Irrigation area liquid level unmanned aerial vehicle detection system fused with image recognition

Abstract

The invention discloses an irrigation area liquid level unmanned aerial vehicle detection system fused with image recognition, and relates to the technical field of liquid level detection, and the system comprises an image recognition processing module which comprises a marker detection unit, a liquid level line recognition unit and a geometric calculation unit. The marker detection unit carries out target area matching on input irrigation area image data and outputs marker pixel coordinates and type codes, the liquid level line recognition unit demarcates a detection area based on the marker pixel coordinates and outputs liquid level line pixel coordinates, and the geometric calculation unit calculates liquid level data through space projection transformation. By designing the image recognition processing module, the functions of automatic recognition of irrigation area image data and high-precision calculation of the liquid level are achieved, the problems that traditional manual inspection efficiency is low and the deployment and maintenance cost of a fixed sensor is high are solved, and the automation degree and measurement precision of irrigation area liquid level monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of liquid level detection technology, specifically to a drone-based liquid level detection system for irrigation areas that integrates image recognition. Background Technology

[0002] With the rapid development of modern agricultural water conservancy projects in my country, the demand for refined management and efficient utilization of water resources in large irrigation areas is becoming increasingly urgent. Liquid level data from key nodes such as irrigation canals, reservoirs, and water storage ponds are core foundational data for implementing precision irrigation, optimizing water allocation, and preventing droughts and floods. Therefore, water conservancy management departments are continuously promoting the automation and intelligent upgrading of irrigation area monitoring technologies to replace traditional and outdated manual inspection methods.

[0003] Currently, water level monitoring in irrigation areas in my country mainly relies on manual handheld measuring devices for inspection and the construction of fixed water level sensor networks. Manual inspection suffers from low efficiency, limited coverage, poor data continuity, and poor accessibility in harsh environments, and cannot meet the needs of real-time monitoring in large-scale irrigation areas. While deploying fixed sensor networks can achieve continuous measurement, its construction and maintenance costs are extremely high, and it is difficult to achieve comprehensive coverage in vast irrigation areas with complex terrain. Sensors are susceptible to water corrosion, siltation, and biological adhesion, leading to inaccurate measurements or equipment damage. Communication cabling also presents significant challenges in remote areas. Furthermore, existing methods using ordinary cameras for visual ranging are easily affected by water surface reflections, wave motion, weather changes, and complex background interference in irrigation areas, making it difficult to guarantee measurement accuracy and stability, and failing to meet the precise requirements of irrigation management.

[0004] Patent CN113744325B discloses a liquid level detection device and method based on image recognition technology. The above patent achieves accurate liquid level detection regardless of the liquid's own characteristics and environmental factors.

[0005] The aforementioned patented container is an inverted frustum, with the radius of the liquid surface inside the container uniquely determining the liquid level. A camera captures images of the liquid level detection, including the top opening of the container under illumination, the liquid surface inside the container, and the inner wall of the container between the top opening and the liquid surface. The image recognition system inputs the liquid level detection images into a trained YOLO-v3 detection model, outputting the position coordinates of the bounding rectangle of the top opening and the bounding rectangle of the liquid surface. Based on these coordinates and the container's dimensions, the liquid level is determined, achieving accurate liquid level detection regardless of the liquid's inherent characteristics and environmental factors. However, there is still room for improvement in the accurate detection of liquid levels in irrigation areas.

[0006] Therefore, this application proposes an unmanned aerial vehicle (UAV) system for detecting irrigation levels in irrigation areas, which integrates image recognition to achieve accurate detection of irrigation levels. Summary of the Invention

[0007] The purpose of this invention is to provide an unmanned aerial vehicle (UAV) detection system for irrigation area liquid level that integrates image recognition, so as to solve the technical problem of difficulty in ensuring the accuracy and stability of irrigation area liquid level measurement mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a drone-based irrigation area liquid level detection system integrating image recognition, comprising an image recognition processing module, wherein the image recognition processing module is used to calculate the liquid level value based on irrigation area image data collected by the drone;

[0009] The image recognition processing module includes: a marker detection unit, a liquid level recognition unit, and a geometric calculation unit;

[0010] The marker detection unit is equipped with a pre-stored marker feature library. Based on the standard pattern data in the feature library, it performs target region matching on the input irrigation area image data and outputs the marker pixel coordinates and type code.

[0011] The liquid level line recognition unit delineates the detection area based on the pixel coordinates of the marker, extracts the continuous horizontal contour through the edge enhancement algorithm, filters the liquid surface boundary by combining the regional grayscale gradient distribution, and outputs the pixel coordinates of the liquid level line.

[0012] The geometric calculation unit calls the corresponding actual size parameters according to the marker type code, and combines the UAV flight parameter data and camera calibration parameters to construct a spatial mapping model between the marker pixel coordinates and the actual height, and calculates the liquid level data through spatial projection transformation.

[0013] Preferably, the irrigation area image data required by the marker detection unit is provided by the image acquisition unit of the UAV platform;

[0014] The drone platform also includes a flight control unit and a wireless transmission unit. The flight control unit is connected to the image acquisition unit and the wireless transmission unit via signals.

[0015] The flight control unit controls the flight status of the UAV according to the route instructions received by the wireless transmission unit, so that the UAV passes through each irrigation area monitoring point along the preset path;

[0016] The image acquisition unit is equipped with a high-resolution camera to collect image data of the irrigation area, including markers, when the drone flies over the monitoring point of the irrigation area;

[0017] The wireless transmission unit transmits the filtered liquid level data to the ground control center and receives control commands from the ground control center.

[0018] Preferably, the route planning unit of the ground control center provides route instructions to the flight control unit of the UAV platform;

[0019] The ground control center also includes a data display unit and a monitoring point division unit. The monitoring point division unit is connected to the data display unit and the flight path planning unit via signals.

[0020] The monitoring points are divided into units based on the topographical features, hydrological features, and key areas of irrigation management in the irrigation area.

[0021] The flight path planning unit generates the UAV flight path based on the distribution of monitoring points in the irrigation area and sends the flight path instructions to the wireless transmission unit of the UAV platform.

[0022] The data display unit receives liquid level data transmitted by the wireless transmission unit of the drone platform and displays it in the form of charts, so that operators can intuitively obtain the liquid level distribution in the irrigation area.

[0023] Preferably, the pattern feature data required by the feature feature library configured in the marker detection unit is provided by the feature template storage unit of the marker parameter library module;

[0024] The marker parameter library module also includes a geographic coordinate storage unit and a physical size storage unit;

[0025] The geographic coordinate storage unit records the latitude, longitude, and elevation data of each marker installation point, providing navigation and positioning references to the flight control unit of the UAV platform;

[0026] The physical size storage unit associates the identifier type encoding to store the actual width and height values, and provides the size conversion parameters of the spatial mapping model to the geometric calculation unit of the image recognition processing module;

[0027] The feature template storage unit stores the standard binary contour matrix of the marker and provides target matching pattern feature data to the marker feature library configured in the marker detection unit of the image recognition processing module.

[0028] Preferably, the flight parameter data required by the geometric calculation unit is provided by the attitude perception unit and the positioning unit of the UAV state monitoring module;

[0029] The UAV status monitoring module also includes a fault diagnosis unit, which is connected to the attitude sensing unit and the positioning unit via signals.

[0030] The positioning unit acquires the real-time position information of the drone to check whether it deviates from the preset route. At the same time, it is equipped with a laser rangefinder to provide the drone's altitude data to the geometric calculation unit of the image recognition and processing module.

[0031] The attitude perception unit collects the flight attitude data of the UAV and provides it to the geometric calculation unit of the image recognition and processing module. The flight attitude data includes pitch angle, roll angle and yaw angle data.

[0032] The fault diagnosis unit analyzes real-time location information and flight attitude data, and generates an alarm signal when an abnormal state is detected and sends it to the ground control center.

[0033] Preferably, the liquid level data calculated by the geometric calculation unit is filtered by the preprocessing unit of the data processing module;

[0034] The data processing module also includes an error analysis unit and a parameter correction unit. The data preprocessing unit is connected to the error analysis unit and the parameter correction unit via signals.

[0035] The preprocessing unit filters the liquid level data calculated by the geometric calculation unit to remove outliers caused by image noise, and then transmits the processed liquid level data to the wireless transmission unit.

[0036] The error analysis unit receives the filtered liquid level data from the data preprocessing unit, compares it with the reference liquid level value obtained by on-site calibration at the marker installation point using a high-precision measuring instrument, calculates the deviation between the liquid level data measured by the UAV and the reference liquid level value, and generates a deviation analysis report.

[0037] Based on the deviation analysis report output by the error analysis unit, when the average deviation of multiple consecutive measurements exceeds a set threshold, the parameter correction unit generates correction instructions for the camera calibration parameters and spatial mapping model, and feeds the correction instructions back to the geometric calculation unit of the image recognition processing module.

[0038] Preferably, the marker has an appearance that contrasts with the surrounding environment and includes a geometric structure with known physical dimensions. The marker is made of weather-resistant material and is fixedly installed on the side wall of the irrigation canal and on the stable structure of key monitoring points.

[0039] High-contrast appearance features regular patterns formed by alternating dark and light areas;

[0040] A geometric structure with known physical dimensions includes precisely defined scale lines and standard geometry with standard side lengths;

[0041] The regular patterns and standard geometric structures of the markers together constitute the feature templates for machine vision recognition, enabling the marker detection unit to achieve robust recognition and accurate positioning in complex backgrounds.

[0042] Preferably, the process by which the geometric calculation unit performs spatial projection transformation to calculate liquid level data includes:

[0043] Based on the received identifier type code, the corresponding actual size parameters of the identifier are retrieved from the identifier parameter library module;

[0044] Acquire real-time flight parameter data of the UAV provided by the UAV status monitoring module, including three-dimensional spatial coordinates and flight attitude angles;

[0045] Read the preset camera internal calibration parameters;

[0046] Based on the pixel coordinates of the markers in the image and their corresponding actual size, combined with flight parameter data and camera calibration parameters, a spatial mapping relationship from the two-dimensional image pixel coordinate system to the three-dimensional world coordinate system is constructed.

[0047] Substituting the liquid level pixel coordinates output by the liquid level recognition unit into the spatial mapping relationship, the three-dimensional spatial coordinates of the liquid point are calculated through coordinate transformation and geometric operations, and its elevation component is the final calculated liquid level data.

[0048] Preferably, the flight parameter data on which the geometric calculation unit performs the spatial projection transformation specifically includes ground altitude data provided by the positioning unit of the UAV status monitoring module and pitch angle, roll angle and yaw angle data provided by the attitude perception unit; the camera calibration parameters specifically include the camera focal length, principal point coordinates and radial distortion coefficient obtained in advance by Zhang Zhengyou calibration method.

[0049] Preferably, the judgment condition for the average deviation of multiple consecutive measurements in the error analysis unit exceeding the set threshold is as follows: the system continuously monitors a specific marker point, and when the absolute value of the difference between the arithmetic mean of the measured values ​​of the point in the most recent N inspections and the pre-stored reference liquid level value is continuously greater than M cm, and its standard deviation is less than K cm, it is determined that there is a systematic deviation; where N is an integer greater than 5, and M and K are positive real numbers set according to the measurement accuracy requirements of the irrigation area.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] 1. This invention, through the design of an image recognition processing module, realizes the functions of automatic recognition of irrigation area image data and high-precision liquid level calculation, solving the problems of low efficiency of traditional manual inspection and high cost of deployment and maintenance of fixed sensors, and improving the automation level and measurement accuracy of irrigation area liquid level monitoring;

[0052] 2. This invention, through the design of an unmanned aerial vehicle (UAV) platform, realizes automated patrol and image acquisition functions in irrigation areas with large-scale and complex terrain, solving the problems of limited coverage and numerous blind spots of fixed monitoring points, and improving the efficiency and scope of monitoring operations;

[0053] 3. This invention, through the design of a data processing module, realizes the functions of filtering and denoising the raw liquid level data, error analysis, and self-calibration of system parameters, which solves the pain point of poor long-term reliability of measurement data caused by equipment drift and environmental interference, and improves the stability and accuracy of the entire system.

[0054] 4. This invention provides machine vision algorithms with a highly robust recognition target and absolute size reference by designing a specially designed marker. It solves the core problems of difficulty in liquid level line recognition and measurement instability caused by complex lighting, waves and backgrounds in natural scenes, thereby improving the success rate of image recognition and the reliability of liquid level calculation. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the structure of an unmanned aerial vehicle (UAV) system for detecting water levels in irrigation areas, which integrates image recognition, as proposed in this invention.

[0056] Figure 2 This is a schematic diagram of the process of an unmanned aerial vehicle (UAV) system for detecting water levels in irrigation areas, which integrates image recognition, as proposed in this invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see Figure 1 and Figure 2This invention provides an embodiment of an irrigation area liquid level UAV detection system that integrates image recognition. The system includes a marker detection unit, a liquid level line recognition unit, and a geometric calculation unit. The marker detection unit uses a pre-stored marker feature library to match and identify markers in irrigation area images captured by the UAV, outputting the precise pixel position and type of the markers. The liquid level line recognition unit uses the marker positions as a reference, employs an edge detection algorithm to enhance horizontal edges, and combines the grayscale statistical features above and below the area to accurately select the liquid level boundary line, outputting its pixel coordinates. The geometric calculation unit calls the actual physical dimensions of the markers according to their type, and combines the UAV's flight parameter data and camera calibration parameters to construct a perspective projection transformation model from the image's two-dimensional pixel coordinate system to the real-world three-dimensional coordinate system. The pixel coordinates of the liquid level line are substituted into this model to finally calculate the absolute liquid level elevation value in meters. The flight parameter data specifically includes ground clearance data provided by the UAV's status monitoring module's positioning unit and pitch, roll, and yaw angle data provided by the attitude perception unit. The camera calibration parameters specifically include the camera focal length, principal point coordinates, and radial distortion coefficient, which are obtained in advance using the Zhang Zhengyou calibration method.

[0059] Furthermore, the marker detection unit is equipped with a pre-stored marker feature library, which contains standard pattern data of various pre-designed markers (such as ArUco codes, checkerboard patterns, and targets of specific shapes). Based on this standard data, the unit performs target region matching and recognition on irrigation area images transmitted by UAVs. It preferably adopts a target detection network based on deep learning, such as YOLOv5s or MobileNet-SSD. These networks achieve a good balance between speed and accuracy and are very suitable for real-time inference on airborne embedded AI computing platforms such as JetsonNano. Before implementation, a dataset of tens of thousands of marker images taken under different lighting, weather, and angle conditions needs to be constructed and accurately labeled to train the network. The trained model can effectively resist partial occlusion, changes in light and shadow, and background interference, and accurately output the pixel coordinates and type codes of the identified markers in the image. There are more than one type of marker deployed in the irrigation area. For example, large scale rulers are used in the main canals and small ArUco code targets are used in the branch canals. The type codes correspond to the actual physical size.

[0060] Based on the pixel coordinates of the marker obtained above, the liquid surface line recognition unit delineates a narrow horizontal detection area near it to eliminate most interference from irrelevant backgrounds. Within this area, an edge enhancement algorithm (such as the Canny operator) is used to extract all continuous contours in the horizontal direction. Then, combined with the gray-level gradient distribution characteristics of the upper and lower parts of this area (typically, the gradient change is gentle above the water surface, while the gradient change below the water surface is complex due to waves and suspended matter), the most likely liquid surface boundary is intelligently selected, and its precise pixel coordinates are output. For calm water surfaces, a classic image processing algorithm chain can be used: first, Gaussian filtering is applied to the horizontal detection area for noise reduction; then, the Canny operator is used for edge detection; then, Hough transform is used to detect straight lines; finally, the most suitable horizontal line is selected from multiple horizontal lines as the liquid surface line. For complex scenes with waves, reflections, and floating objects, the above methods are prone to failure. Therefore, the present invention prefers an improved scheme: First, perform temporal averaging on consecutive frames of images to effectively smooth random wave noise. Then, use a method based on Otsu thresholding and region growing to initially divide the image into "water" and "non-water" regions. Alternatively, a lightweight pixel classification model (such as using LoG filters, Local Binary Pattern (LBP) features to train a Support Vector Machine (SVM)) can be trained to classify each pixel as belonging to the water surface or the background. Finally, perform polynomial fitting on the lower boundary of the segmented water surface region to obtain a smooth and accurate liquid surface line. Although this scheme has a slightly higher computational cost, it can greatly improve the recognition success rate and accuracy under harsh conditions.

[0061] Geometric calculations are crucial for the accuracy of the entire process. The core involves using the world coordinates (calculated from the physical dimensions) of several feature points on a known marker and their corresponding image pixel coordinates to solve for the camera's pose (extrinsic parameters). This process relies precisely on ground-level altitude data (e.g., from a laser rangefinder) and complete attitude angle data (pitch, roll, yaw) provided by the UAV's status monitoring module to construct an accurate camera extrinsic parameter matrix. The camera's intrinsic parameters (focal length, principal point, radial distortion coefficient) are precisely obtained beforehand using the Zhang Zhengyou calibration method. This method involves taking multiple images of a checkerboard calibration board with known geometric information from different angles to solve for the camera's internal parameter model, thereby correcting inherent lens distortion. Subsequently, points on the liquid surface are projected back onto the world coordinate system using the intrinsic and extrinsic parameter matrices. Since the liquid surface is horizontal... The elevation value is the desired liquid level. This process relies on precise camera calibration parameters and high-precision ground clearance and complete attitude data of the UAV, ultimately achieving centimeter-level liquid level measurement accuracy, fully meeting the needs of refined management in irrigation areas. The geometric calculation unit, based on the identifier type code, retrieves the corresponding actual size parameters of the identifier from the system database (e.g., the physical side length of each black and white square is 0.1 meters). Combining the flight parameter data synchronously transmitted by the UAV with the internal parameters of the camera obtained through pre-calibration, a spatial mapping model from the two-dimensional pixel coordinate system of the image to the three-dimensional coordinate system of the real world is constructed. By applying this model to the pixel coordinates of the liquid surface line and performing spatial projection transformation, the absolute liquid level elevation data in meters is finally calculated. The calculation formula is as follows:

[0062]

[0063] Among them, H l The output is the absolute elevation of the liquid surface, H. d The input is the real-time ground altitude from the UAV's laser rangefinder sensor; y l The input is the y-coordinate of the liquid level line pixel output by the liquid level line recognition unit; r The input quantity is the pixel ordinate of a fixed reference point (such as the bottom center) on the marker, output by the marker detection unit; y l -y r S represents the pixel distance between the liquid surface and the reference point. r is the input value, representing the actual physical size of the reference point for the marker retrieved from the database; cosθ is the input value, representing the pitch angle of the UAV camera, with the camera pointing downwards as positive; f is the camera's focal length, a parameter obtained through internal calibration.

[0064] Suppose we install a simple bar marker on the wall of an irrigation canal, with a black reference block on it;

[0065] Identifier parameters: S r=0.20m;

[0066] Camera parameters: Through camera calibration, it is known that the camera's focal length is f=1000 pixels;

[0067] Drone Status: The positioning unit obtains the current drone's altitude H above the ground. d =30.0m, the attitude sensing unit collects the camera pitch angle cosθ=5°, which needs to be converted to radians for calculation: cos(0.0873)=0.9962;

[0068] The y-coordinate of the pixel of a reference point (such as the bottom) in the image. r =600 pixels (the origin of the image coordinate system is at the top left corner, and the y-axis is positive downwards);

[0069] The pixel y-coordinate of the liquid level line in the image l =750 pixels;

[0070] Substitute into the formula to calculate:

[0071] Substitute all known parameters into the formula to calculate:

[0072] Final liquid level elevation: H l =30.0-0.029886=29.970114m;

[0073] ArUco code: A square mark consisting of a black border and an internal binary matrix. It is similar to a QR code, but is more commonly used in camera calibration and object tracking in computer vision. Each ArUco code has a unique ID, which makes it easy for the system to quickly identify and retrieve its known physical dimensions.

[0074] Object detection networks (YOLOv5s, MobileNet-SSD): These are deep learning models used to quickly and accurately locate and classify target objects in images.

[0075] Edge enhancement algorithms (such as the Canny operator): A very popular image processing algorithm used to highlight the edges of objects in an image, i.e., where grayscale values ​​change drastically.

[0076] Hough Transform: An algorithm for detecting simple geometric shapes (such as lines and circles) in an image. It can be used to fit the most probable straight line as the liquid surface line from a large number of edge points detected by the Canny operator.

[0077] Otsu thresholding: an image binarization algorithm (transforming an image into only black and white pixels) that can automatically calculate an optimal grayscale threshold to separate the foreground (water) from the background (shore);

[0078] Region growing: An image segmentation algorithm that starts from a "seed point" and gradually merges surrounding pixels with similar attributes (such as grayscale and color) to form a connected region, which can be used to generate a complete water surface area from the initial segmentation results;

[0079] LoG filter and Local Binary Pattern (LBP) are two feature training methods.

[0080] Support Vector Machine (SVM): A classic machine learning classification algorithm where it can be trained to determine whether a pixel belongs to "water" or "non-water" based on the features surrounding each pixel.

[0081] Please see Figure 1 and Figure 2 The present invention provides an embodiment of an irrigation area liquid level UAV detection system that integrates image recognition. The system integrates a UAV platform, whose flight control unit receives flight route instructions from the ground control center and controls the UAV to autonomously fly to the preset irrigation area monitoring point. When the UAV reaches the target point, the flight control unit triggers the image acquisition unit to take pictures. The acquired images are transmitted back through the airborne data transmission radio and processed by the image recognition processing module. The calculated liquid level data is transmitted back to the data display unit of the ground control center through the wireless transmission unit and presented to the administrator in the form of spatiotemporal curves and channel profile diagrams.

[0082] Furthermore, the administrator first manually delineates or automatically generates a series of monitoring points on the electronic map of the ground station software, based on hydrological characteristics such as channel direction, gate location, and terrain undulations. The flight path planning unit then uses these points, employing a traveling salesman problem optimization algorithm or obstacle avoidance algorithm, to automatically generate a flight path that is the shortest in time, most energy-efficient, and safest. This path is then sent to the drone. The drone's flight control system, such as the Pixhawk, precisely controls its flight along this path, hovering (or cruising at low speed) briefly above each monitoring point to trigger the camera shutter via the flight control system's MAVLink protocol. To ensure image quality, the camera should be set to shutter priority mode and a high shutter speed (such as 1 / 1000 second) should be used to freeze the water surface ripples. To solve the problem of overexposure of the water surface under sunlight, a ring polarizing filter can be installed in front of the lens to effectively suppress reflection. The entire mission process does not require manual intervention. The drone can automatically complete the entire process of take-off, cruise, shooting, data transmission and return landing. One administrator can manage multiple drones at the same time to inspect different irrigation areas, shortening the work that traditionally required several people to complete in a day to less than an hour, increasing efficiency by more than ten times, and completely avoiding the safety risks of manual inspection.

[0083] Traveling Salesman Problem Optimization Algorithm: Given a series of cities and distances, find the shortest path to visit all cities and return to the starting point. In drone path planning, "city" is "monitoring point".

[0084] Obstacle avoidance algorithms: Algorithms that enable drones to perceive and automatically bypass obstacles (such as trees and power lines) in front of them in real time, typically relying on visual sensors, lidar, or ultrasonic sensors.

[0085] Pixhawk: An open-source and widely used autopilot (flight controller) hardware platform. It provides standardized hardware for running flight control software;

[0086] MAVLink protocol: A lightweight messaging protocol designed for small unmanned systems (such as drones) that defines the message format and rules for communication between drones and ground stations.

[0087] Please see Figure 1 and Figure 2 This invention provides an embodiment of an unmanned aerial vehicle (UAV) system for detecting water levels in irrigation areas, which integrates image recognition. The system includes a marker parameter library module and a data processing module. The marker parameter library module, acting as the system's core, pre-stores the latitude, longitude, elevation, physical dimensions, and feature templates of each marker. The data processing module optimizes the initial water level values ​​output by the geometric calculation unit: its preprocessing unit uses Kalman filtering to remove jump values ​​caused by instantaneous waves or image noise; and the error analysis unit compares the filtered data with the baseline water level values ​​recorded in the parameter library and measured on-site by a precision level instrument. The comparison and calculation of long-term deviations are as follows: The judgment condition of "the average deviation of multiple consecutive measurements exceeds the set threshold" in the error analysis unit is as follows: The system continuously monitors a specific marker point. When the absolute value of the difference between the arithmetic mean of the measured values ​​of the point in the most recent N inspections and the pre-stored reference liquid level value is continuously greater than M cm and its standard deviation is less than K cm, it is determined that there is a systematic deviation. If this condition is met, the parameter correction unit determines that the camera calibration parameters (such as focal length) may drift due to temperature and vibration, and automatically generates a correction command, which is fed back to the geometric calculation unit to update its internal model parameters.

[0088] Furthermore, the high temperature in summer may cause a slight thermal expansion of the focal length of the camera lens, resulting in a fixed deviation in the measurement value. Without a self-calibration function, this deviation will persist and be difficult to detect. In this embodiment, the system measures an absolutely true liquid level value at the marker using a higher-precision measuring tool (such as a total station or a laser rangefinder) by the staff at specific calibration moments (such as at the initial installation of the system or during regular maintenance), and enters it into the parameter library as the "reference value". After that, the result obtained by the UAV each time it measures will be compared with this "reference value". The error analysis unit will continuously monitor the deviation sequence within a period of time (such as one month). Once it finds that the mean deviation and fluctuation are abnormal, that is, the mean difference > M and the standard deviation < K, the calibration process will be triggered. The parameter correction unit will then reverse-fit the camera parameter correction amount most likely to cause this deviation (such as correcting the focal length from 35.0 mm to 35.1 mm) by solving an optimization problem (such as the least squares method) and complete the online update. This enables the system to always maintain the measurement accuracy at the initial calibration, solving the industry pain point of the traditional vision measurement system that "the measurement becomes less and less accurate";

[0089] Kalman filter: An optimal estimation algorithm that predicts the next state by considering the "dynamic model" and "measurement value" of the system, and performs a weighted average of the predicted value and the new measurement value to obtain the best estimate closer to the true value. It is very good at extracting the true signal from data containing noise;

[0090] Least squares method: A mathematical optimization technique that finds a set of optimal function parameters by minimizing the "sum of squared errors" so that the calculation result of the model can best fit the actual observed data;

[0091] Please refer to Figure 1 and Figure 2 , an embodiment provided by the present invention: An irrigation area liquid level UAV detection system integrating image recognition includes a marker designed specifically for machine vision. The marker uses an aluminum alloy substrate sprayed with a Teflon coating and has excellent weather resistance. Its pattern design is a fusion of high-contrast black and white staggered checkerboards and ArUco codes. The checkerboard provides rich corner features for precise positioning and attitude calculation, and the ArUco code provides a unique identity recognition, facilitating the system to retrieve the corresponding size and position information from the parameter library. This design ensures that even in the case of backlight, shadow, water surface reflection, or partial splashing with mud and water, the marker detection unit can still stably and reliably distinguish and identify it from the natural environment;

[0092] Furthermore, the installation of the markers is a crucial step. Expansion bolts should be used to securely install them onto a stable structure on the channel sidewall (such as the top of a concrete lining) to ensure long-term vertical stability. During installation, a level must be used for calibration to ensure that its reference plane is parallel to the channel's reference elevation plane or has a known height difference. The size design of the markers must consider the drone's shooting distance: if the cruising altitude is 20 meters, the minimum feature of the marker (such as a checkerboard pattern) should have a pixel size greater than 10x10 pixels in the image to ensure detection accuracy. Therefore, the physical size should not be too small. (This is related to the algorithm layer.) In the face of partial occlusion, the marker detection unit can use the RANSAC algorithm to fit the feature points of the unoccluded parts, thereby estimating the position and orientation of the complete marker. In addition, the template images stored in the feature template storage unit should include multiple sets of images taken under different lighting conditions (front lighting, side lighting, back lighting). Data augmentation is used to improve the generalization ability of the algorithm. This solution, through "dedicated hardware (marker) + dedicated algorithm", increases the success rate of image recognition from about 70% of the traditional method to more than 98%, providing a solid foundation for subsequent liquid level calculation.

[0093] RANSAC algorithm: A very powerful mathematical algorithm used to robustly estimate mathematical model parameters from data containing a lot of "noise" or "outliers". Even if the marker is partially obscured by soil, the RANSAC algorithm can still guess the correct position and shape of the complete marker through the correct feature points that are not obscured, and it has a very strong anti-interference ability.

[0094] Please see Figure 1 and Figure 2 This invention provides an embodiment of an irrigation area liquid level UAV detection system that integrates image recognition. The positioning unit and attitude perception unit of the UAV status monitoring module provide high-precision flight parameter data to the geometric calculation unit, while the data processing module monitors the system output performance. Its fault diagnosis unit analyzes the UAV positioning and attitude data in real time. If it detects GPS loss of lock or abnormal tilt of the UAV, it immediately issues an alarm and terminates the mission to prevent the generation of erroneous data. Its error analysis unit monitors the quality of the liquid level data. Once an abnormal deviation is detected, it triggers the parameter correction process. This enables the system not only to output data but also to evaluate data quality and maintain its own health.

[0095] Furthermore, the attitude perception unit calculates and outputs data such as pitch angle and roll angle in real time at a high-frequency sampling rate of 100Hz. This high refresh rate data stream is sent to the geometry calculation unit in real time to compensate for changes in camera perspective caused by the swaying of the UAV body, ensuring the instantaneous accuracy of the projection transformation model. On the other hand, it is also continuously monitored by the fault diagnosis unit. This unit has preset safety thresholds based on aircraft dynamics and mission requirements. For example, the roll angle safety threshold is set to 25 degrees. If the UAV encounters continuous crosswinds during the inspection process, causing its roll angle to continuously exceed this threshold, the diagnosis unit will determine, based on logical judgment or a more complex state machine model, that the current flight attitude has seriously deviated from the horizontal. At this time, the image captured by the camera has huge geometric distortion. Even if the subsequent algorithm compensates, the accuracy of its liquid level calculation cannot be guaranteed. Therefore, it will immediately send a data packet containing command code to the UAV's flight control unit through the wireless transmission unit, instructing it to immediately execute the "hover" or "automatic return" plan, thereby actively interrupting the data acquisition chain, avoiding the storage of erroneous data, and saving the cost of subsequent data cleaning.

[0096] On the data backend, the error analysis unit does not react to single measurement deviations. Instead, based on statistical principles, it performs trend analysis on historical data of the same monitoring point. For example, through periodic T-tests or control charts, it discovers that at a specific marker, the liquid level values ​​obtained from 10 consecutive measurements all have a fixed negative deviation of approximately -5 cm from the benchmark value calibrated on-site by a precision level. The algorithm determines that this is a systematic error rather than random noise. In this case, it does not directly tamper with the original liquid level data but generates a structured diagnostic report that clearly states: "Suspected camera intrinsic parameter (focal length) drift; it is recommended to lower the focal length parameter by 0.05 mm." After receiving this report, the parameter correction unit calls the built-in optimization algorithm (such as the least squares method) to verify the rationality of the suggestion. Subsequently, it silently adds this correction amount to the camera calibration parameter file called by the geometry calculation unit in the background. The entire process is transparent to the user and requires no manual intervention. When the drone inspects the point again, the system will automatically use the updated intrinsic parameter model for calculation, and the measurement value will then return to normal.

[0097] Working principle: First, the ground control center divides the monitoring points and plans the optimal flight route based on the topography and hydrological characteristics of the irrigation area. The flight route command is sent to the UAV platform through the wireless transmission unit. The flight control unit of the UAV platform receives the command, controls the UAV to fly autonomously to the target monitoring point, and triggers the image acquisition unit to take pictures of the irrigation area including special markers. At the same time, the UAV status monitoring module collects its own positioning and attitude data at high frequency. These images and flight parameter data are transmitted back to the system backend in real time.

[0098] Next, the core algorithm of the image recognition processing module is activated. The marker detection unit quickly identifies and locates the marker in the image based on the feature templates pre-stored in the marker parameter library, and outputs its pixel coordinates and type code. The liquid level line recognition unit then delineates the detection area based on the marker position. Through advanced edge enhancement and feature filtering algorithms, it accurately extracts the pixel coordinates of the liquid level boundary line. The geometric calculation unit calls the actual size parameters corresponding to the marker and integrates the high-precision flight parameter data returned by the UAV with the camera's inherent calibration parameters to build an accurate spatial mapping model. By solving the perspective projection transformation, it finally calculates the true physical elevation value of the liquid level at this moment.

[0099] Finally, the data processing module performs in-depth optimization and system maintenance on the calculated liquid level results. Its preprocessing unit first filters the raw liquid level data to smooth instantaneous fluctuations; the error analysis unit then cross-compares the results with high-precision benchmark values. If a systematic deviation is found, a parameter correction instruction is generated to silently calibrate the geometric calculation model to ensure long-term accuracy. At the same time, the fault diagnosis unit continuously monitors the UAV status. Once an anomaly is detected, an alarm is immediately triggered and the mission is interrupted to prevent data contamination. Finally, the processed and reliable liquid level data is sent to the data display unit of the ground control center for visualization and is used to generate irrigation scheduling decisions, thus forming a complete closed loop from intelligent sensing, accurate calculation to self-correction and reapplication.

[0100] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A drone-based irrigation area liquid level detection system integrating image recognition, characterized in that: It includes an image recognition and processing module, which is used to calculate the liquid level value based on the irrigation area image data collected by the UAV; The image recognition processing module includes: a marker detection unit, a liquid level recognition unit, and a geometric calculation unit; The marker detection unit is equipped with a pre-stored marker feature library. Based on the standard pattern data in the feature library, it performs target region matching on the input irrigation area image data and outputs the marker pixel coordinates and type code. The liquid level line recognition unit delineates the detection area based on the pixel coordinates of the marker, extracts the continuous horizontal contour through the edge enhancement algorithm, filters the liquid surface boundary by combining the regional grayscale gradient distribution, and outputs the pixel coordinates of the liquid level line. The geometric calculation unit calls the corresponding actual size parameters according to the marker type code, and combines the UAV flight parameter data and camera calibration parameters to construct a spatial mapping model between the marker pixel coordinates and the actual height, and calculates the liquid level data through spatial projection transformation.

2. The irrigation area liquid level UAV detection system fused with image recognition according to claim 1, characterized in that: The irrigation area image data required by the marker detection unit is provided by the image acquisition unit of the UAV platform; The drone platform also includes a flight control unit and a wireless transmission unit. The flight control unit is connected to the image acquisition unit and the wireless transmission unit via signals. The flight control unit controls the flight status of the UAV according to the route instructions received by the wireless transmission unit, so that the UAV passes through each irrigation area monitoring point along the preset path; The image acquisition unit is equipped with a high-resolution camera to collect image data of the irrigation area, including markers, when the drone flies over the monitoring point of the irrigation area; The wireless transmission unit transmits the filtered liquid level data to the ground control center and receives control commands from the ground control center.

3. The irrigation area liquid level UAV detection system fused with image recognition according to claim 2, characterized in that: The route planning unit of the ground control center provides route instructions to the flight control unit of the UAV platform; The ground control center also includes a data display unit and a monitoring point division unit. The monitoring point division unit is connected to the data display unit and the flight path planning unit via signals. The monitoring points are divided into units based on the topographical features, hydrological features, and key areas of irrigation management in the irrigation area. The flight path planning unit generates the UAV flight path based on the distribution of monitoring points in the irrigation area and sends the flight path instructions to the wireless transmission unit of the UAV platform. The data display unit receives liquid level data transmitted by the wireless transmission unit of the drone platform and displays it in the form of charts, so that operators can intuitively obtain the liquid level distribution in the irrigation area.

4. The irrigation area liquid level UAV detection system fused with image recognition according to claim 1, characterized in that: The pattern feature data required by the feature feature library configured in the marker detection unit is provided by the feature template storage unit of the marker parameter library module; The marker parameter library module also includes a geographic coordinate storage unit and a physical size storage unit; The geographic coordinate storage unit records the latitude, longitude, and elevation data of each marker installation point, providing navigation and positioning references to the flight control unit of the UAV platform; The physical size storage unit associates the identifier type encoding to store the actual width and height values, and provides the size conversion parameters of the spatial mapping model to the geometric calculation unit of the image recognition processing module; The feature template storage unit stores the standard binary contour matrix of the marker and provides target matching pattern feature data to the marker feature library configured in the marker detection unit of the image recognition processing module.

5. The irrigation area liquid level UAV detection system fused with image recognition according to claim 1, characterized in that: The flight parameter data required by the geometric calculation unit is provided by the attitude perception unit and the positioning unit of the UAV state monitoring module; The UAV status monitoring module also includes a fault diagnosis unit, which is connected to the attitude sensing unit and the positioning unit via signals. The positioning unit acquires the real-time position information of the drone to check whether it deviates from the preset route. At the same time, it is equipped with a laser rangefinder to provide the drone's altitude data to the geometric calculation unit of the image recognition and processing module. The attitude perception unit collects the flight attitude data of the UAV and provides it to the geometric calculation unit of the image recognition and processing module. The flight attitude data includes pitch angle, roll angle and yaw angle data. The fault diagnosis unit analyzes real-time location information and flight attitude data, and generates an alarm signal when an abnormal state is detected and sends it to the ground control center.

6. The irrigation area liquid level UAV detection system fused with image recognition according to claim 1, characterized in that: The liquid level data calculated by the geometric calculation unit is filtered by the preprocessing unit of the data processing module; The data processing module also includes an error analysis unit and a parameter correction unit. The data preprocessing unit is connected to the error analysis unit and the parameter correction unit via signals. The preprocessing unit filters the liquid level data calculated by the geometric calculation unit to remove outliers caused by image noise, and then transmits the processed liquid level data to the wireless transmission unit. The error analysis unit receives the filtered liquid level data from the data preprocessing unit, compares it with the reference liquid level value obtained by on-site calibration at the marker installation point using a high-precision measuring instrument, calculates the deviation between the liquid level data measured by the UAV and the reference liquid level value, and generates a deviation analysis report. Based on the deviation analysis report output by the error analysis unit, when the average deviation of multiple consecutive measurements exceeds a set threshold, the parameter correction unit generates correction instructions for the camera calibration parameters and spatial mapping model, and feeds the correction instructions back to the geometric calculation unit of the image recognition processing module.

7. The irrigation area liquid level UAV detection system fused with image recognition according to claim 1, characterized in that: The markers have an appearance that contrasts sharply with the surrounding environment and contain a geometric structure of known physical dimensions. The markers are made of weather-resistant materials and are fixedly installed on the side walls of irrigation canals and on the stable structures of key monitoring points. High-contrast appearance features regular patterns formed by alternating dark and light areas; A geometric structure with known physical dimensions includes precisely defined scale lines and standard geometry with standard side lengths; The regular patterns and standard geometric structures of the markers together constitute the feature templates for machine vision recognition, enabling the marker detection unit to achieve robust recognition and accurate positioning in complex backgrounds.

8. The irrigation area liquid level UAV detection system fused with image recognition according to claim 1, characterized in that: The process by which the geometric calculation unit performs spatial projection transformation to calculate liquid level data includes: Based on the received identifier type code, the corresponding actual size parameters of the identifier are retrieved from the identifier parameter library module; Acquire real-time flight parameter data of the UAV provided by the UAV status monitoring module, including three-dimensional spatial coordinates and flight attitude angles; Read the preset camera internal calibration parameters; Based on the pixel coordinates of the markers in the image and their corresponding actual size, combined with flight parameter data and camera calibration parameters, a spatial mapping relationship from the two-dimensional image pixel coordinate system to the three-dimensional world coordinate system is constructed. Substituting the liquid level pixel coordinates output by the liquid level recognition unit into the spatial mapping relationship, the three-dimensional spatial coordinates of the liquid point are calculated through coordinate transformation and geometric operations, and its elevation component is the final calculated liquid level data.

9. The irrigation area liquid level UAV detection system fused with image recognition according to claim 1, characterized in that: The flight parameter data on which the geometric calculation unit performs spatial projection transformation specifically includes ground altitude data provided by the positioning unit of the UAV status monitoring module and pitch angle, roll angle and yaw angle data provided by the attitude perception unit; the camera calibration parameters specifically include the camera focal length, principal point coordinates and radial distortion coefficients obtained in advance by Zhang Zhengyou calibration method.

10. The irrigation area liquid level UAV detection system fused with image recognition according to claim 6, characterized in that: The specific condition for determining that the average deviation of multiple consecutive measurements in the error analysis unit exceeds the set threshold is as follows: the system continuously monitors a specific marker point. When the absolute value of the difference between the arithmetic mean of the measured values ​​of this point in the most recent N inspections and the pre-stored reference liquid level value is continuously greater than M cm and its standard deviation is less than K cm, it is determined that there is a systematic deviation. Here, N is an integer greater than 5, and M and K are positive real numbers set according to the measurement accuracy requirements of the irrigation area.

Images