal burrows in real time, and dynamically adjusts the patrol density for dangerous sections. The water quality inspection drone is dispatched to perform spectral scanning of the river section. Before each scan begins, in-situ self-calibration is performed using adjacent water bodies, and the probability of water pollution is calculated. Under high water level conditions, the infrared thermal imaging camera of the daily inspection drone is linked to conduct initial screening of piping and seepage at the embankment toe and structures penetrating the embankment based on temperature difference.

Images