Plant protection unmanned aerial vehicle operation method and system based on Beidou high-precision positioning and hyperspectral nitrogen nutrition estimation

Abstract

The invention discloses a plant protection unmanned aerial vehicle operation method and system based on Beidou high-precision positioning and hyperspectral nitrogen nutrition estimation, and the method comprises the steps: determining a plant protection unmanned aerial vehicle operation region, and calculating the flight height of an unmanned aerial vehicle through hyperspectral camera parameters and navigation positioning data; acquiring a hyperspectral image of crops in the operation area of the plant protection unmanned aerial vehicle in real time through a hyperspectral camera, and determining a nitrogen nutrition deficiency area by using a spectral feature selection algorithm based on spectral data in the hyperspectral image; converting the pixel coordinates of the nitrogen nutrition deficiency region into real geographic coordinates; based on the real geographic coordinates of the nitrogen nutrition deficient area, the performance parameters and task requirements of the unmanned aerial vehicle are combined, a dynamic window method is adopted for sowing path planning, and finally accurate sowing irrigation is achieved. According to the method, a dynamic threshold value and maximum entropy threshold value method is introduced, Canny edge detection and multi-source correction are combined, region boundary jitter and area calculation errors are reduced, and a space reference is provided for precise sowing.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of precision agriculture / smart agriculture technology, and particularly relates to a plant protection unmanned aerial vehicle operation method and system based on Beidou high-precision positioning and high-spectrum nitrogen nutrition estimation. BACKGROUND

[0002] With the development of smart agriculture concept, the application of plant protection unmanned aerial vehicles in spreading fertilizers, plant protection pesticides, microelement supply and irrigation assistance is continuously expanding. High-spectrum imaging technology can capture the spectral characteristics of vegetation in multiple sub-bands, thereby supporting the inversion and diagnosis of leaf nitrogen, chlorophyll, water content and other nutritional status. Beidou high-precision positioning (especially RTK and differential technology) can achieve centimeter-level positioning, which is conducive to realizing precise delivery and repeated route reproduction. At the same time, path planning and variable decision need to take into account: operation coverage, obstacle avoidance safety, energy and load constraints, wind field influence, uniformity and efficiency.

[0003] The limitations of the current methods include: (1) Traditional NDVI / vegetation index single threshold method: the threshold is fixed or set empirically, which is difficult to adapt to the spectral differences of different plots, light and crop growth stages, resulting in insufficient accuracy and stability of the identification of nutrient deficiency areas. (2) Simple binary segmentation + rough edge extraction: without entropy optimization and fine edge detection, the boundary jitter is large, the area estimation error is high, and the accuracy of subsequent variable application areas is affected. (3) Ordinary GPS (meter level) positioning: cannot meet the requirements of fine variable sowing for regional boundary closure accuracy and repeated positioning consistency, causing sowing / spreading drift and overlap. (4) Existing unmanned aerial vehicle tasks rely on manual or rule-based grid flight lines: without optimizing the trajectory according to the irregular shape of the real-time nutrient deficiency patch, over-spraying, under-spraying and redundant flight paths are prone to occur. (5) Traditional path planning ignores dynamic speed / attitude constraints: only generates geometric shortest or coverage paths, ignoring dynamic constraints (acceleration, angular velocity) and temporary obstacles (temporary facilities, moving personnel / machines), reducing safety and execution consistency. (6) Lack of multi-source data fusion: high-spectral images and positioning, attitude, DEM (digital elevation), GCP (ground control point) and other data are not synchronized or misregistered, causing cumulative errors in geographic coordinate inversion. (7) Variable application strategy is extensive: the "deficient" and "normal" areas are simply treated as binary, without considering different deficiency levels or uncertainties, and the fertilizer utilization rate and ecological benefits are not fully improved. (8) Lack of cloud collaboration and closed-loop feedback: flight trajectory, application amount, nitrogen estimation results and other data are not analyzed in real time, and the subsequent decision optimization, cross-season comparison and model self-learning ability are insufficient. (9) Existing differential / RTK applications and high-spectral algorithms are disconnected: the positioning accuracy advantage is not fully converted into the boundary fitting accuracy of the nutrient patch, and a "detection-positioning-variable control" tightly coupled link cannot be formed. (10) Environmental and safety factors are not adapted: wind speed, obstacle distribution, no-fly zone, remaining battery and load-endurance coupling are not optimized comprehensively, affecting the continuity and safety redundancy of the operation. This brings the following problems: (1) Nutrient deficiency area identification is not stable → variable application precision is insufficient; (2) Geographical positioning and spectral data coupling is not tight → boundary offset and area calculation error; (3) Path planning and dynamics / obstacle avoidance are not fully integrated → operation efficiency and safety are reduced; (4) Variable delivery strategy lacks fine grading and closed-loop data feedback → fertilizer utilization rate and resource scheduling optimization are insufficient; (5) The system's functional modules are fragmented → it is difficult to form a data-driven continuous iteration improvement. SUMMARY

[0004] The purpose of the present application is to solve the problems in the prior art, and provide a plant protection unmanned aerial vehicle operation method based on Beidou high-precision positioning and hyperspectral nitrogen nutrition estimation, which introduces a dynamic threshold and a maximum entropy threshold method, combines Canny fine edge detection with RTK, GCP and DEM multi-source correction, uses an extended dynamic window method to search for an optimal action in a feasible set of speed-angle speed in real time, differentiates and supplies to a nitrogen deficiency area based on nitrogen distribution, adapts to different illumination, varieties and background differences in different growth periods, improves the accuracy and robustness of nitrogen deficiency area determination, reduces area boundary jitter and area calculation error, provides a reliable spatial reference for precise variable spreading, realizes high-precision mapping of pixel-level spectral information to real geographic coordinates, and reduces spatial deviation accumulation.

[0005] According to an aspect of the present application, a plant protection unmanned aerial vehicle operation method based on Beidou high-precision positioning and hyperspectral nitrogen nutrition estimation is provided, comprising:

[0006] S1, determining a plant protection unmanned aerial vehicle operation area, and calculating the flight height of the unmanned aerial vehicle through hyperspectral camera parameters and navigation positioning data;

[0007] S2, collecting a hyperspectral image of crops in the plant protection unmanned aerial vehicle operation area in real time through a hyperspectral camera, determining a nitrogen deficiency area by using a dynamic threshold algorithm according to the pixel spectral features in the hyperspectral image, and outputting the boundary shape and area of the nitrogen deficiency area;

[0008] S3, obtaining pixel coordinates of the nitrogen deficiency area based on the boundary shape of the nitrogen deficiency area, and converting the pixel coordinates into real geographic coordinates;

[0009] S4, planning a spreading path by using an extended dynamic window method based on the real geographic coordinates and area of the nitrogen deficiency area, and finally realizing precise spreading irrigation.

[0010] Further, the determination of the nitrogen deficiency area by using the dynamic threshold algorithm according to the pixel spectral features in the hyperspectral image comprises:

[0011] calculating a gray histogram according to the pixel spectral features in the hyperspectral image, normalizing the gray histogram into a probability distribution, and segmenting the hyperspectral image by using a maximum entropy threshold method based on the probability distribution to obtain a binary image; the binary image is divided into a nitrogen deficiency area and a normal area;

[0012] performing edge detection on the binary image by using a Canny algorithm, extracting edge pixel coordinates of the nitrogen deficiency area, and outputting the boundary shape and area of the nitrogen deficiency area.

[0013] Further, the edge detection on the binary image by using the Canny algorithm comprises:

[0014] Gaussian smoothing is performed on the binary image;

[0015] The gradient components of the Gaussian smoothed image in horizontal and vertical directions are calculated by using a Sobel operator to obtain gradient amplitudes and gradient directions;

[0016] For each pixel in the image, the gradient amplitudes of the pixel and its adjacent pixels are compared along the gradient direction, and the local maximum points are retained;

[0017] High and low threshold values are set and , When the gradient amplitude is greater than or equal to the high threshold value, it is marked as a strong edge; when the gradient amplitude is less than the high threshold value and greater than or equal to the low threshold value, it is marked as a weak edge connected to the strong edge; and the edge detection result is obtained according to the strong edge and the weak edge connected to the strong edge.

[0018] Further, the S3 comprises:

[0019] The pixel coordinates of the nitrogen deficiency region are obtained, and the ground coordinates of the nitrogen deficiency region are obtained through the mapping relationship between the pixel coordinates and the real-time spatial coordinates of the unmanned aerial vehicle;

[0020] The ground coordinates of the nitrogen deficiency region are converted into real geographical coordinates, and the formula is:

[0021]

[0022]

[0023]

[0024] wherein, is the longitude, is the latitude, is the height of the ground point, is the radius of the earth, is the point in the ground coordinate system.

[0025] Further, the S4 comprises:

[0026] According to the real geographical coordinates and the area of the nitrogen deficiency region, the region is divided into a plurality of sub-regions, and each sub-region is subjected to grid processing to obtain the boundary coordinates and the grid unit of each sub-region;

[0027] Based on the boundary coordinates and the grid unit of each sub-region, an extended dynamic window method is used for sowing path planning, and the evaluation function is:

[0028] ,

[0029] wherein, represents the path cost to reach the target area; represents the cost of avoiding obstacles; represents the speed optimization cost; represents the weight coefficient, represents the current speed and angular velocity the normalized cost of the newly covered nitrogen deficiency grid area in the prediction time window relative to the maximum newly covered area in the candidate set; represents the normalized measure of the deviation between the predicted variable spreading actual spreading rate and the target spreading rate; represents the insufficient risk cost of the remaining load relative to the expected demand of the nitrogen deficiency area that has not been covered.

[0030] Further, , The expressions of and are respectively:

[0031] ,

[0032] ,

[0033] ,

[0034] wherein, is the newly covered nitrogen deficiency area in the prediction window by the current candidate speed combination, is the maximum newly covered area in the candidate set in the same window; N is the number of prediction trajectory coverage grids; is the i-th grid prediction instantaneous or segment average spreading rate, is the target spreading rate of the grid, is the reference spreading rate; is the remaining unfinished nitrogen deficiency area, is the average target spreading rate of the remaining area, is the real-time remaining load.

[0035] Further, after obtaining the real geographical coordinates of the nitrogen deficiency area, it further includes:

[0036] Geographical control point correction: compare the real geographical coordinates with known geographical control points and correct using the least squares method;

[0037] Terrain elevation data correction: adjust the error caused by terrain undulation using high-resolution terrain elevation data.

[0038] Further, the flight height of the unmanned aerial vehicle is calculated, and the expression is:

[0039]

[0040] wherein, is the flight height; is the sensor width; is the lens focal length; is the image width.

[0041] According to an aspect of the present specification, a plant protection unmanned aerial vehicle operation system based on Beidou high-precision positioning and high-spectral nitrogen nutrition estimation is provided, comprising:

[0042] A data preparation module is configured to determine a plant protection unmanned aerial vehicle operation area and calculate the flight height of the unmanned aerial vehicle through high-spectral camera parameters and navigation positioning data;

[0043] A nitrogen deficiency area determination module is configured to collect high-spectral images of crops in the plant protection unmanned aerial vehicle operation area in real time through a high-spectral camera, determine a nitrogen deficiency area according to the pixel spectral characteristics in the high-spectral images, and output the boundary shape and area of the nitrogen deficiency area by using a dynamic threshold algorithm;

[0044] A coordinate conversion module is configured to obtain pixel coordinates of the nitrogen deficiency area based on the boundary shape of the nitrogen deficiency area and convert the pixel coordinates into real geographic coordinates;

[0045] A path planning module is configured to plan a scattering path by using an extended dynamic window method based on the real geographic coordinates and area of the nitrogen deficiency area, and finally realize precise scattering irrigation.

[0046] Further, the system further comprises:

[0047] A wireless communication module is configured to upload the flight trajectory, nitrogen estimation result, scattering amount, and obstacle information to a cloud data platform in real time;

[0048] A cloud data platform is configured to store and analyze data, and support decision-making and optimization for unmanned aerial vehicle operation;

[0049] A communication module is configured to monitor the state of the unmanned aerial vehicle throughout the entire process and issue an emergency return instruction in a timely manner when an abnormality occurs;

[0050] A flight control module is configured to automatically plan the shortest return path.

[0051] Compared with the prior art, the present application has the following advantages:

[0052] 1. The application introduces dynamic threshold and maximum entropy threshold method, self-adapts to different light, variety and background difference of growth period, improves the accuracy and robustness of nitrogen nutrient deficiency area determination; combined with Canny fine edge detection and RTK, GCP, DEM multi-source correction, reduces the area boundary jitter and area calculation error, provides reliable spatial reference for accurate variable spreading; through the fusion of attitude matrix, sensor model and high-precision positioning data, realizes high-precision mapping of pixel-level spectral information to real geographic coordinates, reduces the accumulation of spatial deviation.

[0053] 2. The application adopts extended dynamic window method to search the optimal action in the speed-angle speed feasible set in real time, considers obstacle avoidance, safety and time consumption, reduces track redundancy and missed spreading; based on nitrogen distribution, the application differentiates the supply of deficiency areas, improves fertilizer utilization rate, and reduces loss and environmental risk caused by excessive application; through regional blocking and grid organization, combined with optimized path planning under the constraint of dynamics, the application shortens the air distance and repeated flight time, and improves the effective working area per unit time. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0055] Figure 1 The method flow chart of the plant protection unmanned aerial vehicle operation method based on Beidou high-precision positioning and high-spectrum nitrogen nutrition estimation is provided. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0057] As Figure 1As shown, the embodiment of the present application provides a plant protection unmanned aerial vehicle operation method based on Beidou high-precision positioning and high-spectrum nitrogen nutrition estimation, comprising: S1, determining a plant protection unmanned aerial vehicle operation area, and calculating the unmanned aerial vehicle flight height through high-spectrum camera parameters and navigation positioning data; S2, collecting high-spectrum images of crops in the plant protection unmanned aerial vehicle operation area in real time through a high-spectrum camera, determining a nitrogen nutrition deficiency area according to pixel spectral characteristics in the high-spectrum image, and outputting the boundary shape and area of the nitrogen nutrition deficiency area; S3, obtaining pixel coordinates of the nitrogen nutrition deficiency area based on the boundary shape of the nitrogen nutrition deficiency area, and converting the pixel coordinates into real geographic coordinates;

[0058] S4, based on the real geographic coordinates and area of the nitrogen nutrition deficiency area, using an extended dynamic window method for scattering path planning, and finally realizing precise scattering irrigation.

[0059] Specifically, the plant protection operation area is determined, the plant protection operation area is identified and delimited using ArcGIS software, the real geographic coordinates, terrain elevation data (DEM) and static obstacle distribution information of the area are obtained, and the accuracy and safety of the operation area are ensured. The map preprocessing module rasterizes the area, marks the obstacles, and integrates the terrain elevation data, thereby providing basic data for subsequent path planning and flight control.

[0060] Specifically, the high-spectrum camera parameters and the unmanned aerial vehicle flight height are determined, the high-spectrum camera module (400-1000nm, resolution 5nm) is combined with the operation accuracy requirement (GSD 5-6cm), and the sensor parameters (width 25mm, resolution 4000pixels, focal length 12.5-15mm) in the structure module are used to calculate the optimal flight height (100-120meters). A multi-source data synchronization module is used to ensure that the high-spectrum camera and the navigation positioning system data are synchronized. The navigation positioning system data is navigation data and positioning information: the navigation data comes from the real geographic coordinates, terrain elevation data (DEM) and static obstacle distribution information during planning by the ArcGIS software, and the positioning information comes from the GPS module in the unmanned aerial vehicle.

[0061] Specifically, a higher flight height will result in a larger ground coverage, but the ground sampling distance (GSD) will become larger and the image resolution will become lower. A lower flight height will improve the image resolution (GSD becomes smaller), but the ground coverage will decrease. Therefore, the optimal flight height of the unmanned aerial vehicle needs to be calculated according to the operation accuracy requirement (GSD).

[0062] Specifically, based on actual calibration test results, this embodiment of the invention selects a hyperspectral camera with a sensor width of 25mm, a resolution of 4000 pixels, and a lens focal length of 12.5-15mm. The GSD needs to be controlled between 5-6cm. The flight altitude is calculated according to the following formula:

[0063] (1)

[0064] in, It is the flight altitude (unit: meters). This is the sensor width (unit: millimeters). It is the lens focal length (unit: millimeters); It is the image width (unit: pixels). Based on calculations, the drone's operating altitude should be between 100-120 meters.

[0065] Specifically, the UAV takeoff operation includes activating the BeiDou high-precision positioning module (RTK). The UAV achieves centimeter-level real-time differential positioning through the collaboration of a base station (with a known location) and a rover station (the UAV itself). The inertial measurement unit (IMU) and millimeter-wave radar work together to improve flight attitude and obstacle detection accuracy. The GNSS module provides global navigation satellite system redundancy, enhancing positioning reliability. The UAV uses the BeiDou high-precision positioning module for real-time differential positioning to obtain its centimeter-level latitude and longitude coordinates. The differential positioning system uses carrier phase measurements from the UAV and the ground base station, combined with pseudorange differences and integer ambiguity, to calculate the UAV's high-precision position. Using the BeiDou-3 RTK differential positioning system, positioning accuracy can reach the centimeter level. The differential positioning formula is as follows:

[0066] (2)

[0067] in, It is the carrier phase difference value between the UAV and the base station; This is the current carrier phase measurement value of the drone; It is the carrier phase measurement value of the reference station; It is the pseudorange difference between the UAV and the base station to the satellite; It is the carrier wavelength (BeiDou-3 system), used to convert pseudorange difference values ​​into phase difference values; It is integer ambiguity, which needs to be solved through differential positioning; This refers to the measurement error term, which includes errors introduced by environmental noise, signal interference, etc. It is estimated through experimental calibration and compensated and corrected in subsequent processing.

[0068] Specifically, hyperspectral data acquisition includes real-time acquisition of crop spectral images by a hyperspectral camera, which is then synchronously recorded with the high-precision positioning coordinates of the UAV. Land-based sensors and environmental sensors (temperature, humidity, air pressure, wind speed, soil moisture, light intensity, etc.) synchronously acquire environmental data, providing support for subsequent parameter optimization and model correction. During UAV flight, hyperspectral images of crops are acquired in real-time via the hyperspectral camera module. Images are acquired and recorded synchronously with the UAV's high-precision positioning coordinates, ensuring accurate spatial calibration of the images. The acquired hyperspectral images undergo preprocessing: 1. Spectral correction: eliminating the influence of ambient light variations and sensor noise. 2. Band selection: selecting key band ranges related to nitrogen content (550-700nm red and near-infrared bands). 3. Image filtering: using Gaussian filtering to remove random noise from the images.

[0069] Specifically, nitrogen nutrient estimation includes: an image preprocessing module performing radiometric correction, atmospheric correction, image registration, and multimodal data integration on the acquired images to eliminate environmental and sensor interference and improve data consistency. A multi-source data synchronization module ensures accurate temporal and spatial registration of the acquired data. Based on the acquired spectral data of crops, each pixel contains multiple bands (features), which are closely related to the crop's nitrogen content. The nitrogen content is estimated using a spectral feature selection algorithm, specifically: Step 1, establishing a crop input feature matrix X, where each row of X represents a pixel sampling point, and each column represents a band. For example, , For the sample size, This represents the number of spectral bands. This is for the real number domain. The sample size is set to 2000 by default, and the spectral bands are set to 6. Step 2: Generate Q, K, and V, and project the input features into three new features using three sets of weight matrices: These weight matrices are three independent trainable matrices, each learning a different linear projection. W is a trainable linear transformation matrix reflecting information on "which spectral band combinations are most helpful in predicting nitrogen." Step 3: Use an attention mechanism to focus on the bands / features most relevant to nitrogen content, as shown in the following formula:

[0070] (3)

[0071] in, It is a scaling factor for the feature dimension, used for normalization calculations. Based on actual testing, this embodiment of the invention sets it to 64.

[0072] Specifically, the determination of nutrient deficiency areas includes: a nitrogen inversion model module that estimates crop nitrogen content through spectral feature extraction, feature selection, model training and optimization, and model validation and evaluation; a nitrogen nutrient analysis module that generates a nitrogen content distribution map to identify and accurately locate nutrient deficiency areas; a nitrogen inversion model module that converts hyperspectral multi-band data into a continuous estimation raster reflecting crop nitrogen levels, providing high-discrimination input for subsequent classification; and a nitrogen nutrient analysis module that performs statistical distribution modeling, grading, and anomaly removal on the continuous nitrogen raster, outputting threshold selection parameters and nutrient level labels, providing a structured data foundation for dynamic threshold segmentation and path planning. The dynamic decision system performs dynamic threshold segmentation and edge detection on the hyperspectral image, and outputs the spatial location and area of ​​nutrient deficiency areas through differential region location and mutually exclusive region area calculation.

[0073] Specifically, based on spectral feature analysis of hyperspectral images and combined with a dynamic thresholding algorithm, the nitrogen content of crops is assessed in real time, and nutrient-deficient areas are marked. The specific steps are as follows:

[0074] 1. Dynamic thresholding segmentation analyzes images across different wavelengths to select the wavelength that best reflects vegetation health. Commonly used wavelengths include near-infrared (NIR) and red light, as they are sensitive to differences in vegetation reflectance. Gray-level histograms of the Normalized Difference Vegetation Index (NDVI) and crop vegetation index images are calculated. The bimodal distribution of the histogram corresponds to nutrient-deficient and normal regions. The maximum entropy thresholding method is used to determine the optimal threshold, segmenting the image into two parts: nutrient-deficient and normal regions. Specific steps include calculating the entropy of the foreground and background, and selecting the threshold that maximizes the total entropy. Based on the calculated threshold, regions in the image below the threshold are identified as nutrient-deficient regions, and regions above the threshold are identified as normal regions.

[0075] The specific calculation process is as follows: First, calculate the gray-level histogram of the image. Assuming the image gray-level range is 0 to L-1, the histogram can be represented as a vector of length L, where each element represents the number of pixels at the corresponding gray level. Then, normalize the histogram to a probability distribution: ,in, It's the probability of grayscale. It refers to the number of pixels at the grayscale level. This represents the total number of pixels in the image. For each possible threshold t, the image is divided into foreground (gray level 0 to t) and background (gray level t+1 to L–1), and the entropy of the foreground is calculated:

[0076] (4)

[0077] in, The total probability of the foreground is used to calculate the entropy of the background:

[0078] (5)

[0079] in, It represents the total probability of the background.

[0080] The choice makes the total entropy The largest threshold t is taken as the optimal threshold.

[0081] 2. Use the Canny algorithm to perform edge detection on the binarized image and extract the edge pixel coordinates of nutrient-deficient regions. Output the boundary shape and area of ​​the nutrient-deficient regions. The specific formula is as follows:

[0082] 2.1 Gaussian filtering (denoising) for binarized images Gaussian smoothing is applied to suppress noise interference. The Gaussian kernel function is:

[0083] (6)

[0084] in, The standard deviation of the Gaussian kernel controls the smoothness (take...). ).

[0085] Smoothed image Convolution of the original image with a Gaussian kernel:

[0086] (7)

[0087] 2.2. The Sobel operator is used to calculate the gradient components of the image in the x-direction (horizontal) and y-direction (vertical). and :

[0088] (8)

[0089] Among them, gradient magnitude and direction The calculation formula is:

[0090] (9)

[0091] gradient direction Quantization is performed in four principal directions (0°, 45°, 90°, 135°) for nonmaximum suppression.

[0092] 2.3 For each pixel, compare its gradient magnitude with that of its neighboring pixels along the gradient direction, retain only local maxima, and suppress non-edge pixels:

[0093] (10)

[0094] The horizontal gradient obtained by the Sobel operator with vertical gradient Composite, representing pixels The strength of grayscale change indicates the point's location; a larger value suggests that the point is more likely to be located at an edge. It is a refined edge response map after "non-maximum suppression". Its main function is to refine the edges and shrink the originally wide and blurry gradient response into a single pixel wide candidate edge.

[0095] 2.4 Dual-threshold hysteresis tracking (edge ​​connection), setting high and low thresholds and , ,Pick Strong edge: amplitude Directly marked as an edge. Weak edge: amplitude Edges connected to strong edges are preserved; otherwise, they are suppressed. Final edge image. Composed of pixels that meet the following conditions:

[0096] (11)

[0097] 2.5 Edge pixel coordinates extracted through the above steps After coordinate transformation, the boundary shape and area of ​​the nutrient-deficient region are output.

[0098] Specifically, the coordinate calculation of nutrient-deficient areas includes converting the pixel coordinates of these areas into true geographic coordinates (latitude, longitude, and elevation) using a mapping model between pixel coordinates and UAV real-time spatial coordinates. Utilizing GNSS modules, IMUs, base station / rover RTK data, and DEM data, combined with Geographic Control Points (GCPs) and attitude matrices, accurate coordinate calculation and error correction are performed. The coordinate calculation of nutrient-deficient areas involves converting the pixel coordinates of hyperspectral images into true geographic coordinates, ensuring the accurate location of nutrient-deficient areas in geospatial space. The specific steps are as follows:

[0099] 1. Pixel coordinate acquisition involves segmenting the preprocessed hyperspectral image into different regions using image processing algorithms (such as machine learning algorithms). Each region represents a different health status. Within the segmented image, regions of interest are identified, determined by analyzing spectral features. For example, identifying anomalies in reflectance at specific wavelengths can help pinpoint potential nutritional deficiencies. Pixel coordinates are then extracted from the identified regions. Pixel coordinates are expressed in row and column numbers. The form represents the position of the pixel in the image matrix.

[0100] 2. Spatial coordinate transformation model, including establishing the mapping relationship between pixel coordinates and the real-time spatial coordinates (latitude, longitude, and elevation) of the UAV. The transformation formula is as follows:

[0101] (12)

[0102] in, These are pixel coordinates. These are the pixel coordinates of the image center point. This refers to the camera's focal length. This refers to the flight altitude of the drone. These are the transformed ground coordinates.

[0103] 3. Geographic coordinate calculation, the specific calculation process is as follows:

[0104] 3.1 Data Preparation: Before calculating geographic coordinates, the following data and information need to be prepared: Hyperspectral image data: including pixel coordinates (row and column numbers) and corresponding spectral information. Sensor parameters: including imaging geometry (e.g., focal length, field of view), sensor position (e.g., satellite orbit data), attitude angles (pitch angle, yaw angle, roll angle). Georeferenced data: Digital elevation model (DEM) provides terrain height information. Geographic control points (GCPs): points with known geographic coordinates, used for calibration and registration. Projection information: the projection system of the target geographic coordinates, WGS84.

[0105] 3.2 Conversion from pixel coordinates to image coordinates. Pixel coordinates in hyperspectral images are converted to row and column coordinates. This means that it needs to be converted to physical coordinates in the image coordinate system. The formula is:

[0106] (13)

[0107] in, The reference point for the image (top left corner of the image); The physical size of a pixel (unit: degrees).

[0108] 3.3. Transformation from image coordinates to sensor coordinates. Transform to the sensor coordinate system (i.e., the camera coordinate system). The sensor is a linear projection model, using the following formula:

[0109] (14)

[0110] in, This is the focal length of the sensor. Image height (provided by DEM). These are the three-dimensional coordinates in the sensor coordinate system.

[0111] 3.4 Transformation from Sensor Coordinates to Ground Coordinates: This step transforms points in the sensor coordinate system to the ground coordinate system (i.e., the Earth coordinate system). This requires considering the sensor's attitude angles (pitch, yaw, roll) and its absolute position (e.g., GPS data). The sensor's attitude matrix is ​​used. (Calculated from attitude angles) Perform coordinate rotation:

[0112] (15)

[0113] in, Represents a point in the ground coordinate system. Indicates the absolute position of the sensor; This represents the attitude rotation matrix.

[0114] 3.5. Conversion from ground coordinates to geographic coordinates: Converting ground coordinates (Cartesian coordinates) to geographic coordinates (latitude, longitude, and altitude) uses the following formula:

[0115] (16)

[0116] (17)

[0117] (18)

[0118] in, This represents the Earth's radius (assuming it is an ellipsoid; the specific value is determined by the ellipsoid model, such as WGS84). Indicates the height of the ground point (corrected from DEM data).

[0119] 3.6 Projection Conversion: If it is necessary to convert geographic coordinates (latitude and longitude) to planar projected coordinates (such as UTM coordinates), a projection conversion formula can be used. This paper adopts the general transverse Mercator projection model.

[0120] 3.7 Accuracy Correction: Due to sensor errors, attitude errors, and terrain influences, the initially calculated geographic coordinates may have deviations, requiring accuracy correction.

[0121] Geographic Control Point (GCP) Correction: The calculated coordinates are compared with known geographic control points, and correction is performed using the least squares method.

[0122] DEM Correction: Adjustments are made for errors caused by terrain undulations using high-resolution DEM data.

[0123] 3.8 Output geographic coordinates. Finally, the output geographic coordinates are represented in the following form: latitude and longitude coordinates: That is, longitude, latitude, and altitude. Projected coordinates: such as UTM coordinates. That is, the east coordinate, the north coordinate, and the altitude.

[0124] 4. Coordinate output: Output the actual geographic coordinates (latitude and longitude) of the nutrient-deficient area for subsequent seeding path planning.

[0125] Specifically, this invention proposes a seeding path planning method. This method generates an efficient and safe seeding path based on the geographical coordinates of nutrient-deficient areas, combined with UAV performance parameters and mission requirements. The specific steps are as follows:

[0126] 1. Nutrient-deficient areas are segmented and gridded. Based on the actual geographic coordinates and area of ​​the nutrient-deficient region, the area is divided into several sub-regions to facilitate drone operations on a segment-by-segment basis. The segmentation criteria include region shape, terrain complexity, and the coverage capability of a single drone flight.

[0127] 1.1. Each sub-region is gridded. The grid size is determined based on the drone's spraying width (usually 5-10 meters, adjusted according to actual conditions) and accuracy requirements to ensure uniform spraying.

[0128] 1.2 Record the boundary coordinates and grid cells of each sub-region as input data for path planning.

[0129] 1.3. Based on the shape of the sub-region and the task requirements, select the round-trip (Z-shaped) path coverage mode. Ensure 100% coverage of the broadcast area to avoid omissions, while minimizing duplicate coverage areas to save resources.

[0130] 1.4. Based on the drone's endurance and payload, calculate the area that can be covered by a single flight, and rationally plan the number of take-offs and landings and the location of resupply points.

[0131] 1.5 To ensure that the route planning meets safety requirements, it is recommended that the operating wind speed be less than level 5, avoid rainy days and high temperatures, and pay attention to no-fly zone restrictions.

[0132] 2. Extended Dynamic Window (DWA) path optimization, within the drone's velocity space. Search for the optimal speed combination By optimizing the cost function of direction, distance, and velocity, a safe and efficient path is generated.

[0133] (19)

[0134] in, Indicates based on current speed With angular velocity The normalized cost of the area of ​​uncovered nutrient-deficient grids that can be newly covered within the prediction time window relative to the maximum newly covered area in the candidate set, i.e., the candidate velocity combination for a certain dynamic window ( The normalized result of comparing the area of ​​nutrient-deficient grids that can be "added" to the coverage of currently "uncovered" grids using this velocity (and the short-term flight trajectory derived from it) within a finite forward-looking time window T with the "maximum area that can be added to the coverage" among all candidate velocity combinations in this evaluation period. A normalized measure of the deviation between the actual application rate and the target application rate of the predictor variable; This represents the cost of insufficient remaining load relative to the projected demand in nutrient-deficient areas that are not yet covered.

[0135] in, , and The calculations are as follows:

[0136] (20)

[0137] (twenty one)

[0138] (twenty two)

[0139] in, The area of ​​unseeded, nutrient-deficient regions newly covered within the prediction window for the current candidate velocity combination. The maximum newly added coverage area in the same window candidate set; N is the number of grid cells covered by the predicted trajectory; To predict the instantaneous or segmental average application rate for the i-th grid, The application rate for this grid target. The baseline application rate; The remaining unresolved nutrient deficiency area. The average target application rate for the remaining area. This represents the real-time remaining load.

[0140] 3. Route Simulation and Optimization: The generated dispersal path is simulated using MissionPlanner software to check coverage, missed areas, and potential risks (such as sharp turns). The path is optimized based on the simulation results to reduce unnecessary turns and flight distances, ensuring maximum utilization of power and time. For large-area operations, multiple takeoffs and landings and resupply plans are rationally arranged, and return and resupply routes are optimized.

[0141] Based on the optimization results, a dissemination path for the drone is generated, including flight direction, speed (3-5 m / s recommended), altitude (7-10 m recommended to ensure dissemination effect), and dissemination switch trigger point. The path planning results are output in geographic coordinates and converted into a KML format file executable by the drone. Emergency strategies are set in the task file, including low battery return-to-home, loss of connection return-to-home, and detour strategies when encountering obstacles.

[0142] Specifically, path planning includes: a cost evaluation submodule, used for evaluating dynamic window candidates. Trajectory calculation , , , , , and output the overall cost. The variable-rate application prediction submodule is used to predict the future short-term application rate based on the application mechanism calibration curve, flight altitude, velocity, particle size distribution, and remaining payload. .

[0143] Specifically, set the weight adaptive strategy: when Time Improvement and The weights are adjusted to enhance coverage integrity and load matching; after weight adjustment, the following conditions are met. Or it can be normalized and mapped to [0,1]; The threshold for online learning is adaptively updated based on historical assignment statistics.

[0144] Specifically, this invention proposes precise seeding irrigation. The flight control system adjusts the flight altitude to a suitable level for seeding irrigation (7-10 meters). Based on the path planning structure, the pesticide / fertilizer storage and supply device and the seeding device implement precise variable seeding according to the nitrogen distribution map. Dynamic real-time obstacle avoidance is simultaneously achieved during the operation. The specific method is as follows:

[0145] Variable-rate application control based on spectral data: Combining the crop nitrogen content distribution estimated from previous hyperspectral images, the application rate was increased to 110%-120% of the baseline amount in nitrogen-deficient areas, while the application rate was uniformly sprayed at the baseline amount in normal areas.

[0146] Dynamic obstacle avoidance and path adaptation: During operation, the drone analyzes obstacle information (such as fixed obstacle positions) in the flight path in real time based on built-in algorithms, and avoids obstacles by optimizing flight speed and direction. Specifically, the optimal path is calculated using an improved dynamic window method (DWA), and the flight trajectory is dynamically adjusted to avoid missed spraying due to obstacle avoidance. When a sudden obstacle is detected, the drone triggers an automatic selection mode to determine whether to hover or detour, ensuring operational safety.

[0147] Specifically, this embodiment of the invention records operational data in real time, including the wireless communication module uploading flight trajectories, nitrogen estimation results, seeding amounts, obstacle information, and other data to a cloud data platform. The data transmission and reception module is responsible for the efficient exchange of data such as base station correction information and rework instructions. The cloud data platform stores and analyzes the data, facilitating subsequent operational optimization and decision support.

[0148] Specifically, the operation's completion and return process includes the flight control system automatically planning the shortest return path, taking into account remaining battery power and obstacle distribution to ensure a safe return. The communication system monitors the drone's status throughout the process and promptly issues emergency return commands in case of abnormalities.

[0149] Specifically, this invention integrates millimeter-wave radar / obstacle databases to dynamically adjust local speed and heading; an emergency return-to-base strategy is triggered in abnormal states to enhance operational safety and redundancy. Data closed-loop construction: Flight, sowing, estimation, environmental, and positioning data are uploaded to the cloud in real time, supporting historical comparison, model iteration, and cross-seasonal decision optimization, forming a traceable operational archive. Resource utilization optimization: The planning comprehensively considers payload, power, and resupply point locations to achieve phased task scheduling and reduce the proportion of non-productive flights. Strong adaptability: It can be extended to multispectral, thermal imaging, pest and disease identification, or multi-aircraft collaborative algorithms, possessing good modularity and scalability. Stability and consistency: Multi-source data synchronization and fusion reduce the impact of single sensor drift on results, improving the comparability of repeated operations and long-term monitoring. Balancing environmental and economic benefits: It reduces redundant fertilizer application, suppresses nutrient loss and non-point source pollution, and improves crop nutrient balance and yield quality while reducing input costs. Improved coverage uniformity: By coordinating target, obstacle avoidance, and speed terms in the cost function, spray width / sowing parameters are stabilized, reducing the probability of missed / double spraying.

[0150] The implementation of the various embodiments of this invention is based on programmed processing by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of this invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, this invention provides a plant protection drone operation system based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation. This system is used to execute a plant protection drone operation method based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation from the above method embodiments.

[0151] The system includes: a data preparation module, used to determine the operating area of ​​the agricultural drone and calculate the drone's flight altitude using hyperspectral camera parameters and navigation positioning data; and a nitrogen deficiency area determination module, used to acquire hyperspectral images of crops within the operating area of ​​the agricultural drone in real time using a hyperspectral camera, determine nitrogen deficiency areas using a dynamic threshold algorithm based on the pixel spectral characteristics in the hyperspectral images, and output the boundary shape and area of ​​the nitrogen deficiency areas.

[0152] The coordinate transformation module is used to obtain the pixel coordinates of the nitrogen-deficient area based on the boundary shape of the nitrogen-deficient area and convert them into real geographic coordinates; the path planning module is used to plan the sowing path based on the real geographic coordinates and area of ​​the nitrogen-deficient area, using the extended dynamic window method, and finally achieves precise sowing irrigation.

[0153] The agricultural drone operation system based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation provided in this embodiment of the invention

[0154] To address existing technical challenges, this paper employs several modules, introducing dynamic thresholding and maximum entropy thresholding methods. It combines Canny fine-grained edge detection with RTK, GCP, and DEM multi-source correction. An extended dynamic window method is used to search for the cost-optimal action in real-time within the feasible velocity-angular velocity set. Based on nitrogen distribution, differentiated supplementation is applied to nitrogen-deficient areas, adapting to background differences in light intensity, variety, and growth stage. This improves the accuracy and robustness of nitrogen deficiency area identification, reduces regional boundary jitter and area calculation errors, provides a reliable spatial benchmark for precise variable seeding, and achieves high-precision mapping from pixel-level spectral information to real geographic coordinates, reducing spatial offset accumulation.

[0155] Finally, it should be noted that the above specific embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and many variations are possible. Any simple modifications, equivalent changes, and alterations made to the above specific embodiments based on the technical essence of the present invention should be considered within the protection scope of the present invention.

Claims

1. A method for operating agricultural drones based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation, characterized in that, include: S1. Determine the operating area of ​​the agricultural drone and calculate the drone's flight altitude using hyperspectral camera parameters and navigation positioning data; S2. Hyperspectral images of crops in the operating area of ​​the plant protection drone are acquired in real time by a hyperspectral camera. Based on the pixel spectral characteristics in the hyperspectral images, a dynamic threshold algorithm is used to determine the nitrogen nutrient deficiency area and output the boundary shape and area of ​​the nitrogen nutrient deficiency area. S3. Obtain the pixel coordinates of the nitrogen-deficient region based on the boundary shape of the nitrogen-deficient region, and convert them into real geographic coordinates; S4. Based on the real geographical coordinates and area of ​​nitrogen-deficient regions, the extended dynamic window method is used for seeding path planning, ultimately achieving precision seeding irrigation.

2. The method for agricultural drone operations based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation according to claim 1, characterized in that, Based on the pixel spectral features in the hyperspectral image, a dynamic thresholding algorithm is used to determine nitrogen deficiency regions, including: The gray-level histogram is calculated based on the pixel spectral features in the hyperspectral image. The gray-level histogram is then normalized to a probability distribution. Based on the probability distribution, the hyperspectral image is segmented using the maximum entropy thresholding method to obtain a binarized image. The binarized image is divided into nitrogen-deficient regions and normal regions. The Canny algorithm is used to perform edge detection on the binarized image, extract the edge pixel coordinates of the nitrogen-deficient region, and output the boundary shape and area of ​​the nitrogen-deficient region.

3. The method for agricultural drone operation based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation according to claim 2, characterized in that, Edge detection of binarized images is performed using the Canny algorithm, including: Apply Gaussian smoothing to the binarized image; The Sobel operator is used to calculate the gradient components in the horizontal and vertical directions of the Gaussian smoothed image, and the gradient magnitude and gradient direction are obtained. For each pixel in the image, compare its gradient magnitude with that of its neighboring pixels along the gradient direction, and retain local maxima. Set high and low thresholds and , When the gradient magnitude is greater than or equal to the high threshold, it is marked as a strong edge; when the gradient magnitude is less than the high threshold but greater than or equal to the low threshold, it is marked as a weak edge connected to the strong edge; the edge detection result is obtained based on the strong edges and the weak edges connected to the strong edges.

4. The method for agricultural drone operation based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation according to claim 1, characterized in that, The S3 includes: Obtain the pixel coordinates of the areas identified as nitrogen-deficient, and obtain the ground coordinates of the nitrogen-deficient areas by mapping the pixel coordinates to the real-time spatial coordinates of the UAV. The formula for converting ground coordinates of nitrogen-deficient areas to actual geographic coordinates is as follows: , , , in, Longitude Latitude The height of the ground point. For the Earth's radius, It is a point in the ground coordinate system.

5. The method for agricultural drone operation based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation according to claim 1, characterized in that, The S4 includes: Based on the actual geographical coordinates and area of ​​the nitrogen-deficient region, the region is divided into several sub-regions, and each sub-region is gridded to obtain the boundary coordinates and grid cells of each sub-region. Based on the boundary coordinates and grid cells of each sub-region, the extended dynamic window method is used for seeding path planning, and its evaluation function is: ， in, This represents the path cost to reach the target area; This indicates the cost of avoiding obstacles; Indicates the cost of speed optimization; Indicates the weighting coefficient. Indicates based on current speed With angular velocity The normalized cost of the area of ​​uncovered nitrogen-deficient grids newly covered within the prediction time window relative to the largest newly covered area in the candidate set. A normalized measure of the deviation between the actual application rate and the target application rate of the predictor variable; This represents the cost of the remaining load being insufficient relative to the projected demand in nitrogen-deficient areas that are not yet covered.

6. The method for agricultural drone operation based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation according to claim 5, characterized in that, , The expressions are as follows: ， ， ， in, The area of ​​unseeded, nutrient-deficient regions newly covered within the prediction window for the current candidate velocity combination. The maximum newly added coverage area in the same window candidate set; N is the number of grid cells covered by the predicted trajectory; For the i-th grid, predict the instantaneous or segmental average application rate. The application rate for this grid target. The baseline application rate; The remaining unresolved nutrient deficiency area. The average target application rate for the remaining area. This represents the real-time remaining load.

7. The method for agricultural drone operation based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation according to claim 4, characterized in that, After obtaining the actual geographical coordinates of the nitrogen-deficient area, the following is also included: Geographic control point correction: Compare the actual geographic coordinates with the known geographic control points and use the least squares method for correction; Terrain elevation data correction: Adjustments are made for errors caused by terrain undulations using high-resolution terrain elevation data.

8. The method for agricultural drone operation based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation according to claim 1, characterized in that, The expression for calculating the drone's flight altitude is: , in, It is the flight altitude; It is the sensor width; It is the lens focal length; It is the image width.

9. A plant protection drone operation system based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation, characterized in that, include: The data preparation module is used to determine the operating area of ​​agricultural drones and calculate the drone's flight altitude using hyperspectral camera parameters and navigation and positioning data. The nitrogen nutrient deficiency area determination module is used to acquire hyperspectral images of crops in the operation area of ​​the plant protection drone in real time through a hyperspectral camera, determine the nitrogen nutrient deficiency area based on the pixel spectral characteristics in the hyperspectral image using a dynamic threshold algorithm, and output the boundary shape and area of ​​the nitrogen nutrient deficiency area. The coordinate transformation module is used to obtain the pixel coordinates of the nitrogen-deficient area based on the boundary shape of the nitrogen-deficient area and convert them into real geographic coordinates. The path planning module, based on the real geographical coordinates and area of ​​nitrogen-deficient regions, uses an extended dynamic window method to plan the sowing path, ultimately achieving precise sowing irrigation.

10. A plant protection drone operation system based on BeiDou high-precision positioning and hyperspectral nitrogen nutrient estimation according to claim 9, characterized in that, The system also includes: The wireless communication module is used to upload flight trajectory, nitrogen estimation results, seeding amount, and obstacle information to the cloud data platform in real time; A cloud-based data platform is used for data storage and analysis, providing optimization and decision support for drone operations. The communication module is used to monitor the drone's status throughout the entire process and issue emergency return-to-home commands in a timely manner in case of abnormalities; The flight control module is used to automatically plan the shortest return route.

Images