Method for optimizing pasture seed sowing path based on air flow simulation data

Abstract

The application discloses a method for optimizing a pasture seed sowing path based on airflow simulation data, relates to the technical field of agricultural seeding and unmanned aerial vehicle application, and comprises the following steps: S1, collecting wind speed time series and terrain airflow disturbance data at a current flight height by using a multi-point wind field sensor carried on an unmanned aerial vehicle, analyzing local wind field characteristics in combination with a terrain interference factor, constructing an airflow simulation model of a target region based on the collected data, and obtaining a wind field dynamic distribution map of the target region; and S2, simulating a pasture seed throwing trajectory according to the wind field dynamic distribution map, calculating a trajectory offset amount caused by wind field disturbance, and determining an initial drift compensation vector of seed falling by fusing a wind field influence weight and a trajectory simulation error; the method can greatly improve pasture reseeding efficiency and grassland vegetation restoration effect.

Description

Technical Field

[0001] This invention relates to the field of agricultural seeding and drone application technology, specifically to a method for optimizing the seeding path of forage grass based on airflow simulation data. Background Technology

[0002] In grassland ecological restoration and forage reseeding operations, drone seeding technology has become an important means of vegetation restoration due to its wide coverage, high efficiency, and low reliance on manual labor. During forage seeding, drones typically fly along preset routes and scatter seeds at a fixed frequency to achieve the most uniform seeding distribution possible within the target area. However, the complex and volatile wind characteristics in actual grassland environments make it difficult to achieve stable and uniform seeding results using traditional fixed-path or simple feedback seeding strategies.

[0003] In existing technologies, drone-based seeding methods generally rely on path planning based on experience or historical wind field data. However, wind speed and direction in grassland areas typically exhibit significant spatial heterogeneity and temporal dynamic changes due to altitude, terrain morphology, and local airflow disturbances. As small particles significantly affected by wind, forage seeds are easily influenced by factors such as instantaneous wind speed, sudden changes in wind direction, and terrain-induced eddies, resulting in significant drift deviations. This often leads to seed landing points that do not match the planned locations, resulting in both sparse and concentrated areas of seeding, which in turn affects the uniformity of subsequent forage growth and ecological restoration. Existing technologies attempt to correct seed offsets using wind speed sensors or simple real-time compensation, but these methods are usually based on single-point or low-dimensional wind field data and cannot effectively characterize the dynamic wind field structure in complex grassland environments. Furthermore, wind field changes exhibit significant short-term variability and spatial coupling characteristics; relying solely on instantaneous measurements cannot accurately predict wind field evolution during seeding. If the compensation strategy is lagging or the model accuracy is insufficient, trajectory errors will be further amplified. Especially in areas with complex terrain, the local airflow rises, falls, recirculation, or vortices caused by terrain disturbances make it difficult to accurately estimate seed trajectories using traditional methods. Furthermore, existing methods generally lack a dynamic evaluation mechanism for the uniformity of seeding distribution. When local deviations occur in the seeding area, the system often fails to identify the degree of deviation in a timely manner, nor can it make targeted adjustments to the flight path or seeding parameters based on real-time wind field changes. Therefore, when there are continuous wind field disturbances or changes in wind field trends, traditional methods struggle to perform continuous, adaptive path optimization during seeding, resulting in difficulties in ensuring overall seeding quality. Summary of the Invention

[0004] The purpose of this invention is to provide a method for optimizing the seed dispersal path of forage grass based on airflow simulation data, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing the seed dispersal path of forage grass based on airflow simulation data, comprising:

[0006] S1. Use multi-point wind field sensors mounted on the UAV to collect wind speed time series and terrain airflow disturbance data at the current flight altitude, and combine them with terrain interference factors to analyze the local wind field characteristics. Based on the collected data, construct an airflow simulation model of the target area to obtain a dynamic wind field distribution map of the target area.

[0007] S2. Simulate the trajectory of pasture seed scattering based on the wind field dynamic distribution map, calculate the trajectory offset caused by wind field disturbance, and combine the wind field influence weight and trajectory simulation error to determine the initial drift compensation vector for seed fall.

[0008] S3. The residual calculation method is used to evaluate the accuracy of the deviation between the initial drift compensation vector and the actual landing point. Combined with the landing point distribution density and the granularity of the region division, it is determined whether the unevenness of the seed landing point distribution exceeds the preset threshold standard.

[0009] S4. When the unevenness of the landing point distribution exceeds the preset threshold standard, construct a state vector dimension that includes wind speed time series information and terrain airflow disturbance parameters, integrate historical wind field data and disturbance spatial distribution characteristics, and obtain the airflow simulation state description of the current wind field.

[0010] S5. The state vector is updated based on real-time observation input, and the system deviation in the wind speed change process is corrected by combining the state update cycle and noise interference separation model to obtain the short-term wind field change trend prediction results.

[0011] As can be seen from the above technical solution, the present invention has the following beneficial effects:

[0012] This method for optimizing forage seed dispersal paths based on airflow simulation data can construct a multi-dimensional, multi-temporal dynamic wind field simulation model to accurately predict the trajectory of forage seeds, addressing the complex wind field distribution, dramatic instantaneous fluctuations, and significant terrain-induced disturbances in grassland environments. By fusing wind speed time series, terrain airflow disturbance factors, and historical wind field data, this invention can effectively capture short-term wind field change trends and update wind field status in real time, overcoming the shortcomings of traditional technologies that rely on fixed paths or single-point wind speed measurements, resulting in compensation lag and low prediction accuracy. By introducing residual evaluation and landing point distribution density analysis mechanisms, this invention can promptly identify uneven landing points within the dispersal area and dynamically adjust the seed dispersal timing, dispersal angle, and UAV flight attitude when thresholds are exceeded, achieving adaptive path correction throughout the entire process. Simultaneously, through continuous updates of drift compensation vectors and dynamic path adjustment parameters, this invention enables the UAV to maintain real-time control over seed trajectory deviations in complex wind fields, fundamentally improving the uniformity and accuracy of dispersal. In large-scale grassland ecological restoration scenarios, this invention not only significantly reduces sowing errors and improves sowing uniformity, but also enhances the operational stability and adaptability of drones under complex airflow conditions, which can greatly improve the efficiency of pasture reseeding and the effect of grassland vegetation restoration. Attached Figure Description

[0013] Figure 1 This is a flowchart of the pasture seed dispersal path optimization method based on airflow simulation data according to the present invention. Detailed Implementation

[0014] 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.

[0015] like Figure 1As shown, this invention provides a technical solution: a method for optimizing the seed dispersal path of forage grass based on airflow simulation data, including: S1, using a multi-point wind field sensor mounted on a UAV to collect wind speed time series and terrain airflow disturbance data at the current flight altitude, and analyzing the local wind field characteristics in conjunction with terrain interference factors, constructing an airflow simulation model of the target area based on the collected data, and obtaining a dynamic wind field distribution map of the target area; S2, simulating the seed dispersal trajectory of forage grass based on the dynamic wind field distribution map, calculating the trajectory offset caused by wind field disturbance, and integrating the wind field influence weight and trajectory simulation error to determine the initial drift compensation vector for seed fall; S3, using residual calculation to evaluate the accuracy of the deviation between the initial drift compensation vector and the actual landing point, and combining the landing point distribution density and the granularity of the area division to determine whether the unevenness of seed landing point distribution exceeds a preset threshold standard; S4, when the landing point distribution is uneven... When the uniformity exceeds the preset threshold standard, a state vector dimension containing wind speed time series information and terrain airflow disturbance parameters is constructed. Historical wind field data and disturbance spatial distribution characteristics are integrated to obtain a simulated airflow state description of the current wind field. S5. The state vector is updated based on real-time observation input, and the system deviation in the wind speed change process is corrected by combining the state update cycle and noise interference separation model to obtain the short-term wind field change trend prediction result. S6. According to the short-term wind field change trend and terrain interference factor, the seed release timing and release angle are dynamically adjusted, and the landing point distribution is monitored in real time by combining the data update frequency to determine the dynamic path adjustment parameters used to correct the flight trajectory. S7. The UAV flight altitude and flight speed are updated based on the dynamic path adjustment parameters. The landing point distribution density and threshold judgment results are integrated, and the final path plan that meets the uniform seeding requirements is generated based on continuous monitoring of the release trajectory deviation.

[0016] This implementation method is based on a dynamic disturbance compensation mechanism for UAV seed dispersal under complex wind conditions. Multi-point wind field sensors on the UAV are used to collect wind speed time series and airflow disturbance information caused by different terrains within the flight altitude range. By introducing force field disturbance coefficients and terrain interference factors, a wind field simulation model can be constructed in the target area, allowing the dynamic characteristics of wind speed changes with time and terrain to be expressed. Based on the dynamic wind field distribution map, the seed falling trajectory can be physically simulated. The trajectory offset is calculated based on gravity, drag, horizontal disturbance components, and a turbulence model to obtain an initial drift compensation vector. By comparing the expected landing point after compensation with the actual landing point, the fit of the compensation vector can be determined through residual analysis, and the dispersal uniformity can be judged by combining the landing point distribution density within the area. When uniformity is insufficient, the system constructs a data state vector containing wind field time series, airflow disturbance intensity, and spatial gradient. By fusing historical wind field evolution patterns, a real-time dynamic description of the current wind field is obtained. As the UAV continues to fly, the state vector can be updated using real-time observation input, and the wind speed estimation error can be calibrated through a noise interference separation model to predict wind field change trends over short time scales. After obtaining the prediction results, the system adjusts the seed dispersal angle, timing, and related dynamic path parameters based on wind field trends and terrain interference factors, enabling the UAV's altitude, speed, and dispersal direction to adaptively adjust with changes in the wind field. Finally, by continuously monitoring the distribution of seed landing points, a path plan that meets the requirements for dispersal uniformity is generated, achieving dynamic dispersal control based on an airflow compensation mechanism.

[0017] S1 includes acquiring wind speed time series and terrain airflow disturbance data at the current flight altitude using multi-point sensors mounted on the UAV, obtaining data acquisition results; performing interference analysis on local wind field characteristics based on the data acquisition results and terrain interference factors to determine the changing trend of local characteristics; constructing an airflow simulation model of the target area using the changing trend of local characteristics and flight altitude data, and outputting simulation results by inputting the changing trend and altitude parameters to obtain the simulation output; generating a dynamic wind field distribution map of the target area based on the simulation output, and identifying abnormal points in the dynamic distribution; if the abnormal points in the dynamic distribution exceed a preset threshold, marking the abnormal points to obtain the distribution characteristics of wind field anomaly detection.

[0018] In this embodiment, a multi-point sensor mounted on the UAV first continuously samples the wind speed at a preset time interval at the current flight altitude. The time interval is determined during the mission planning phase based on the UAV's maximum flight speed and the desired spatial resolution. In this embodiment, the time interval is fixed at 1 second. This results in a series of wind speed measurements arranged in chronological order at each sensor location, forming a wind speed time series at the current flight altitude. Each sensor records its fixed relative position on the UAV body during installation. During flight, the UAV uses its own navigation device to record the latitude, longitude, and altitude along its flight path in real time, and correlates the UAV's position at each wind speed sampling moment with the corresponding sensor position. The relative installation positions are converted to assign a corresponding geographic location coordinate and altitude information to each wind speed measurement. While collecting wind speed time series data, the multi-point sensors also record airflow disturbance data caused by terrain. This data is obtained by comparing the instantaneous wind speed at each sampling location with the average wind speed at multiple sampling locations at the same altitude and within the same time period. The comparison steps are as follows: at the end of each sampling period, the average wind speed measurements from all sensors within that period are calculated; the wind speed measurement at a given point is subtracted from this average value, and the difference is taken as the terrain airflow disturbance intensity at that point. This disturbance intensity, together with the corresponding terrain type, constitutes terrain airflow disturbance data, thus forming a data set including time, location, wind speed, and disturbance intensity. The data collection results are used to determine the terrain interference factor. Before the task begins, the terrain of the target area is surveyed, and the area is divided into several levels based on slope, elevation difference, and surface undulation. For example, flat areas are set as level 1, gentle slopes as level 2, and steep slopes and valleys as level 3. A terrain interference factor value is assigned to each level. In this embodiment, the terrain interference factor for level 1 is set to 1, for level 2 to 2, and for level 3 to 3. This factor is then mapped one-to-one with the terrain level of each sampling location. Subsequently, the system performs interference analysis on the local wind field characteristics based on the data collection results and the terrain interference factor. The local area is divided into a certain spatial and temporal range. The spatial range is achieved by dividing the target area into several regular grids in the horizontal direction. The grid side length is set according to the seeding accuracy requirements during task planning. In this embodiment, the grid side length is set to 10 meters. The temporal range is achieved by selecting several consecutive sampling times on the wind speed time series to form a time window. In this embodiment, the time window length is set to 30 seconds. Within each grid and each time window, the average wind speed, maximum wind speed, and minimum wind speed of all sampling points within that range are statistically analyzed. The change in the average wind speed with the time window sequence is calculated. At the same time, the terrain interference factor of all sampling points within the grid is averaged to obtain the comprehensive terrain interference factor of the grid.The disturbance intensity index for the local area is obtained by multiplying the wind speed change by the comprehensive terrain disturbance factor. The changing trend of the local wind field characteristics is determined by the direction and magnitude of the disturbance intensity index changes over multiple consecutive time windows. When the disturbance intensity index continuously increases in adjacent time windows, the wind speed in the local area is determined to be increasing; when the disturbance intensity index continuously decreases, it is determined to be decreasing; when the disturbance intensity index frequently changes between positive and negative with large amplitude, it is determined that there is a significant turbulent trend in the area. When constructing the airflow simulation model of the target area, the average wind speed, disturbance intensity, and... (The sentence is incomplete and requires further context to be fully translated.) The system combines trend analysis with UAV flight altitude data, which is recorded by the UAV's altitude measurement device at each sampling moment. The system divides the target area into several altitude levels along the altitude direction. In this embodiment, the interval between altitude levels is set to 5 meters. The wind speed at each grid at different altitude levels is estimated through interpolation extrapolation. That is, based on the real-time measured wind speed at that grid, wind speed records at other altitudes at the same location during historical flights, and local trend analysis, a wind speed estimate is given for each altitude level within the current time window, thus forming an airflow simulation model that includes three-dimensional spatial location and time dimensions.

[0019] When running the airflow simulation model, the changing trend information and height parameters corresponding to each grid, each height level, and each time window are taken as input. The model determines the direction of wind speed increase or decrease over time based on the changing trend, and determines the change of wind speed with height based on the height parameters. It then calculates the predicted wind speed for each grid and height level in subsequent time windows, thus obtaining the simulation results. These simulation results constitute the output of the simulation. When generating the dynamic wind field distribution map of the target area, the output of the simulation is arranged according to the spatial grid position. Each grid corresponds to a wind speed magnitude and a wind direction at a certain time. The system maps the wind speed magnitude to different numerical values, etc. The system maps wind direction to different directional markers and overlays this information onto a planar location within the target area, thus obtaining a dynamic wind field distribution map at different times. When identifying anomalies in the dynamic distribution, the system compares the wind speed and direction of each grid at the current moment with the average wind speed and direction of that grid over the previous few time windows, as well as the average wind speed and direction of neighboring grids at the current moment. If the difference between the current wind speed of a grid and its historical average wind speed is greater than twice the average historical wind speed variation of that grid, or if the current wind direction of that grid is completely opposite to the average current wind direction of neighboring grids, then that grid is removed from the list. The location corresponding to a grid cell is identified as an anomaly and recorded as an anomaly marker in the system. Subsequently, the system counts the number of anomalies in the current wind field dynamic distribution map and divides this number by the total number of grid cells involved in the analysis to obtain the percentage of anomalies. This percentage is then compared with a preset threshold, which is determined before system deployment using extensive historical UAV seeding task data. The specific steps are as follows: multiple historical tasks are selected, and the percentage of anomalies in each task is calculated using the above method. Simultaneously, the results of the forage seed landing site survey after the task are used to determine whether the seeding meets the uniformity requirements. Finally, the percentage of anomalies is calculated among all tasks that meet the uniformity requirements. The maximum value is increased by a fixed safety margin. In this embodiment, the safety margin is set to 0.02, and the value after increasing the safety margin is determined as the preset threshold. In actual operation, when the proportion of abnormal points in the current wind field dynamic distribution map is greater than the preset threshold, the system determines that the abnormal points in the dynamic distribution exceed the preset threshold and marks all abnormal points uniformly. The marking content includes at least the grid index where the abnormal point is located, the corresponding geographical coordinates, the current time, and the abnormality type. By summarizing and superimposing the spatial locations of all abnormal points, a wind field anomaly detection distribution feature in the target area is formed. This distribution feature directly indicates which areas have strong or unstable airflow disturbances.

[0020] S2 includes using offset results to fuse influence weight values, adjusting local wind characteristics, and determining the disturbance distribution in local wind characteristics; constructing a simulation error term based on the disturbance distribution and height data parameters, and outputting error correction by inputting the disturbance distribution and height parameters to obtain the output of the error term; generating outlier markers through the output of the error term, and if the outlier markers exceed a preset threshold, marking the distribution feature map to obtain the feature value of the distribution feature map; and using the feature value of the distribution feature map to fuse the wind field disturbance factor to determine the initial drift compensation vector for seed fall.

[0021] In this embodiment, firstly, based on airflow simulation and seed dispersal simulation, the offset of the seed landing point relative to the target landing point is determined for each location within the target area. This offset is the distance difference in the horizontal direction. Specifically, during the trajectory simulation stage, the target landing point coordinates and simulated landing point coordinates of each seed are recorded. The coordinate differences are calculated separately in the east-west direction and in the north-south direction. Then, the squares of the differences in the two directions are summed and the square root is taken to obtain the offset distance value. Simultaneously, the offset direction is recorded. The offset distances obtained from multiple simulations at the same location are averaged to form the offset result for that location. After obtaining the offset result, to ensure that different locations affect the overall seeding quality... The impact of quantity is differentiated. During the task planning phase, an impact weight value is pre-set for each location. The impact weight value is set to a range of 0 to 1, specifically three fixed levels: 0.3, 0.6, and 0.9. Locations with gentle terrain, minimal historical wind field disturbance, and general seeding density requirements are uniformly assigned a value of 0.3. Locations with some terrain undulation or moderate historical wind field disturbance and higher seeding density requirements are assigned a value of 0.6. Locations with complex terrain or located in key seeding areas and significant historical wind field disturbance are assigned a value of 0.9. Subsequently, the system calculates the offset result for each location and its corresponding impact weight value one by one, and multiplies the offset result by the impact weight value to obtain the offset correction. The system adjusts the local wind characteristics based on the magnitude and direction of the offset correction value. The adjustment method is as follows: when the offset direction at a certain location points upwind and the offset correction value is large, the effective disturbance intensity along the offset direction in the local wind characteristics at that location is increased; when the offset direction points downwind, the effective disturbance intensity in that direction is decreased, so that the adjusted local wind characteristics can reflect the true impact of the trajectory simulation feedback. After the system completes the adjustment of the local wind characteristics at all locations throughout the entire area, it calculates the difference in disturbance intensity before and after the adjustment for each location. The absolute value of the difference is taken as the disturbance intensity value for that location, and the locations are sorted from smallest to largest. The top 30% of locations are classified as low disturbance, and the remaining 30% are classified as medium disturbance. The middle 40% of the positions are classified as medium disturbances, and the bottom 30% are classified as high disturbances, thus obtaining a disturbance distribution covering the entire target area. This disturbance distribution is used to describe the spatial variation of local wind characteristics. When constructing the simulation error term, the system combines the aforementioned disturbance distribution with altitude data parameters. The altitude data parameters are determined as several fixed altitude layers during the mission configuration phase based on the available flight altitude range of the UAV and the requirements for pasture seed dispersal. In this embodiment, the altitude layer interval is set to 5 meters, and several altitude layers are sequentially divided starting from the minimum safe altitude. During the test flight or simulation, the aforementioned trajectory simulation is repeated on each altitude layer to obtain the offset results of the same position on different altitude layers.

[0022] When constructing the simulation error term, for each location and each height layer, the theoretical offset distance calculated from the perturbation distribution at the current height layer is compared with the offset result obtained from the simulation. The difference between the two is calculated, and the absolute value of the difference is taken as the error component of that location at that height layer. Then, the error components of all height layers at the same location are weighted and summed according to the pre-set weights of the height data parameters to obtain the total error value of that location. The height weights are set as follows: the weight of the height layer where the main seeding is carried out is set to 0.5, the weight of the adjacent height layers above and below it is set to 0.25, and the weight of the other height layers is set to 0. This way, the total error value focuses more on the error performance near the main seeding height. The system combines the total error values ​​of all locations to form the simulation error term, and uses the perturbation distribution and height data parameters as inputs to perform error correction calculations on the simulation error term one by one according to the location. The error correction method is to compare the total error value with the upper limit of the allowable error. The upper limit of the allowable error is determined before system deployment based on the seeding uniformity requirements. The error correction value is set to a fixed value, which is 1 meter in this embodiment. When the total error value at a certain location is less than the upper limit of the allowable error, the error correction value at that location is set to 0. When the total error value is greater than the upper limit of the allowable error, the error correction value is set to the total error value minus the upper limit of the allowable error, thus forming the output of the error item. The system generates anomaly point markers based on the output of the error item. The generation method is as follows: for each location, when the error correction value is greater than 0, the location is marked as an anomaly point. The error correction value is divided into 3 levels according to the magnitude of the value. The error correction value in the range of 0 to 0.5 meters is marked as a slight anomaly, the range of 0.5 meters to 1 meter is marked as a moderate anomaly, and the value greater than 1 meter is marked as a severe anomaly. The anomaly point marker records the location index, error correction value, and anomaly level. The system counts the number of locations marked as anomalies among all current locations and calculates the ratio between the number of anomalies and the total number of locations. This ratio is compared with a preset threshold, which is determined through historical task data before system deployment.

[0023] The specific steps are as follows: Select multiple completed sowing tasks, calculate the outlier ratio for each task using the method described above, and simultaneously determine whether the sowing meets the uniformity requirements based on the actual landing point survey results. Take the maximum outlier ratio from all tasks that meet the uniformity requirements, and then add a safety margin of 0.05 to this value. Set the sum as the preset threshold. During actual operation, when the outlier ratio of the current task exceeds the preset threshold, it is determined that the outlier marking exceeds the preset threshold. The system explicitly marks all outlier locations on the distribution feature map. The distribution feature map is based on the target area grid and displays the disturbance distribution and anomaly level of each location simultaneously. Based on this, the feature value of the distribution feature map is calculated. The steps for calculating the feature value are as follows:

[0024] First, for all locations marked as outliers, their error correction values ​​are multiplied by their corresponding influence weight values ​​and summed. Then, this sum is divided by the total number of outliers to obtain the average weighted error of the outlier area. This average weighted error is then multiplied by the proportion of outliers to obtain a single value, which serves as the feature value of the distribution feature map, used to comprehensively characterize the combined impact of current wind field disturbances and simulation errors on seeding. Finally, the system uses the feature value of the distribution feature map to fuse with the wind field disturbance factor to determine the initial drift compensation vector for seed descent. The wind field disturbance factor has already been given specific values ​​for each location in the aforementioned airflow simulation stage, and the value is proportional to the wind speed change and local disturbance intensity at that location. In this implementation, the wind field disturbance factor is uniformly normalized to between 0 and 1. When calculating the initial drift compensation vector, the system calculates the east-west compensation distance and the north-south compensation distance for each location. The calculation steps are as follows: in the east-west direction... In terms of direction, the offset direction of the location is determined to be east or west. The characteristic value of the distribution feature map is multiplied by the wind field disturbance factor and the offset result at that location. If the offset direction is east, the product is used as the compensation distance to the west; if the offset direction is west, the product is used as the compensation distance to the east. The same method is used in the north-south direction. The reverse compensation distance is determined according to whether the offset direction is south or north. Then, the compensation distances of all locations in the east-west direction are spatially averaged to obtain the overall initial drift compensation distance in the east-west direction. The compensation distances of all locations in the north-south direction are spatially averaged to obtain the overall initial drift compensation distance in the north-south direction. The two together constitute the initial drift compensation vector for seed falling. This vector is used in subsequent control steps to correct the seed release timing and direction, so that the seeds still fall into the target sowing area as much as possible even in the presence of wind field disturbance, thereby achieving quantitative compensation for wind field disturbance and simulation error.

[0025] S3 includes using a residual calculation method to obtain an accuracy assessment result by subtracting the actual landing point deviation from the initial drift compensation vector, thus obtaining the deviation value in the accuracy assessment result; weighting and fusing the landing point distribution density by the deviation value to determine the distribution density characteristics, thus obtaining the density distribution map in the distribution density characteristics; calculating the non-uniformity index based on the density distribution map and the regional division granularity, judging the non-uniformity of seed landing point distribution, thus obtaining the non-uniformity judgment index; if the judgment index exceeds the preset threshold standard, then correcting and adjusting the compensation vector according to the wind direction variation based on the density distribution map, thus determining the initial drift compensation vector for seed falling.

[0026] In this embodiment, after obtaining the initial drift compensation vector for seed descent through the aforementioned steps, the actual landing points of trial sowing or simulated sowing are statistically analyzed within the target area according to a pre-defined regular grid. Each grid corresponds to a target landing point position and one or more actual landing point positions. The initial drift compensation vector is the overall compensation distance calculated in the previous steps based on wind field disturbance and distribution characteristic maps, containing two components: east-west and north-south. It is used to offset the nominal landing point of the seed during scattering control. The actual landing point deviation is the offset distance of the average position of the actual landing points within the same grid relative to the target center position of that grid, also containing two components: east-west and north-south. Its determination process is as follows: in each trial sowing or operation, the actual landing point coordinates of all seeds in each grid are collected, and the values ​​of these coordinates in the east-west direction are averaged and compared with... The coordinates of the grid center in the east-west direction are compared to obtain the deviation distance in the east-west direction; the same method is used to obtain the deviation distance in the north-south direction, thus obtaining the actual landing point deviation of the grid. The system uses a residual calculation method to subtract the component of the initial drift compensation vector in each direction from the actual landing point deviation in the corresponding direction of the grid, obtaining the residual distance of the grid in the east-west direction and the residual distance in the north-south direction. Then, the squares of the residual distances in the two directions are summed and the square root is taken to obtain the residual value of the grid. This residual value is used as the deviation value in the accuracy evaluation result. The smaller the deviation value, the more accurate the correction of the grid by the initial drift compensation vector is; the larger the deviation value, the more significant the undercompensation or overcompensation of the grid. The system performs the above residual calculation on all grids in the target area one by one to form a deviation value distribution covering the entire area.Subsequently, to comprehensively consider the impact of landing point deviation and the number of landing points on uniformity, the system determines the distribution density characteristics by weighted fusion of landing point distribution density based on deviation values. The landing point distribution density is the number of seeds per unit area within each grid, determined as follows: after sowing, the number of seeds in each grid is counted, and the count result is divided by the actual area of ​​the grid to obtain the landing point distribution density for that grid. The deviation value weighted fusion process is as follows: first, the maximum value among all grid deviation values ​​in the current task is found; then, the deviation value of each grid is divided by this maximum value to obtain a scaling factor between 0 and 1; finally, this scaling factor is incremented by 1. The system obtains weighting coefficients between 1 and 2, and then multiplies the landing point distribution density of each grid with the corresponding weighting coefficient to obtain the weighted density value of that grid. Grids with larger deviations have their weighted densities amplified more, thus receiving more attention in subsequent non-uniformity calculations. The system treats the weighted density values ​​of all grids as a whole as a distribution density feature and plots a density distribution map on a plane based on the spatial location of each grid. Each grid in the density distribution map corresponds to a specific weighted density value, represented by different color shades or numerical labels, thus visually displaying the seed distribution under the combined effect of the number of landing points and deviations in space. Next, the system calculates the non-uniformity index based on the density distribution map and the granularity of the region division. The granularity of the region division is a comprehensive parameter of grid size and number of grids, determined during the system design phase based on the seeding accuracy requirements and the UAV's flight altitude. In this embodiment, the target area is divided into square grids with sides of 10 meters in the horizontal direction, completely covering the entire area. The smaller the granularity, the finer the grid, and the higher the spatial resolution of the density distribution map.

[0027] The calculation process for the non-uniformity index is as follows: First, calculate the average density of all grid weighted density values. Sum the weighted density values ​​of all grids and divide by the total number of grids to obtain a definite average density value. Then, for each grid, calculate the absolute value of the difference between the weighted density value of that grid and the average density to obtain the density deviation of that grid. Next, sum the density deviations of all grids to obtain the total density deviation. Divide the total density deviation by the product of the total number of grids and the average density to obtain a dimensionless non-uniformity index. The larger the non-uniformity index value, the greater the deviation of the weighted density of each grid from the average density, and the more uneven the seed landing point distribution. This non-uniformity index is the judgment index for judging the non-uniformity of seed landing point distribution. In each task, the system calculates a unique non-uniformity judgment index in the above manner for comparison with a preset threshold standard.

[0028] The process for determining the preset threshold standard is as follows: Before the system is put into actual use, select several historical sowing tasks under different airflow conditions and terrain conditions. For each task, calculate the non-uniformity judgment index using the above method. At the same time, measure the pasture coverage effect of each task through field investigation. Manually or statistically, divide these tasks into two categories: those that meet the sowing uniformity requirements and those that do not. Select the maximum value of the non-uniformity judgment index from all tasks identified as meeting the sowing uniformity requirements. Add a fixed safety margin to this maximum value. In this embodiment, the safety margin is set to 0.05. The result of the sum is determined as the preset threshold standard and is fixed in the system parameters so that it does not change automatically with the task in subsequent tasks.

[0029] In actual operation, when the non-uniformity judgment index calculated by the current task is less than or equal to the preset threshold standard, the system determines that the uniformity of the seed landing point distribution meets the requirements and there is no need to make significant adjustments to the initial drift compensation vector; when the non-uniformity judgment index is greater than the preset threshold standard, the system determines that the non-uniformity of the seed landing point distribution exceeds the allowable range and it is necessary to correct the wind direction variation and adjust the compensation vector according to the density distribution map. To address this issue, the system first uses a density distribution map to identify areas with significantly overly dense and underly sparse landing points. Specifically, after calculating the average density of all grid weighted density values, grids with weighted densities higher than the average density are sorted from largest to smallest, and the top 20% are selected as overly dense areas. Similarly, grids with weighted densities lower than the average density are sorted from smallest to largest, and the top 20% are selected as underly sparse areas. Then, the spatial locations of these overly dense and underly sparse areas are compared with the wind direction obtained from previous wind field simulations. When overly dense areas are generally biased downstream of the prevailing wind direction, while underly sparse areas are biased upstream, it indicates that the initial drift compensation vector is insufficient in the prevailing wind direction. The system then adjusts the compensation distance in the prevailing wind direction upwards by a certain proportion, based on the non-uniformity judgment index exceeding the pre-defined... The threshold standard is set to a certain range. When the exceedance is small, the compensation distance adjustment ratio is set to 0.1; when the exceedance is moderate, it is set to 0.2; and when the exceedance is large, it is set to 0.3. When the overly dense area is biased upstream and the overly sparse area is biased downstream, it indicates that the current compensation distance is too large and the compensation distance needs to be adjusted downward in the same way. If the density distribution shows obvious asymmetry in both the east-west and north-south directions, the above adjustment ratios are calculated independently along the east-west and north-south directions respectively, and the two components of the initial drift compensation vector are corrected respectively. When correcting for wind direction variation, wind direction variation refers to the deflection of the main wind direction at different time periods or different spatial locations. The system corrects the reference direction of the wind direction in the seeding control by comparing the positional relationship between the overly dense and overly sparse areas and the main wind direction, so that the direction of the compensation vector is more consistent with the actual effective wind direction. After completing the above corrections, the system recombines the corrected east-west and north-south compensation distances to form a new initial drift compensation vector for seed fall. This vector replaces the original compensation vector for subsequent sowing or the next round of simulation, so that the unevenness of seed fall distribution is gradually reduced and controlled within the preset threshold standard, taking into account actual fall point deviations and wind direction variations, thereby ensuring that forage seeds are sown as evenly as possible in the target area.

[0030] S4 includes obtaining wind speed time series information and terrain airflow disturbance parameters by identifying uneven distribution of landing points exceeding a preset threshold, and constructing a state vector dimension; fusing historical wind field data from the state vector dimension to determine the spatial distribution characteristics of the disturbance; combining the spatial distribution characteristics of the disturbance with wind direction variation correction to obtain airflow disturbance simulation results; using the airflow disturbance simulation results to integrate the time series to determine the current wind field state; and obtaining the airflow simulation state description under the current wind field state.

[0031] In this embodiment, the state vector dimension is first constructed using the wind speed time series information and terrain airflow disturbance parameters recorded in the current task. The wind speed time series information consists of wind speed data continuously collected at fixed time intervals during UAV operation. In this embodiment, the time interval is set to 1 second during the task configuration phase based on the UAV's flight speed and the desired time resolution, and the collection duration is set to 300 seconds, meaning 300 wind speed samples are continuously recorded at each location. The terrain airflow disturbance parameters are pre-determined based on the terrain conditions of the target area before the task begins, dividing the area into three categories: flat terrain, gently undulating terrain, and complex undulating terrain, with corresponding terrain disturbance levels set to 1, 2, and 3, respectively. In actual operation, the system binds a fixed terrain disturbance level to each grid location and calculates the disturbance intensity based on the difference between the instantaneous wind speed in the wind speed time series and the average wind speed at the same height and within the same time window. The disturbance intensity is then combined with the terrain disturbance level to form the terrain airflow disturbance parameters for that location. When constructing the state vector dimension, the system summarizes the above information for each spatial grid location within a time window, and organizes multiple parameters describing the current wind field state into multidimensional data in a fixed order. In this embodiment, the state vector dimension contains 8 parameters: the first parameter is the average wind speed in the last 60 seconds; the second parameter is the maximum change in wind speed in the last 60 seconds; the third parameter is the deflection angle of the wind direction at the current moment relative to the wind direction at the previous moment; the fourth parameter is the turbulence intensity level calculated from wind speed fluctuations, with turbulence intensity divided into three levels from smallest to largest: 1, 2, and 3; the fifth parameter is the aforementioned terrain disturbance level; the sixth parameter is the height layer number of the location, which is sequentially numbered in 5-meter intervals starting from the lowest safe height during the task configuration phase; the seventh parameter is the long-term average wind speed value of the location at the same height layer extracted from historical wind field data; and the eighth parameter is the long-term wind speed fluctuation amplitude of the location at the same height layer extracted from historical wind field data.

[0032] Historical wind field data consists of wind speed time series and terrain disturbance records collected and saved in the region during multiple seeding and test flight missions. In this embodiment, data from the most recent 20 missions are selected as the source of historical wind field data, and the data is classified and stored according to grid location, altitude layer, and seasonal characteristics. When constructing the state vector, the record that is closest to the current conditions is selected based on the geographical location, flight altitude, and operation time of the current mission, and used as the source of the 7th and 8th parameters. After completing the state vector dimension construction, the system integrates historical wind field data to determine the spatial distribution characteristics of disturbances. The specific steps are as follows: For each grid location, the average wind speed measured in the current task is compared with the historical long-term average wind speed, and the two are weighted according to a fixed weight. In this embodiment, the weight of the current task is set to 0.7, and the weight of historical data is set to 0.3. The comprehensive wind speed reference value of the location is obtained by weighting. Then, the current turbulence intensity level and the historical wind speed fluctuation amplitude are jointly analyzed. When the current turbulence level is higher than 2 and the historical fluctuation amplitude is large, the disturbance level of the location is marked as high disturbance. When the current turbulence level is 2 or the historical fluctuation amplitude is in the medium range, it is marked as medium disturbance. The rest are marked as low disturbance. These disturbance levels are combined in the entire spatial grid to form the spatial distribution characteristics of disturbances in the plane and height directions. Based on the spatial distribution characteristics of the disturbance, the system combines the wind direction variation correction results obtained in the previous stage according to the density of the landing points and the wind direction variation to adjust the disturbance direction. The wind direction variation correction results give the deflection direction and deflection angle of the current effective wind direction relative to the original main wind direction. At each grid position, the system corrects the original wind direction value in the spatial distribution characteristics of the disturbance according to the deflection angle, so that the disturbance direction is consistent with the actual effective wind direction. At the same time, it checks whether the high disturbance area is concentrated on the downwind or upwind side after deflection. If a large number of high disturbance areas are concentrated on the downwind side, the disturbance intensity is appropriately amplified in that direction. If they are concentrated on the upwind side, the disturbance intensity is slightly reduced. In this way, the airflow disturbance simulation results after wind direction variation correction are obtained.Subsequently, the system integrates the airflow disturbance simulation results with a time series to construct the basis for determining the current wind field state. Specifically, at each grid location, the measured wind speed time series within the most recent 300 seconds is spliced ​​with the wind speed change trend within the next 60 seconds calculated based on the disturbance simulation results, forming a continuous wind speed time series of 360 seconds. On this time series, the overall average wind speed, the overall wind speed standard deviation, and the rate of change of the average wind speed within several consecutive time periods are calculated. In this embodiment, the 360 ​​seconds are divided into 6 time periods, each 60 seconds long. The average wind speed is calculated for each time period, and the difference between the average wind speeds of adjacent time periods is compared. When the average wind speed difference between all time periods is less than a preset stability threshold (set to 0.5 m / s in this embodiment) and the overall wind speed standard deviation is less than 1 m / s, the system determines the current wind field state as basically stable. When the average wind speed difference between adjacent time periods is partially between 0.5 m / s and 1.5 m / s, or the overall wind speed standard deviation is between 1 m / s and 2 m / s, it is determined to be a slowly changing state. When the average wind speed difference between adjacent time periods exceeds 1.5 m / s, or the overall wind speed standard deviation exceeds 2 m / s, and the high disturbance area is clearly distributed in a spatially concentrated manner, it is determined to be a violently disturbed state. Based on the above determination, the system obtains a clear wind field state classification at each grid location and altitude layer. It then summarizes the location of all grids, the comprehensive wind speed reference value, the disturbance level, the corrected effective wind direction, the wind field state category, and the predicted wind speed change trend in the next 60 seconds in a unified format to form a simulated airflow state description under the current wind field state. This description includes both a numerical description at the grid level and spatial distribution information showing the boundaries of different wind field state areas within the entire target area. This provides a direct and quantifiable wind field basis for subsequent seed release timing adjustment, release angle correction, and dynamic flight path planning.

[0033] S5 includes updating the wind field state vector through real-time observation input, combining the state update cycle with a noise interference separation model. The noise interference separation model separates interference by comparing observed data with historical noise patterns, and obtains interference noise data from the updated vector to obtain the noise separation result. A deviation correction process is then fused to the noise separation result. This process uses the difference between the observed input and the vector deviation to calculate and correct the system deviation during wind speed changes, obtaining the corrected wind speed data. The corrected wind speed data is combined with real-time data processing to determine the terrain impact assessment under the interference filtering mechanism. The terrain impact assessment is obtained by comparing the data with terrain parameters, resulting in a terrain assessment index. The terrain assessment index is then used to integrate system deviation adjustments. This adjustment is based on a weighted fusion of the index and the deviation, used to optimize the state vector and determine the optimized vector features. Finally, the optimized vector features are fused to produce a short-term prediction output. This short-term prediction output is generated by matching features with a trend sequence, resulting in a short-term wind field change trend prediction result.

[0034] In this embodiment, after completing the airflow simulation state description, the system updates the wind field state vector in real time during the actual operation of the UAV at a fixed state update cycle. The state update cycle is determined during the task configuration stage based on the UAV's flight speed and the typical time scale of wind field changes. In this embodiment, the state update cycle is set to 5 seconds, that is, every 5 seconds, the system updates the wind field state vector at each grid position using the latest collected wind speed and wind direction observation data. The real-time observation input includes the instantaneous wind speed value, instantaneous wind direction angle value, and corresponding sampling time marker at each grid position at the current moment. The system matches these observation data with the wind field state vector saved during the last update one by one, replaces the corresponding short-term statistics in the vector with the new wind speed and wind direction information, and retains the historical average value and historical fluctuation, thereby forming a temporary state vector before the update. Subsequently, the system processes the temporary state vector using a pre-defined noise interference separation model. This model separates interference signals by comparing observed data with historical noise patterns. The historical noise patterns are statistical results obtained before system deployment through numerous static tests and flight tests under windless or weak wind conditions. Specifically, under conditions without significant environmental wind field influence, wind speed readings from multiple sensors at different altitudes and attitudes are recorded. The mean and fluctuation range of these readings are calculated based on location and altitude to obtain the measurement noise distribution range of the sensor under ideal still air conditions. During actual operation, the model compares the real-time observed wind speed at each grid location with the aforementioned noise range. When a sampled value falls within the noise range or is slightly above the upper limit of the noise range for a very short duration, that part is considered a noise component. When the sampled value is significantly higher than the upper limit of the noise range and maintains a consistent trend over multiple consecutive update cycles, that part is considered a valid wind field signal. The system decomposes the observed wind speed at each grid location into an effective wind speed component and a noise interference component. The noise interference component is then accumulated over time to form a noise interference data set for each location within the current update cycle. This set is the noise separation result, which includes the magnitude of the noise component at each grid location and the frequency of noise occurrence, used in the subsequent bias correction process. Next, the system integrates the noise separation result with the bias correction process to correct system biases during wind speed changes. The bias correction process uses the difference between the observed input and the vector bias, where the vector bias refers to the difference between the predicted wind speed recorded in the state vector before the update and the effective observed wind speed within the current cycle.

[0035] The specific procedure is as follows: First, noise components are removed from the noise separation results, retaining only the effective observed wind speeds. Then, for each grid location, the effective observed wind speed in the current update cycle is compared with the predicted wind speed stored in the state vector of the previous cycle. The difference between the two is calculated and defined as the instantaneous deviation for that location. Then, within a fixed time window, such as the last five update cycles, all instantaneous deviations are averaged to obtain the system deviation estimate for that location. The system deviation estimate reflects the systematic overestimation or underestimation of wind speed predictions for that location over a continuous period. To correct this deviation, the system subtracts the corresponding system deviation estimate from the predicted wind speed for each location to obtain corrected wind speed data. This corrected wind speed data is then written into the state vector to replace the original predicted wind speed components, while keeping other parameters describing terrain and historical features unchanged. This completes the correction of system deviations during wind speed changes. Subsequently, the system combines the corrected wind speed data with real-time data processing to determine the terrain impact assessment under the interference filtering mechanism. The interference filtering mechanism refers to removing noise and system biases, retaining only those wind speed variation components that are stably correlated with terrain features for analysis. Specifically, for each grid location, the corrected wind speed data for that location is correlated and compared with its corresponding terrain parameters. Terrain parameters include the terrain type level, slope, and relative altitude of the location. This information is acquired and stored in a fixed location through terrain mapping before the task begins. Within a fixed time window, such as the last 60 seconds, the system statistically analyzes the average corrected wind speed and the range of corrected wind speed fluctuations for all grid locations under each terrain type level. These statistical results are compared with typical wind speed statistics for the same terrain type under similar climatic conditions in the past. When the difference between the current statistical result and the historical typical result is within a preset tolerance range, it indicates that the wind speed change at that location is mainly caused by the terrain, and the terrain influence is relatively stable. When the difference is significantly larger or smaller and persists for a long period, it indicates that other factors besides terrain are also influencing the wind speed. The system uses this comparison method to calculate a terrain impact assessment index for each terrain type. This index is a value between 0 and 1. A value close to 1 indicates that the terrain type has a significant impact on the current wind speed distribution and is consistent with historical characteristics, while a value close to 0 indicates that the terrain type has a weak impact under current conditions or is masked by other factors. After completing the terrain impact assessment, the system uses the terrain assessment index to integrate system bias adjustments and optimize the state vector.

[0036] Specifically, for each grid location, its system bias estimate is weighted and fused with the terrain assessment index of its corresponding terrain type. The weighting method is predetermined during the system configuration phase. In this embodiment, the weight of the terrain assessment index is set to 0.6, and the weight of the system bias estimate is set to 0.4. When the terrain assessment index is high, the system considers the long-term impact of terrain on the wind field more when adjusting the state vector, reducing the effect of random bias. When the terrain assessment index is low, it relies more on the system bias estimate obtained from recent observations for adjustment. The weighted fusion result is used to jointly optimize multiple components in the state vector, including fine-tuning the predicted wind speed, predicted wind direction, and disturbance level. The optimized state vector features are mainly reflected in the predicted wind speed being closer to the observed wind speed after bias correction, the predicted wind direction being more consistent with the terrain guidance direction, and the overlap between high-disturbance areas and typical complex terrain being more reasonable. Finally, the system obtains the optimized vector feature fusion short-term prediction output. The short-term prediction output is generated by matching features with a trend sequence. The trend sequence is a sequence of wind speed changes over a recent period, after noise removal and bias correction, arranged in chronological order. In this embodiment, the length of the trend sequence is set to 120 seconds. The system pre-extracts several typical wind speed change patterns from historical wind field data, such as a continuous strengthening pattern, a continuous weakening pattern, a periodic fluctuation pattern, and a sudden fluctuation pattern, and establishes a corresponding reference change sequence for each pattern. During actual operation, the current trend sequence is aligned with these reference change sequences one by one, and the changes within the same time period are compared. The system calculates the degree of matching between a typical pattern and the current trend sequence. When the degree of matching between a typical pattern and the current trend sequence is the highest and exceeds the set matching threshold, the system uses the wind speed change pattern corresponding to the typical pattern for the future time period as the short-term prediction output for the current grid location. The prediction amplitude is amplified or reduced according to the disturbance level and terrain assessment index contained in the optimized state vector. Finally, the short-term wind field change trend prediction result is obtained in the future for several update cycles. The prediction result is output in the form of the predicted average wind speed, predicted wind direction change and disturbance change level for each grid location in the future, for example, 60 seconds. This provides a reliable basis for the dynamic adjustment of the subsequent seeding timing and angle.

[0037] S6 includes obtaining wind field trend predictions through observation input fusion, determining the initial release timing adjustment based on the wind field trend predictions and terrain factor assessments, and obtaining integrated wind speed variation results; performing noise separation processing and fusion bias correction on the integrated wind speed variation results, dynamically adjusting the release angle optimization, and obtaining seed density control data; monitoring the landing point distribution in real time based on the seed density control data and data frequency updates, and generating integrated prediction outputs if the landing point distribution exceeds a preset threshold, and judging the distribution deviation value; obtaining the distribution deviation value and dynamically adjusting the path through trajectory parameter correction processing, fusing observation data offset compensation, and determining the correction vector group; extracting real-time feedback from the observation input fusion based on the correction vector group to obtain dynamic path adjustment parameters for correcting the flight trajectory.

[0038] In this implementation, after completing the short-term wind field trend prediction, the system first obtains the wind field trend prediction by fusing observation inputs in each state update cycle. The observation inputs include the instantaneous wind speed and instantaneous wind direction at each grid location in the current cycle, as well as the predicted wind speed and predicted wind direction for that location over a future period given in the short-term prediction output of the previous cycle. The fusion process is as follows: for each grid location, the wind speed actually measured in the current cycle is compared with the wind speed predicted at the same time point in the previous cycle, the difference between the two is calculated, and then the difference is averaged over the most recent several cycles to obtain a smoother correction amount. This correction amount is then superimposed on the original predicted wind speed to obtain the updated wind field trend prediction. At the same time, according to the flight route pre-planned by the UAV, the system arranges the grids along the flight route in the target area in chronological order. For each grid to be seeded, the initial seeding timing is adjusted based on the wind field trend prediction and the corresponding terrain factor assessment. The terrain factor assessment is performed before the mission based on slope, Relative height and surface roughness are divided into several levels and assigned specific numerical levels. In this embodiment, the terrain factor level is divided into three levels: flat area, general undulating area, and complex undulating area, with values ​​set to 1, 2, and 3 respectively. When adjusting the seeding timing, if the wind trend prediction shows that the wind speed at the location will increase and the terrain factor is large in the next few seconds, the system will appropriately advance the seeding timing relative to the original planned time. If the wind speed decreases or the terrain factor is small, it will be appropriately delayed. The seeding timing adjustment amount is calibrated to a fixed ratio based on the seeding height and seed fall time during the task configuration stage. In this embodiment, the advance seeding time corresponding to every 1 meter per second increase in wind speed is set to 0.5 seconds, and the delay seeding time corresponding to every 1 meter per second decrease in wind speed is set to 0.5 seconds. By performing the above correction on the seeding timing of all the grids to be seeded, the system obtains the wind speed variation integration result over time. This result records the adjusted seeding time and the corresponding predicted wind speed change for each seeding location. In response to the wind speed variation integration results, the system employs noise separation processing and fusion deviation correction in each state update cycle to dynamically adjust the throwing angle optimization. The noise separation processing follows a pre-established noise pattern, considering wind speed mutations with a duration of less than one cycle and an amplitude below a set threshold as noise. In this embodiment, mutations with a duration of less than 5 seconds and a wind speed change of less than 0.5 meters per second are considered noise and are removed from the wind speed variation integration results, retaining only effective variations with longer durations and larger amplitudes.

[0039] Deviation correction is performed using the difference between the observed input and the integrated result. Specifically, during the actual seeding process, the seed landing point offset obtained through trial seeding is compared with the theoretical offset calculated based on the integrated result of wind speed variation. The difference is calculated and averaged over several recent seeding events to form a correction amount in the seeding angle direction. This correction amount is then superimposed on the currently set seeding angle to obtain the dynamically adjusted seeding angle. In this embodiment, the seeding angle adjustment range is limited to the safe working angle of the UAV nozzle, for example, the vertical downward deflection angle is within ±20 degrees. Within this range, the system makes fine adjustments according to the above correction amount. By jointly optimizing the seeding angle and seeding timing of all seeding positions, the system calculates the relationship between the theoretical number of seeds that should fall in each grid and the target seeding density. The grid target seeding density, the adjusted seeding timing, the seeding angle, and the seed flow rate per unit time are combined to form seed density control data. This data clearly indicates the number of seeds to be deployed and the corresponding spatial location on each flight path in each update cycle. The system monitors the distribution of landing points in real time by combining seed density control data with data frequency updates. Data frequency updates refer to the system refreshing the landing point statistics once at a fixed time interval. In this embodiment, the time interval is set to 10 seconds. At the end of each time interval, the system counts the actual number of landing points in each grid cell based on the airborne or ground acquisition device, and compares the actual landing point density with the target density recorded in the seed density control data. When the deviation between the actual density and the target density of a certain grid exceeds a preset threshold, it is marked as an abnormal grid. This preset threshold is selected from historical tasks during the task configuration stage based on the requirements for uniformity of pasture sowing.

[0040] In this embodiment, a threshold is set when the difference between the actual density and the target density of a single grid exceeds 20% of the target density. When the proportion of the number of grids marked as abnormal in a monitoring period exceeds another preset proportion threshold, such as exceeding 30%, the system determines that the current distribution of landing points exceeds the preset threshold and dynamic path adjustment is required. In this case, the system integrates the difference between the actual density and the target density of all abnormal grids in the current period with the short-term wind field prediction output of the corresponding location to calculate the distribution deviation value of each abnormal area in the main wind direction and the direction perpendicular to the main wind direction. The distribution deviation value is the degree to which the average density of the current area deviates from the target density and whether the deviation is mainly concentrated on the downwind side or the upwind side, which is used to indicate the direction and degree of seeding offset. After acquiring the distribution deviation value, the system dynamically adjusts the path through trajectory parameter correction. The trajectory parameters include the UAV's heading angle, lateral offset, and flight speed over the next time period. Without changing the overall operational safety boundary, when the distribution deviation value indicates that the seed landing points are generally biased towards the leeward side, the system shifts the centerline of the next trajectory towards the leeward side by a certain distance. This distance is determined during mission calibration according to the correspondence between the distribution deviation value and the seeding height and average wind speed. In this embodiment, the lateral offset is set to 5 meters when the distribution deviation value reaches 20% of the target density, and 10 meters when the distribution deviation value reaches 40%. At the same time, the system appropriately adjusts the flight speed according to the relationship between the distribution deviation value and the current flight speed, appropriately increasing the speed over areas with excessively dense landing points and appropriately decreasing the speed over areas with excessively sparse landing points. Through the above methods, a set of trajectory correction parameters is formed to jointly correct the heading, offset, and speed.

[0041] To further reduce the impact of real-time observation errors, the system also integrates observation data offset compensation when generating trajectory correction parameters. The landing point offset trend observed in the most recent monitoring cycles is superimposed on the trajectory correction parameters. For example, when the landing point is continuously biased to one side for several consecutive cycles, the system adds a fixed additional offset to the correction vector to compensate for the residual deviation of the sensor or model that has not yet been calibrated. All these correction information with direction and magnitude are combined to form a correction vector group, where each vector corresponds to a small segment of the track interval to be executed. The direction of the vector indicates the direction of track adjustment, and the length of the vector indicates the magnitude of the adjustment to be implemented. Finally, the system extracts real-time feedback from the fusion of observation inputs based on the correction vector group. The real-time feedback includes the execution status of the corrected flight path, the latest wind speed and direction observations, and the latest landing point statistics in the corresponding area. The system compares the predicted landing point distribution with the actual landing point distribution before and after executing each correction vector. When the correction effect is good, the current correction strategy is maintained; when the correction effect is insufficient, the correction vector group is fine-tuned in the next cycle. After several consecutive iterations, the system converges to obtain dynamic path adjustment parameters for correcting the flight path. These parameters specifically include the heading angle sequence that the UAV should execute in the next few time periods, the lateral offset distance of each flight path segment, and the corresponding flight speed. These parameters are input into the UAV flight control device in a definite numerical form, enabling the UAV to continuously adjust its flight path in a short period of time, thereby keeping the seed landing point distribution as close as possible to the target uniform seeding requirement under actual wind field changes.

[0042] S7 includes obtaining flight altitude updates from path adjustment parameters, integrating wind field change trends to determine flight speed optimization, and obtaining landing point distribution density; using threshold judgment results for landing point distribution density to determine the seeding trajectory deviation, and if the deviation exceeds a preset threshold, integrating terrain interference factors to obtain the seeding timing; extracting soil adaptation adjustments by combining seeding timing with dynamic monitoring of landing points to determine uniform sowing requirements; generating seeding trajectory deviation compensation by integrating threshold judgment results from flight speed optimization based on uniform sowing requirements, and obtaining the final path planning; integrating landing point distribution density for soil adaptation adjustments during the final path planning, judging dynamic monitoring of landing points, and obtaining a path output that meets the uniform sowing requirements.

[0043] In this embodiment, after obtaining the dynamic path adjustment parameters for correcting the flight trajectory through the aforementioned steps, the system reads the adjustment amount for the altitude of each segment of the flight path from the path adjustment parameters, adds the original planned flight altitude to the adjustment amount to obtain the updated flight altitude, where each segment of the flight path corresponds to a specific altitude value. In this embodiment, the range of the altitude adjustment amount is limited to within 20 meters above and below the original planned altitude to ensure flight safety. While updating the flight altitude, the system integrates the updated altitude with the short-term wind field change trend prediction results to optimize the flight speed. Specifically, a basic flight speed curve is set during the task configuration phase to meet the requirements of seeding time and operational efficiency.

[0044] In this embodiment, the base flight speed is set to 5 meters per second. Then, for each track grid, based on the average wind speed and direction at that location within the prediction time window, if the predicted wind speed component along the flight direction is large, the flight speed is appropriately increased; if the predicted wind speed is opposite to or perpendicular to the flight direction, the flight speed is decreased. The adjustment range is obtained by multiplying the predicted wind speed component along the flight direction by a preset influence coefficient. In this embodiment, the influence coefficient is set to 0.2, meaning that for every 1 meter per second increase in the predicted wind speed component along the flight direction, the corresponding speed adjustment is 1 meter per second. Finally, the base speed is added to the adjustment amount to obtain the optimized flight speed at that location. After updating the flight altitude and optimized flight speed, the system calculates the theoretical number of seeds that each ground grid should cover based on the seed dispersal rate per unit time on each track segment, the UAV's flight speed on that segment, and the track length. Then, combined with the actual number of landing points monitored, the actual number of seeds in each grid is divided by the grid area to obtain the landing point distribution density, and a landing point distribution density map covering the entire target area is formed.

[0045] Subsequently, the system uses a threshold judgment result to determine the deviation of the sowing trajectory based on the density of the landing points. During the task configuration phase, a target sowing density and allowable deviation range are pre-set according to the requirements of pasture growth. In this embodiment, the target density is set to 20 seeds per square meter, and the allowable deviation range is set to ±20%. That is, when the actual density of the landing points in a grid is between 16 and 24 seeds per square meter, it is considered acceptable; otherwise, it is considered a deviation from the target. The system compares all grids one by one, marking grids with a density of less than 16 seeds per square meter as under-sowing areas and grids with a density of more than 24 seeds per square meter as over-sowing areas. The degree of under-sowing or over-sowing is calculated, and the difference between the actual density and the target density is divided by the target density to obtain the relative deviation ratio of each abnormal grid. This ratio is considered the degree of sowing trajectory deviation. When the proportion of abnormal grids to the total number of grids exceeds a preset threshold during a certain monitoring period (in this embodiment, the preset threshold is set to 0.3), an abnormal grid is considered an abnormal grid. When the grid ratio exceeds 30%, the system determines that the overall deviation of the seeding trajectory exceeds a preset threshold. At this time, the system integrates the aforementioned terrain interference factors and recalculates the seeding timing. The terrain interference factors have been divided into multiple levels based on terrain undulation and wind tunnel effects before the mission begins. In this embodiment, three levels are set and assigned values ​​of 1, 2, and 3, respectively. The larger the value, the stronger the terrain disturbance to the airflow. When adjusting the seeding timing, if a certain area is determined to be under-seeded and the terrain interference factor in that area is high, the system advances the seeding time over that area relative to the time when the UAV flies to the center of the grid by a certain amount. The advance amount is obtained by multiplying the relative deviation ratio of the area by the terrain interference factor and the seed's fall time from the airborne seeding device to the ground. In this embodiment, the fall time is fixed at 2 seconds through experiments during mission calibration. If the area is over-seeded and the terrain interference factor is high, the same calculation result is delayed relative to the grid center time, thereby obtaining a clear seeding timing for each key area.Next, the system extracts soil adaptation adjustments by combining seed sowing timing with dynamic monitoring of seed landing points. Dynamic monitoring involves periodically counting the number of newly added seeds in each grid during the operation; in this embodiment, the monitoring cycle is set to once every 30 seconds. Soil adaptation adjustments are determined before and during the task, taking into account soil texture, moisture content, and fertility level. Before the task, sampling and testing divide the target area into several soil types, assigning a soil adaptation coefficient to each type. In this embodiment, soil suitable for lower sowing densities is assigned a value of 0.8, moderate soil a value of 1, and infertile or poorly water-retaining soil a value of 1.2, indicating a relatively higher sowing density is required. The system multiplies the target sowing density by the soil adaptation coefficient of the corresponding grid to obtain the target soil adaptation density for that grid, and compares it with the actual density obtained from the dynamic monitoring landing points. When the deviation is within the allowable range of the target soil adaptation density, it is considered acceptable. To meet the uniform seeding requirements under local soil conditions, if the deviation exceeds the allowable range, the grid is considered not to meet the uniform seeding requirements. Based on this, the system statistically analyzes the proportion of grids that meet the requirements and the types of grids that do not meet the requirements across the entire area, obtaining the evaluation result of the uniform seeding requirements for the current operation stage. According to this evaluation result, the system integrates the threshold judgment result from the flight speed optimization data to generate seeding trajectory offset compensation. Specifically, for areas with continuous under-seeding, the system further reduces the flight speed and slightly increases the seeding flow rate over these areas. For areas with continuous over-seeding, the opposite is true. The system compensates for the impact of trajectory offset by jointly adjusting the flight speed and seeding flow rate. At the same time, the compensation amount is superimposed with the aforementioned seeding timing adjustment amount to form a comprehensive correction of the seeding trajectory in time and space, thereby obtaining the final path planning that includes the altitude, speed, seeding time, and seeding flow rate of each segment of the trajectory. Finally, during the final path planning, the system re-integrates the landing point distribution density based on soil adaptation adjustments and determines the dynamic monitoring landing points. Specifically, after each monitoring cycle, the deviation between the actual density of each grid and its soil adaptation target density is recalculated. When the proportion of all grids with deviations within the allowable range reaches or exceeds a preset coverage threshold (in this embodiment, the coverage threshold is set to 0.9, meaning 90% of the grids meet the uniform spreading requirements), the system determines the current path planning track sequence as a path output that meets the uniform spreading requirements and registers this path output as the recommended operational path for this region under the current season and wind field conditions. If, within several cycles, there are still areas of similar type that fail to meet the requirements, the system iteratively fine-tunes the local paths in these areas without changing the overall operational boundary and repeats the above evaluation process until the landing point distribution density meets the uniform spreading requirements under soil adaptation adjustment constraints, thereby forming a stable and reliable final path output.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the seed dispersal path of forage grass based on airflow simulation data, characterized in that, include: S1. Use multi-point wind field sensors mounted on the UAV to collect wind speed time series and terrain airflow disturbance data at the current flight altitude, and combine them with terrain interference factors to analyze the local wind field characteristics. Based on the collected data, construct an airflow simulation model of the target area to obtain a dynamic wind field distribution map of the target area. S2. Simulate the trajectory of pasture seed scattering based on the wind field dynamic distribution map, calculate the trajectory offset caused by wind field disturbance, and combine the wind field influence weight and trajectory simulation error to determine the initial drift compensation vector for seed fall. S3. The residual calculation method is used to evaluate the accuracy of the deviation between the initial drift compensation vector and the actual landing point. Combined with the landing point distribution density and the granularity of the region division, it is determined whether the unevenness of the seed landing point distribution exceeds the preset threshold standard. S4. When the unevenness of the landing point distribution exceeds the preset threshold standard, construct a state vector dimension that includes wind speed time series information and terrain airflow disturbance parameters, integrate historical wind field data and disturbance spatial distribution characteristics, and obtain the airflow simulation state description of the current wind field. S5. The state vector is updated based on real-time observation input, and the system deviation in the wind speed change process is corrected by combining the state update cycle and noise interference separation model to obtain the short-term wind field change trend prediction results. S6. Based on the short-term wind field change trend and terrain interference factor, dynamically adjust the seed scattering timing and scattering angle, and combine the data update frequency to monitor the landing point distribution in real time, and determine the dynamic path adjustment parameters used to correct the flight trajectory. S7. Based on the dynamic path adjustment parameters, update the UAV's flight altitude and speed, integrate the landing point distribution density and threshold judgment results, and generate the final path plan that meets the uniform dispersal requirements based on continuous monitoring of the dispersal trajectory deviation.

2. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 1, characterized in that: S1 includes: The data collection results are obtained by acquiring time series of wind speed and terrain airflow disturbance data at the current flight altitude through multi-point sensors carried by the drone; Based on the data collection results and the terrain interference factor, the local wind field characteristics are analyzed for interference to determine the changing trend of local characteristics. A simulation model of airflow in the target area is constructed by combining the changing trends in local characteristics with flight altitude data. The simulation results are obtained by inputting the changing trends and altitude parameters. The simulation output generates a dynamic wind field distribution map of the target area, and anomalies in the dynamic distribution are identified. If the number of outliers in the dynamic distribution exceeds the preset threshold, the outliers are marked to obtain the distribution characteristics of the wind field anomaly detection.

3. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 1, characterized in that: S2 includes: The offset results are fused with the influence weight values ​​to adjust the local wind characteristics and determine the disturbance distribution in the local wind characteristics; The simulation error term is constructed based on the perturbation distribution and height data parameters. The error is corrected by inputting the perturbation distribution and height parameters, and the output of the error term is obtained. Anomaly markers are generated by the output of the error term. If the number of anomaly markers exceeds a preset threshold, the distribution feature map is marked to obtain the feature values ​​of the distribution feature map. The initial drift compensation vector for seed fall is determined by fusing the eigenvalues ​​of the distribution feature map with the wind field disturbance factor.

4. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 1, characterized in that: S3 includes: The residual calculation method is used to obtain the accuracy evaluation result by subtracting the actual landing point deviation from the initial drift compensation vector, and the deviation value in the accuracy evaluation result is obtained. By weighting and fusing the landing point distribution density using deviation values, the distribution density characteristics are determined, and the density distribution map in the distribution density characteristics is obtained; Based on the density distribution map and the regional granularity, the unevenness index is calculated to determine the unevenness of seed landing point distribution, and the unevenness judgment index is obtained. If the judgment index exceeds the preset threshold standard, the wind direction variation is corrected and the compensation vector is adjusted according to the density distribution map to determine the initial drift compensation vector for seed fall.

5. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 1, characterized in that: S4 includes: By obtaining wind speed time series information and terrain airflow disturbance parameters by the uneven distribution of landing points exceeding a preset threshold, a state vector dimension is constructed. By fusing historical wind field data from the state vector dimension, the spatial distribution characteristics of the disturbance can be determined. By combining the spatial distribution characteristics of the disturbance with wind direction variation correction, the simulation results of airflow disturbance are obtained; The current wind field status is determined by integrating time series data with airflow disturbance simulation results. The simulation state description of the airflow under the current wind field condition is obtained.

6. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 1, characterized in that: S5 includes: The wind field state vector is updated by real-time observation input. Combined with the state update period and the noise interference separation model, the noise interference separation model separates the interference by comparing the observed data with historical noise patterns. The noise interference data is obtained from the updated vector to obtain the noise separation result. For the noise separation result fusion deviation correction process, the deviation correction process uses the difference between the observed input and the vector deviation to calculate the correction, correct the system deviation in the wind speed change process, and obtain the corrected wind speed data; By combining the corrected wind speed data with real-time data processing, the terrain impact assessment under the interference filtering mechanism is determined. The terrain impact assessment is obtained by comparing the data with the terrain parameters, and the terrain assessment index is obtained. The terrain assessment index is used to integrate system bias adjustment. The system bias adjustment is based on the weighted fusion of index and bias to process the state vector optimization and judge the characteristics of the optimized vector. The optimized vector feature fusion short-term prediction output is obtained. The short-term prediction output is generated by matching features with trend sequences to obtain the short-term wind field change trend prediction result.

7. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 1, characterized in that, S6 specifically includes: Wind field trend prediction is obtained by fusing observation inputs. The initial release timing is adjusted by combining the wind field trend prediction with topographic factor assessment, and the integrated result of wind speed variation is obtained. Noise separation and fusion bias correction were applied to the integrated results of wind speed variation, and the seeding angle was dynamically adjusted to optimize seed density control data.

8. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 7, characterized in that: S6 further includes: The seed density control data is combined with the data frequency update to monitor the landing point distribution in real time. If the landing point distribution exceeds the preset threshold, the prediction output is integrated to generate and determine the distribution deviation value. The distribution deviation value is obtained and the path is dynamically adjusted by correcting the trajectory parameters. The offset compensation of the observation data is fused to determine the correction vector group. Based on the correction vector group, real-time feedback is extracted from the fusion of observation inputs to obtain dynamic path adjustment parameters for correcting the flight trajectory.

9. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 7, characterized in that, S7 specifically includes: Flight altitude updates are obtained from path adjustment parameters, and flight speed optimization is determined by integrating wind field change trends to obtain landing point distribution density. The threshold determination result is used to judge the deviation of the seeding trajectory based on the density of the landing point distribution. If the deviation exceeds the preset threshold, the terrain interference factor is integrated to obtain the seeding timing. By combining the timing of seed sowing with dynamic monitoring of the seed landing point, soil adaptation adjustments are extracted to determine the requirements for uniform sowing.

10. The method for optimizing the seed dispersal path of forage grass based on airflow simulation data according to claim 9, characterized in that: The S7 also includes: Based on the uniform dispersal requirements, the threshold judgment results are integrated from the flight speed optimization to generate dispersal trajectory offset compensation, and the final path planning is obtained; During the final path planning, the distribution density of the landing points is adjusted and integrated based on soil adaptation, the landing points are judged and dynamically monitored, and the path output that meets the requirements of uniform spreading is obtained.

Images