A method and system for precision pesticide application by unmanned aerial vehicles (UAVs) in mountainous areas

Abstract

This application relates to the field of drone-based pesticide application technology, and discloses a method and system for precise drone-based pesticide application in mountainous areas. The method includes: acquiring three-dimensional point cloud data of the operation area and performing gridding processing; calculating local slope values; generating a slope distribution map; extracting slope values ​​of uphill and downhill areas; determining the drone nozzle adjustment angle based on the slope values; when the angle exceeds a preset threshold, calculating a compensation coefficient by fusing the slope values ​​with historical flight data to obtain nozzle attitude parameters; acquiring plant images of the operation area and performing segmentation processing; analyzing leaf density and health indicators; generating a control matrix by combining state classification results with slope values; and determining an initial pesticide spraying distribution scheme by comprehensively considering real-time environmental data, vegetation cover density, and terrain factors; if the amount of pesticide in the uphill area is insufficient in the initial scheme, adjusting the pump pressure cyclically through feedback control, and controlling the nozzle to perform real-time attitude switching and flow rate regulation. This application improves the accuracy of pesticide spraying in mountainous environments.

Description

Technical Field

[0001] This application relates to the field of drone-based pesticide application technology, and in particular to a method and system for precise pesticide application by drones in mountainous areas. Background Technology

[0002] With the widespread adoption of drone technology in agriculture, it has demonstrated its advantages in pesticide spraying operations, offering high efficiency and flexibility. However, in complex terrain areas such as hills and mountains, the precision and adaptability of traditional drone spraying methods are severely lacking, leading to unsatisfactory spraying results and pesticide waste.

[0003] Specifically, existing technologies face several key challenges: First, mountainous terrain exhibits significant slope variations, resulting in substantial differences in the relative spatial relationship between plant canopies and drone sprayers between uphill and downhill areas. Existing methods lack the ability to perceive and analyze terrain slope in real-time, making it impossible to dynamically adjust the sprayer's orientation to maintain the spray direction perpendicular to the plant canopy surface. This leads to uneven pesticide coverage and even severe drift. Second, the plant's growth status (e.g., leaf density, health) is spatially uneven and coupled with the terrain. For example, plants on uphill areas may be sparse due to soil erosion, while those on downhill areas may be more lush. Existing technologies fail to deeply integrate terrain information with plant status information, resulting in static and coarse application plans that cannot achieve "on-demand distribution." This means they cannot guarantee sufficient pesticide application in affected areas, nor can they prevent over-application in healthy or sparsely planted areas. Furthermore, existing systems suffer from lag in control loop response. When insufficient pesticide distribution is detected in uphill areas, traditional control methods struggle to adjust the spray pump pressure in a timely and precise manner to change the pesticide flow rate. They are unable to achieve coordinated control of flow rate and attitude during rapid flight, resulting in the final pesticide application path failing to truly adapt to continuous terrain changes.

[0004] Therefore, there is an urgent need for a precision pesticide application solution for mountain drones that can integrate real-time terrain perception, plant status recognition, and intelligent decision control to solve the aforementioned technical bottlenecks. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method and system for precision pesticide application using mountain drones, which solves the problems of inaccurate and uneven pesticide application and poor environmental adaptability in the prior art, thereby improving the accuracy and efficiency of agricultural drone operations.

[0006] Firstly, this application provides a method for precise pesticide application using unmanned aerial vehicles (UAVs) in mountainous areas, the method comprising: The terrain of the work area is scanned by a terrain scanning sensor to obtain three-dimensional point cloud data. The three-dimensional point cloud data is then processed into a grid, and the local slope value of each grid point is calculated to generate a slope distribution map on the current flight path. The slope values ​​of the upslope and downslope areas are extracted from the slope distribution map, and the required adjustment angle of the drone nozzle is determined based on the slope values. The adjustment angle is the angle perpendicular to the plant canopy. If the adjustment angle exceeds a preset threshold, the slope values ​​and historical flight data are combined to calculate the compensation coefficient and obtain the nozzle attitude parameters of the upslope and downslope areas. The plant images in the work area are acquired and segmented. Leaf areas are extracted and leaf density and health indicators are analyzed to determine the plant health status and obtain status classification results. A control matrix is ​​generated by combining the status classification results and the slope value. The initial allocation scheme of the pesticide spraying amount is determined by combining real-time environmental data, vegetation cover density and terrain factors. If the amount of pesticide in the uphill area is lower than that in the downhill area in the initial distribution plan, the pump pressure is adjusted cyclically through feedback control to obtain the amount of pesticide sprayed to match the slope and plant condition. The instructions of the UAV flight controller are updated to control the nozzle to perform real-time attitude switching and flow regulation, so as to achieve precise pesticide application that adapts to changes in terrain.

[0007] Secondly, this application provides a mountain unmanned aerial vehicle (UAV) precision pesticide application system, the system comprising: The terrain scanning module is used to scan the terrain of the work area through the terrain scanning sensor, obtain three-dimensional point cloud data, perform gridding processing on the three-dimensional point cloud data, calculate the local slope value of each grid point, and generate a slope distribution map on the current flight path. Angle decision module is used to extract the slope values ​​of the upslope and downslope areas according to the slope distribution map, and determine the required adjustment angle of the drone nozzle based on the slope values. The adjustment angle is the angle perpendicular to the plant canopy. If the adjustment angle exceeds a preset threshold, the slope value and historical flight data are combined to calculate the compensation coefficient and obtain the nozzle attitude parameters of the upslope and downslope areas. The pesticide application module is used to acquire plant images in the work area and perform image segmentation, extract leaf areas and analyze leaf density and health indicators, determine the health status of the plants, obtain status classification results, generate a control matrix by combining the status classification results and the slope value, and determine the initial distribution scheme of pesticide application amount by combining real-time environmental data, vegetation cover density and terrain factors. The real-time adjustment module is used to adjust the spray pump pressure cyclically through feedback control if the amount of pesticide in the uphill area is lower than that in the downhill area in the initial allocation scheme. This obtains the amount of pesticide sprayed to match the slope and plant status, and updates the instructions of the UAV flight controller to control the nozzle to perform real-time attitude switching and flow regulation, thereby achieving precise pesticide application that adapts to changes in terrain.

[0008] Compared with the prior art, the beneficial effects of the present invention are at least as follows: This application provides a method and system for precise pesticide application using unmanned aerial vehicles (UAVs) in mountainous areas. By comprehensively considering terrain, plant health status, and environmental factors, it achieves precise adjustment of pesticide spraying. First, it acquires three-dimensional point cloud data of the work area through terrain scanning sensors, generates a slope distribution map, and dynamically adjusts the nozzle attitude based on the slope change rate and surface undulation, ensuring that the spraying direction is perpendicular to the plant canopy and avoiding pesticide loss or uneven distribution. Second, by combining plant image analysis and health status classification, the system can automatically adjust the pesticide spraying volume according to different plant conditions, ensuring that diseased and weak plants receive sufficient drug treatment while avoiding waste on healthy plants.

[0009] Furthermore, the real-time adjustment module dynamically adjusts the spray pump pressure through a feedback control mechanism, optimizing the spray volume and nozzle attitude based on the deviation between the actual and target spray volumes. When a mismatch is detected, a gradient descent algorithm is used to optimize the spray volume, ensuring precise matching between the spray volume and plant condition, terrain, and environmental conditions. Finally, by integrating real-time environmental data and terrain factors, the allocation of spray volume is optimized to achieve precise pesticide application. Therefore, this application improves application efficiency, reduces pesticide waste, avoids environmental pollution, ensures crop health, and provides an efficient and intelligent pesticide application solution under complex terrain and environmental conditions. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating a method for precise pesticide application using a mountain drone, as described in an embodiment of this application. Figure 2 This is a slope distribution map of the current flight path in an embodiment of this application; Figure 3 This is a schematic diagram of the topographic profile of the uphill area according to an embodiment of this application; Figure 4 This is a schematic diagram of the topographic profile of the downhill area according to an embodiment of this application; Figure 5 This is a schematic diagram of a mountain drone precision pesticide application system according to an embodiment of this application. Detailed Implementation

[0012] This application provides a method and system for precision pesticide application using a mountain drone. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0013] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of a method for precise pesticide application using a mountain drone in this application includes: Step S1: Scan the terrain of the work area using a terrain scanning sensor to obtain three-dimensional point cloud data, perform gridding processing on the three-dimensional point cloud data, calculate the local slope value of each grid point, and generate a slope distribution map on the current flight path.

[0014] The process of generating a slope distribution map along the current flight path includes: processing the 3D point cloud data into a grid to generate a set of regular grid points; calculating the local slope value of each grid point based on the set of regular grid points to obtain a set of slope values; marking any grid point in the set of slope values ​​that has a slope value exceeding a preset threshold as a high-slope area, generating a set of high-slope areas; overlaying the set of high-slope areas with the flight path data to determine the slope distribution characteristics along the path; and using the slope distribution characteristics to generate a slope distribution map of the current flight path.

[0015] Specifically, in mountainous drone-based pesticide application operations, the steepness and undulation of the terrain are the primary factors affecting the accuracy of pesticide application. To achieve precise control, it is first necessary to digitally model the terrain of the area to be sprayed. The core of this is to generate a high-precision slope distribution map, providing a data foundation for subsequent nozzle attitude and pesticide flow rate decisions. In practice, the terrain scanning sensor carried by the drone, such as a LiDAR, emits laser pulses into the work area and receives the return signals. By measuring the flight time of the laser, a large number of three-dimensional coordinates of surface points are obtained, forming three-dimensional point cloud data. This data is the basic raw data for all subsequent terrain analysis; its essence is digital sampling of the ground surface, containing terrain elevation and planar position information. To transform discrete, irregularly distributed 3D point clouds into a regular data structure suitable for terrain analysis, the 3D point cloud data needs to be meshed. This process first filters and denoises the original point cloud, then uses spatial interpolation algorithms, such as Delaunay triangulation, to generate a continuous surface model. Finally, it is regularly discretized into a set of regular grid points at a specific resolution. Each grid cell contains the 3D coordinates of its center point or corner points, thus representing the continuous terrain as a regular numerical matrix, facilitating batch and uniform slope calculations. Based on the regular grid point set, the local slope value of each grid cell is calculated. The local slope value represents the degree of inclination of the ground surface at that point. The calculation method is based on the elevation values ​​of the grid points, using spatial difference algorithms such as the Horn algorithm to solve for the elevation change rate of the surface surface in the X and Y directions, and then calculating the slope angle using the arctangent function. After performing this calculation on all grid cells in the area, a complete set of slope values ​​is obtained, which quantitatively describes the steepness distribution of the terrain within the entire working area.

[0016] To identify steep slope areas that significantly impact flight safety and pesticide application effectiveness, a slope threshold was set. This threshold was based on the critical requirements for drone flight stability and effective pesticide deposition, comprehensively considering multiple factors such as the drone's maximum safe flight tilt angle, the pesticide drift risk range at typical application altitudes, and the canopy structure characteristics of the target crop. The entire slope value set was traversed, and all grid cells with slope values ​​exceeding the threshold were marked, generating a high-slope area set. This set identified all steep slope terrain requiring special attention within the application area. Subsequently, the high-slope area set was overlaid with the drone's preset flight path data. This analysis was achieved through spatial query and statistical methods: First, a buffer zone for the drone's flight path was established. This buffer zone was defined by establishing the trajectory range of the drone's flight path, defining an area of ​​predetermined width, for example, based on flight altitude and safe distance. High-slope grid cells within the buffer zone were then extracted. Next, spatial cluster analysis was performed on these grid cells to identify continuously distributed high-slope segments. Finally, statistical analysis was performed on the slope values ​​within each high-slope segment, calculating its maximum, minimum, average, and slope change rate. The slope distribution characteristics determined through this series of analyses include the start and end points of high-slope sections, their spatial distribution range, slope statistics, and slope variation trends. These characteristics quantitatively describe the steepness and changes in the terrain traversed by the UAV's flight path. Based on these determined slope distribution characteristics, a slope distribution map of the current flight path is generated, such as... Figure 2 As shown, this is a slope distribution map along the current flight path. This map is usually presented as a raster image or a vector path with slope attributes. It can intuitively and clearly show the slope changes along the flight path. As a key terrain information output, this slope distribution map provides an indispensable terrain input for adjusting the vertical attitude of the nozzle and the initial allocation of the spray volume in subsequent steps.

[0017] The system acquires 3D point cloud data by using terrain scanning sensors and performs meshing to calculate local slope values ​​and generate accurate slope distribution maps. By calculating slope values ​​and identifying high-slope areas, combined with UAV flight path data, the system accurately describes the terrain change characteristics along the flight path. The generated slope distribution map provides a crucial terrain information foundation for subsequent nozzle attitude adjustment and liquid distribution.

[0018] Step S2: Extract the slope values ​​of the uphill and downhill areas based on the slope distribution map, and determine the required adjustment angle of the drone nozzle based on the slope values. The adjustment angle is the angle perpendicular to the plant canopy. If the adjustment angle exceeds the preset threshold, calculate the compensation coefficient by integrating the slope value with historical flight data to obtain the nozzle attitude parameters of the uphill and downhill areas.

[0019] The process of determining the required adjustment angle of the drone nozzle based on slope values ​​includes: extracting slope values ​​of uphill and downhill areas from a slope distribution map to generate a slope value set; using a convolutional neural network to extract features from the slope value set to obtain a terrain feature set; calculating the relative angle between each terrain feature point and the plant canopy based on the terrain feature set to generate an angle distribution set; if the relative angle of the terrain feature points in the angle distribution set exceeds a preset threshold, it is marked as a first region, generating a first region set; and calculating the adjustment angle of the drone nozzle through the first region set to obtain the nozzle adjustment angle set.

[0020] Specifically, after obtaining the slope distribution map, to achieve precise coverage of the plant canopy with the pesticide, the nozzle posture needs to be dynamically adjusted according to the terrain slope to ensure that the spray direction is always perpendicular to the plant canopy surface, ensuring that the pesticide can evenly and effectively cover the entire plant canopy and achieve the best application effect. In practice, the slope values ​​of the uphill and downhill areas are first extracted from the generated slope distribution map. This process is achieved by setting positive and negative slope thresholds: areas with a slope greater than the positive threshold, such as 5°, are classified as uphill areas, and areas with a slope less than the negative threshold, such as -5°, are classified as downhill areas, thereby generating a set of slope values ​​containing area identifiers and corresponding slope values. To extract deep features from the slope value set, a pre-trained convolutional neural network model was used to process the slope value set. The input of this model was a gridded slope value matrix, and its training data included a large number of historical datasets of slope maps and corresponding terrain feature annotations. During training, the model automatically learned local terrain patterns and spatial correlations through convolutional and pooling layers, establishing a mapping relationship from the original slope to high-level features. In practical applications, the trained model can output a terrain feature set containing key parameters such as terrain orientation and slope change rate, providing a richer and more reliable terrain information foundation for subsequent sprinkler angle calculations. Based on the terrain feature set, the relative angle between each terrain feature point and the plant canopy was determined through spatial geometric calculations. Specifically, with the UAV position as the origin, spatial vectors pointing to the plant canopy normal vector and the sprinkler current orientation vector were constructed respectively. The relative angle was obtained by calculating the angle between the two vectors. This calculation considered the UAV flight altitude, the average plant height, and the terrain slope direction. The angle between the sprinkler current orientation and the canopy normal vector was solved through vector operations, generating an angle distribution set describing the angle deviation of the entire working area. By combining the slope value set and the terrain feature set extracted by the convolutional neural network, the adjustment angle of the drone nozzle is accurately calculated, thereby ensuring that the spraying direction is always perpendicular to the surface of the plant canopy, achieving the effect of uniform coverage of the pesticide.

[0021] To ensure the accuracy of the adjustments, an angle adjustment threshold, such as 10°, is set. The angle distribution set is traversed, and all terrain feature points with relative angles exceeding this threshold are marked as areas requiring adjustment, i.e., the first region, forming a first region set. This set accurately identifies areas where the nozzle orientation deviates too much from the canopy normal vector due to terrain undulations, requiring further adjustment. By marking areas with relative angles exceeding the preset threshold and generating the first region set, special terrain areas are identified and optimized to maximize spraying effectiveness. Finally, the specific UAV nozzle adjustment angle is calculated using the first region set. For each area requiring adjustment (i.e., the first region), if the area is uphill, the nozzle's relative angle is set to a downward angle, with the specific value potentially adjusted based on the slope. For example, the steeper the slope, the greater the downward angle the nozzle needs to adjust to; conversely, if it's downhill, the nozzle needs to be adjusted to a downward angle. For each region, the nozzle adjustment angle is calculated and recorded, forming a nozzle adjustment angle set. By calculating the nozzle angle in the first region, precise nozzle attitude adjustment for different terrains is achieved, ensuring accurate pesticide application in both uphill and downhill areas. Figure 3 The diagram shows a topographic profile of an uphill area. It clearly illustrates the relationship between terrain elevation and horizontal distance. Analysis of this profile reveals that during flight, UAV A's spray nozzle needs to be adjusted to a 25° downward angle to adapt to the continuously rising slope. This angle adjustment ensures that the spray direction remains perpendicular to the slope, allowing the pesticide to accurately cover the plant canopy surface and effectively preventing the pesticide from drifting forward due to the terrain slope. Figure 4 As shown, this is a schematic diagram of the terrain profile of the downhill area. The diagram clearly shows the terrain features and nozzle attitude adjustment strategy of UAV A during the downhill flight phase. Through this profile, it can be observed that as the horizontal distance increases, the terrain height shows a significant decreasing trend. Under these terrain conditions, the UAV nozzle needs to be adjusted to a 20° upward viewing angle to ensure that the liquid spray direction remains perpendicular to the downhill slope.

[0022] When the calculated adjustment angle exceeds the preset safety threshold, the compensation mechanism is activated. By integrating the real-time slope value with the successful application parameters under similar terrain conditions in historical flight data, the compensation coefficient is calculated using an optimization algorithm to correct the initially calculated nozzle attitude parameters. Finally, the downward spray attitude parameters suitable for uphill areas and the upward spray attitude parameters suitable for downhill areas are obtained, ensuring that the optimal application angle can be maintained under different slope conditions. Details will be explained later.

[0023] The process of obtaining nozzle attitude parameters for uphill and downhill areas includes: if the adjustment angle exceeds a preset threshold, obtaining the slope value from the slope distribution map and acquiring historical flight data of the UAV; using a particle swarm optimization algorithm to fuse the slope value and historical flight data to obtain a terrain compensation coefficient; calculating the downward spray attitude parameters based on the terrain compensation coefficient and the slope value of the uphill area to obtain the nozzle depression angle set for the uphill area; calculating the upward spray attitude parameters based on the terrain compensation coefficient and the slope value of the downhill area to obtain the nozzle elevation angle set for the downhill area; if the angle values ​​in the nozzle depression angle set or the nozzle elevation angle set exceed the safe range, a preset threshold table is used for correction to obtain the corrected angle set.

[0024] Specifically, when the nozzle adjustment angle exceeds a preset safety threshold, an intelligent compensation mechanism is activated to optimize the nozzle attitude parameters. The specific implementation process is as follows: First, the real-time slope value of the current working area is extracted from the slope distribution map. Simultaneously, historical flight data under similar terrain conditions is retrieved from the database. This historical flight data records the drone's flight trajectory and attitude information, providing feedback for subsequent attitude adjustments. Next, a particle swarm optimization algorithm is used to fuse the real-time slope value and historical data. This algorithm finds the optimal solution by simulating the foraging process of bird flocks. It can calculate the optimal terrain compensation coefficient based on multiple input variables, such as slope value and historical flight data. This compensation coefficient can dynamically adjust the drone's spraying attitude to cope with slope changes during actual flight. Specifically, the particle swarm optimization algorithm updates the terrain compensation coefficient through multiple iterations to ensure that the spraying angle adapts to different terrains and plant needs in real time during flight.

[0025] Based on the obtained terrain compensation coefficient, the nozzle attitude parameters for uphill and downhill areas are calculated separately: First, the base angle is determined according to the type of the first area. If the first area type is an uphill area, the nozzle needs a downward depression angle to be perpendicular to the canopy; if the first area type is a downhill area, it needs an upward elevation angle. Second, the terrain compensation coefficient is multiplied by the base angle to optimize the parameters. This compensation coefficient is derived by the particle swarm optimization algorithm based on historical operation data, which can intelligently correct the theoretical calculation value based on actual operation experience, effectively dealing with the influence of environmental factors such as airflow disturbance and liquid characteristics. Finally, the optimized angles are safety checked. By querying a preset safety threshold table, it is ensured that all angles are within the allowable working range of the equipment. Finally, a set of nozzle depression angles and a set of nozzle elevation angles containing the optimized angles of each location point are generated, which are sent to the flight control system as control commands for the servo mechanism for execution. By integrating real-time terrain data and historical operation experience, intelligent compensation optimization of nozzle attitude parameters is achieved, which not only ensures the accuracy of liquid spraying but also ensures the safety of UAV flight and operation, significantly improving adaptability and reliability in complex mountainous environments.

[0026] Step S3: Acquire plant images of the work area and perform image segmentation, extract leaf areas and analyze leaf density and health indicators to determine the health status of the plants, obtain status classification results, generate a control matrix by combining status classification results and slope values, and determine the initial allocation scheme of pesticide spraying amount by combining real-time environmental data, vegetation cover density and terrain factors.

[0027] The process of determining the health status of plants and obtaining status classification results includes: capturing plant images within the work area using an image acquisition device from a drone; denoising and enhancing the plant images using a preset image preprocessing algorithm to obtain optimized plant images; segmenting the optimized plant images using a semantic segmentation algorithm to extract leaf regions, obtaining a set of leaf regions; calculating the leaf pixel ratio and distribution uniformity using the set of leaf regions to obtain leaf density parameters; extracting color and texture features based on the leaf density parameters and a preset health indicator database to obtain a set of health indicators; and classifying the plant status using a random forest algorithm if the color or texture features in the health indicator set exceed a preset threshold range to obtain a status classification result.

[0028] Specifically, after completing terrain scanning and nozzle attitude parameter calculation, the process enters the pesticide application decision-making stage. Before pesticide application, it is necessary to analyze the plant status based on plant images, generate status classification results, and adjust the spraying volume and distribution of the pesticide according to different plant statuses. The specific implementation process is as follows: First, high-definition camera devices mounted on drones are used to collect plant images of the work area. Median filtering and histogram equalization algorithms are used to denoise and enhance the contrast of the plant images, resulting in optimized plant images. Next, a deep learning-based semantic segmentation model is used to analyze the preprocessed images. This model can accurately identify and segment the leaf regions in the images, generate corresponding binary masks, and form a complete set of leaf regions. Based on this set, the leaf density parameter is calculated by statistically analyzing the proportion of leaf pixels to total pixels, and the spatial distribution characteristics of the leaf regions are analyzed to obtain a uniformity index. Then, color features, including hue, saturation, and texture features in the HSV color space, are extracted from the segmented leaf regions. The LBP algorithm is used to calculate texture feature values, and these features are compared and analyzed against a pre-set health indicator database. This database contains typical health characteristic parameter ranges for different crops at different growth stages. When the extracted feature values ​​exceed the pre-set health threshold range, a random forest classifier is activated for state discrimination. This classifier uses leaf density, color features, and texture features as input feature vectors. By integrating the voting results of multiple decision trees, the plant status is divided into three categories: "healthy," "sub-healthy," and "unhealthy," outputting the final state classification result. Based on the state classification result, it helps identify which plants are in poor condition and which are in good condition, thereby assigning different application weights. In practice, the distribution of healthy and unhealthy plants is accurately marked to ensure the accuracy of pesticide spraying volume and application target areas. By using a drone image acquisition device to capture plant images, image preprocessing, semantic segmentation algorithms, and random forest algorithms, accurate determination of plant health status is achieved, and state classification results are obtained through classification, thus providing accurate data support for subsequent pesticide application decisions.

[0029] The initial allocation scheme for pesticide spraying includes: generating an initial control matrix using a grid partitioning method based on state classification results and slope values; smoothing the initial control matrix using a Kriging interpolation algorithm to obtain a smoothed control matrix; calculating vegetation cover density based on leaf density parameters; acquiring real-time environmental data; calculating the spatial distribution of pesticide spraying based on the smoothed control matrix, combined with vegetation cover density and real-time environmental data; obtaining the spraying distribution; if a local area in the spraying distribution exceeds a preset threshold range, adjusting the spraying amount using a support vector machine algorithm to obtain an adjusted spraying distribution; and generating an initial allocation scheme for pesticide spraying based on the adjusted spraying distribution and by incorporating terrain factors.

[0030] Specifically, after obtaining the plant status classification results and terrain slope data, the specific process of the pesticide application scheme is initiated. First, a standard grid system is established based on the planar coordinates of the work area. The status classification results and slope values ​​corresponding to each grid cell are digitally encoded to construct an initial control matrix. Each element in this matrix contains two types of key information: plant status weight coefficient and slope value. The weight coefficient reflects the impact of plant health status on pesticide demand; the more unhealthy the plant, the greater its pesticide demand, and the larger the weight coefficient. Next, the initial control matrix is ​​smoothed using a Kriging spatial interpolation algorithm. First, based on the slope and plant health status data in the initial control matrix, the spatial autocorrelation between grid points is calculated. This process typically relies on a semi-variogram, which represents the relationship between numerical differences between different locations and their spatial distances. Then, the weighted average of the 12 surrounding adjacent grid points is used to correct the value of each grid. This correction can effectively eliminate outliers caused by sensor sampling errors, making the control values ​​of each grid cell more accurate and stable, generating a spatially continuously changing smooth control matrix.

[0031] Furthermore, vegetation cover density is calculated based on leaf density parameters. Leaf density reflects the crop's growth status and vegetation cover degree in the area; generally, higher density indicates better vegetation cover, and vice versa. Next, vegetation cover density can be estimated through the spatial relationship between leaf density and vegetation distribution. Within the operational area, the leaf density parameter of each grid cell is correlated with the vegetation cover density of that cell. Regression analysis or exponential models are typically used to establish the mapping relationship between the two. For example, historical data and experimental results may show that under certain conditions, the relationship between leaf density and vegetation cover density is linear or follows an exponential growth pattern. In this case, a linear regression model can be used: Vegetation cover density = α × y + β, where y is the leaf density, and α and β are regression coefficients obtained by fitting historical data, which determine the relationship between leaf density and vegetation cover density. If the relationship exhibits exponential growth characteristics, an exponential model can be used: Vegetation cover density = C × C and K are constants that need to be fitted using historical or experimental data, describing the exponential relationship between leaf density and vegetation cover density. This method allows for the precise calculation of vegetation cover density within the application area. Vegetation cover density directly affects the absorption and distribution of pesticides. Dense vegetation canopy increases the residence time of pesticides on the surface, thus requiring a higher spraying rate to ensure the pesticide penetrates the leaves and roots of the plants. Conversely, sparse vegetation means a more even distribution of pesticides, typically requiring a lower spraying rate, as the pesticide can penetrate the soil or be absorbed by the plants more quickly. Furthermore, areas with high vegetation cover density may have more leaves absorbing the pesticide, reducing waste; while areas with low density may have uneven pesticide penetration or runoff, thus requiring a higher spraying rate to ensure each plant receives sufficient pesticide. By considering vegetation cover density, the spraying rate can be precisely adjusted to avoid waste or inadequacy, ensuring uniform coverage of the entire application area, improving application effectiveness and crop health.

[0032] Furthermore, environmental data, such as temperature, humidity, and real-time wind speed, is collected in real time by environmental sensors. Firstly, temperature directly affects the evaporation rate of the pesticide solution. In high-temperature environments, the solution evaporates more easily, resulting in insufficient coverage of the plants. To avoid pesticide loss, the spraying amount needs to be increased. Low temperatures may lead to uneven pesticide deposition, so the spraying amount can be appropriately reduced to avoid waste. Humidity affects the diffusion and adsorption efficiency of the sprayed solution. In high-humidity environments, the solution does not evaporate easily and remains for a longer time, so the spraying amount can be appropriately reduced. In low-humidity environments, the solution evaporates easily, requiring an increased spraying amount to compensate for evaporation losses. Wind speed affects the drift and distribution of the pesticide solution. In areas with high wind speeds, the solution is easily dispersed, leading to uneven distribution. Therefore, the spraying amount needs to be increased and the nozzle orientation adjusted. In low-wind-speed environments, the solution tends to deposit on the plants, allowing for a reduction in the spraying amount to avoid waste. Therefore, incorporating real-time environmental data allows for adjustments to the spraying amount to ensure the uniformity and effectiveness of the pesticide solution.

[0033] By combining a smoothing control matrix, real-time environmental data, and vegetation cover density, a multivariate regression model is used to further adjust the spatial distribution of pesticide spraying. The specific process is as follows: The multivariate regression model uses plant status weighting coefficients, slope values, deviations between real-time and baseline temperatures, deviations between real-time and baseline humidity, and real-time wind speed data as input features. It is trained on a large amount of historical data to determine the relationship between each feature and the pesticide spraying amount. The training data typically includes optimal spraying amounts under different slopes, plant health conditions, and various environmental conditions. Once the model is trained, it calculates the target spraying amount for each grid cell based on the input feature data and through regression. For example, a grid cell in a sub-healthy state, with a certain slope, and under specific temperature and humidity conditions might be assigned a spraying amount of 0.65 liters per acre after model calculation. By traversing all grid cells, a spatialized spraying amount distribution map that meets both crop needs and environmental conditions is finally generated, providing a quantitative basis for precision pesticide application.

[0034] Finally, the adjusted spraying distribution is further integrated with detailed terrain factors. Through terrain complexity analysis, spraying redundancy is increased in areas with drastic slope changes, and spraying is reduced on windward slopes. Ultimately, an initial pesticide spraying distribution scheme that meets crop needs and adapts to terrain characteristics is generated. Details will be explained later.

[0035] The initial allocation scheme for pesticide spraying volume, which integrates topographic factors, includes: multi-source coupling processing of the adjusted spraying volume distribution with the topographic factors of the work area, including slope change rate, surface undulation amplitude, and soil type; constructing a topographic-sensitive allocation matrix based on multi-source input, which represents the influence weight of different topographic units on the pesticide spraying volume; in the allocation matrix, the slope change rate is used to correct the migration trend of the pesticide along the slope direction, the surface undulation amplitude is used to correct the spraying uniformity, and the soil type is used to correct the permeability and adsorption characteristics of the pesticide; and dynamically adjusting the spraying volume of each grid unit according to the influence weight to generate an initial allocation scheme for pesticide spraying volume that integrates topographic factors.

[0036] Specifically, after obtaining the spray volume distribution that has undergone threshold verification and support vector machine optimization, a deep fusion process with terrain factors is initiated. First, the spray volume distribution, after threshold verification and support vector machine optimization, is coupled with terrain factors extracted from a high-precision digital terrain model through multi-source data analysis. These terrain factors include: slope change rate, obtained by calculating the elevation difference between adjacent grid cells, used to quantify the rate of change in terrain steepness; surface undulation amplitude, calculated by analyzing the standard deviation of local elevations, reflecting the micro-undulation characteristics of the surface; and soil type, mainly used to correct the permeability and adsorption characteristics of the pesticide solution. Different soil types, such as sandy soil, clay, and loam, have a significant impact on the absorption, penetration, and diffusion characteristics of the pesticide solution. Based on these multi-source inputs, a terrain-sensitive allocation matrix is ​​constructed. The matrix uses the analytic hierarchy process (AHP) to determine the influence weights of each factor: slope change rate accounts for 40%, surface undulation amplitude accounts for 30%, and soil type accounts for 30%. The comprehensive adjustment coefficient for each grid cell is calculated through weighted summation, specifically using the formula: Adjustment coefficient = 0.4 × slope correction value + 0.3 × undulation correction value + 0.3 × soil type correction value. The slope correction value is dynamically adjusted based on the slope change rate; when the change rate exceeds 5% / meter, the spraying amount is increased by 3-5% on the uphill direction to compensate for pesticide runoff loss. The undulation correction value is determined based on the surface undulation amplitude; when the undulation exceeds 0.15 meters, the spraying amount is increased by 8% in raised areas and by [missing value] in depressed areas. Reduce the spraying amount by 5%. Soil type correction values ​​can be calculated according to the following rules: Sandy soil: Sandy soil has high permeability, and the pesticide solution easily penetrates into the ground. Therefore, sandy soil areas may need to reduce the spraying amount, for example, by 5%-10%, because the pesticide solution will seep down quickly, which may result in the crop not being able to absorb it effectively. Clay soil: Clay soil has poor permeability, and the pesticide solution will stay on the surface for a longer time. Therefore, clay soil areas need to increase the spraying amount, for example, by 10%-15%, which can ensure that the pesticide solution can be effectively absorbed by the plants. Loam soil: Loam soil has moderate permeability, and the penetration and absorption of the pesticide solution are relatively balanced. Therefore, the spraying amount is usually kept at the standard level and no major adjustments are needed.

[0037] Subsequently, the grid cells are dynamically adjusted based on the allocation matrix: first, a comprehensive adjustment coefficient is applied to the base spraying amount; then, Gaussian smoothing is performed using a 3×3 spatial convolution kernel to ensure a natural transition in spraying amount between adjacent grids; finally, a constrained optimization algorithm is used to accurately allocate each grid cell while ensuring the total spraying amount does not exceed the tank capacity. The generated initial allocation scheme is stored in GeoTIFF format, with each pixel containing latitude and longitude coordinates and spraying amount information, providing standardized input for subsequent real-time control, while retaining a 10% dynamic adjustment margin to cope with environmental changes in actual operations. By establishing a quantitative relationship between terrain factors and spraying parameters, precise adaptation of pesticide distribution to terrain features is achieved, effectively solving the problem of uneven pesticide distribution in complex terrain and significantly improving the application accuracy and resource utilization efficiency of mountain agriculture.

[0038] Step S4: If the amount of pesticide in the uphill area is lower than that in the downhill area in the initial distribution plan, the spray pump pressure is adjusted cyclically through feedback control to obtain the amount of pesticide sprayed to match the slope and plant status. The instructions of the UAV flight controller are updated to control the nozzle to perform real-time attitude switching and flow regulation, so as to achieve precise pesticide application that adapts to changes in terrain.

[0039] The process of obtaining the appropriate pesticide spraying volume based on slope and plant condition includes: if the difference in pesticide spraying volume between uphill and downhill areas exceeds a preset threshold, real-time slope and plant condition data are acquired via sensors to generate an initial regional distribution; based on the initial regional distribution, a linear regression algorithm is used to fit the relationship between slope data and pesticide spraying volume to obtain a spraying volume adjustment model; using the spraying volume adjustment model, combined with a feedback control mechanism, the dynamic adjustment value of the spray pump pressure is calculated to generate a pressure control scheme; if the flow rate in a local area of ​​the pressure control scheme deviates from the requirements of the plant condition data, the spray pump pressure is optimized using a gradient descent algorithm to obtain an optimized pesticide flow distribution; based on the optimized pesticide flow distribution, a pesticide spraying volume matching the slope and plant condition is generated.

[0040] Specifically, an initial spraying volume allocation plan was generated based on factors such as plant health status and terrain slope. However, the initial allocation plan may have issues with unreasonable spraying volumes in certain local areas, such as insufficient spraying in uphill areas. Therefore, after obtaining the initial allocation plan, a feedback control loop is used to ensure that the spraying volume is dynamically adjusted based on real-time data to achieve a more precise spraying effect. The specific implementation process is as follows: First, the difference in pesticide distribution between the uphill and downhill areas is monitored in real time. When the difference exceeds a preset threshold, such as when the average spraying amount in the uphill area is 15% lower than that in the downhill area, it means that the pesticide spraying amount in the uphill area is significantly lower than that in the downhill area. This may cause pesticide to run off in the steep slope area, thus affecting the application effect. In this case, a feedback control loop is triggered to compensate for the pesticide runoff caused by the slope and ensure that the spraying amount in the uphill area is reasonably increased. The specific process is as follows: The latest slope data, plant status data, and real-time environmental data are obtained through lidar and multispectral sensors to generate an initial regional distribution map containing spatial location information. This distribution map combines the slope, plant health status, and other environmental data such as temperature and humidity of each grid unit to provide a complete spatial layout that reflects the terrain changes and crop growth status within the operation area.

[0041] Next, based on the initial regional distribution map, and using historical data and the least squares method to fit the relationship between slope data and pesticide spraying volume, a spraying volume adjustment model is obtained. Specifically, the optimal pesticide spraying volume is calculated based on different slope conditions. First, the historical data includes spraying volumes and actual spraying effects under different slope conditions, such as crop growth and pesticide coverage. Using the least squares method, the slope value from the historical data is used as the independent variable, and the actual spraying volume as the dependent variable, for linear regression analysis to solve for the optimal regression equation. The regression equation is typically expressed as: Spraying volume = a × Slope + b, where a and b are regression coefficients obtained by fitting historical data, reflecting the relationship between slope and spraying volume. For example, if the slope increases, more pesticide may be needed to compensate for pesticide loss. Through the regression model, the optimal spraying volume for each grid unit under different slope conditions can be accurately calculated, providing a reasonable basis for subsequent spraying volume allocation, thereby ensuring uniform pesticide distribution and optimal crop protection. This spray volume adjustment model, combined with a PID feedback control mechanism, calculates the dynamic adjustment value of the spray pump pressure. First, it compares the deviation between the actual spray volume and the target value. Then, it uses a proportional term to quickly respond to the deviation, an integral term to eliminate steady-state error, and a derivative term to predict the trend, ultimately generating a precise pressure control scheme. When the pressure control scheme does not match the plant's needs in a localized area, such as when a region has dense vegetation but insufficient pressure adjustment, a gradient descent algorithm is activated for optimization. This algorithm uses spray uniformity as the loss function, iteratively adjusting the spray pump pressure parameters and searching for the optimal solution along the reverse gradient direction. This yields spatially distributed spray parameters, which are then sent to the flight controller in real time. By adjusting the PWM duty cycle, the spray pump motor speed is precisely controlled, and the servo mechanism is coordinated to switch the nozzle attitude, achieving true three-dimensional precision spraying. By establishing a closed-loop feedback control system, real-time dynamic optimization of the pesticide flow rate is achieved, effectively solving the problem of uneven pesticide distribution caused by terrain changes and ensuring optimal spraying results across different slope areas.

[0042] The process involves controlling the nozzles to perform real-time attitude switching and flow regulation to achieve precise pesticide application that adapts to terrain changes. This includes: the drone applying pesticide based on the matched slope and plant condition; if the drone's spraying angle deviates from the actual terrain features by more than a preset threshold during application, the nozzle attitude is optimized using a gradient descent algorithm to obtain an adjusted spraying angle distribution; based on the adjusted spraying angle distribution and real-time wind speed data, the drone control commands are updated to generate a spatially adaptive pesticide application command set; and the nozzles are controlled to perform real-time attitude switching and flow regulation, enabling the drone to obtain a precise pesticide application path that adapts to terrain changes along its flight path.

[0043] Specifically, after obtaining the appropriate pesticide spraying amount to match the slope and plant condition, the precision application phase begins. The specific implementation process is as follows: First, the UAV flight control system, based on the spatial coordinates obtained by the real-time positioning system, calls the pesticide flow rate value at the corresponding location to drive the proportional valve for precise application. During this process, the UAV attitude is monitored in real time through inertial measurement unit (IMU) and global positioning system (GPS) data, and terrain features are continuously scanned by lidar to calculate the deviation between the actual spraying angle and the ideal angle of the nozzle. When the spraying angle deviation exceeds a preset threshold, the gradient descent optimization algorithm is immediately activated. This algorithm uses pesticide coverage uniformity as the objective function and iteratively calculates the adjustment amount of the nozzle in each degree of freedom to find the optimal solution that minimizes the objective function. Specifically, the algorithm first calculates the gradient of the coverage loss function under the current attitude, and then adjusts the nozzle pitch and yaw angles in the opposite direction of the gradient with a preset step size, such as 0.1 rad / s. After multiple iterations, it converges to the optimal attitude and generates the adjusted spraying angle distribution.

[0044] While adjusting the nozzle attitude, real-time wind speed data is integrated to adapt to weather changes. Wind speed variations affect pesticide drift and deposition; therefore, the spray volume and nozzle angle need to be adjusted in real-time based on wind speed. The adjustment of spray volume is not detailed here, but for nozzle angle adjustment, for example, when the crosswind speed exceeds 2 m / s, a 3-5° offset compensation is added in the upwind direction; when the downwind speed exceeds 3 m / s, the nozzle pressure is reduced by 5-8% to reduce drift. Based on the optimized spray angle distribution and wind speed data, the system generates a spatially adaptive pesticide application command set. This command set includes not only nozzle attitude adjustment commands but also spray flow control commands. The UAV control system adjusts the nozzle attitude switching and flow control in real-time according to these commands, ensuring that the pesticide spray volume and angle match the actual terrain and plant condition. Finally, through updated nozzle control commands, the UAV flight controller controls the nozzle attitude switching and spray flow adjustment in real-time. The nozzle attitude is optimized based on the flight path and real-time adjustment needs, while the spray flow is dynamically adjusted according to different terrain and environmental conditions. Through real-time adjustments following the above steps, the drone can precisely apply pesticides along its flight path, ensuring that the spraying adapts to both terrain changes and plant conditions. This allows for adaptive adjustments to the spray volume and nozzle orientation, regardless of whether the plant is on a steep slope, on flat ground, or in areas with varying plant health conditions, achieving truly precise pesticide application.

[0045] By combining gradient descent algorithms, real-time wind speed data, and nozzle attitude optimization, the nozzle angle can be dynamically adjusted to ensure that the pesticide spraying is always perpendicular to the plant canopy, avoiding waste or uneven spraying caused by changes in terrain and wind speed. This allows for automatic adaptation to different terrain conditions and weather environments, ensuring effective pesticide coverage while reducing pesticide drift, evaporation, or loss, improving application efficiency, and guaranteeing crop health protection. Furthermore, the spatially adaptive application command set enables the drone to intelligently respond to real-time environmental changes, enhancing the intelligence level of the application process.

[0046] The above describes a method for precision pesticide application using a mountain drone in an embodiment of this application. The following describes a system for precision pesticide application using a mountain drone in an embodiment of this application. Please refer to [link / reference]. Figure 5 One embodiment of a mountain drone precision pesticide application system in this application includes: The terrain scanning module is used to scan the terrain of the work area using terrain scanning sensors, acquire three-dimensional point cloud data, perform gridding processing on the three-dimensional point cloud data, calculate the local slope value of each grid point, and generate a slope distribution map on the current flight path.

[0047] The angle decision module is used to extract the slope values ​​of the uphill and downhill areas based on the slope distribution map, and determine the required adjustment angle of the drone nozzle based on the slope values. The adjustment angle is the angle perpendicular to the plant canopy. If the adjustment angle exceeds the preset threshold, the slope value and historical flight data are combined to calculate the compensation coefficient and obtain the nozzle attitude parameters of the uphill and downhill areas.

[0048] The pesticide application module is used to acquire plant images in the work area and perform image segmentation, extract leaf areas and analyze leaf density and health indicators, determine the health status of plants, obtain status classification results, generate a control matrix by combining status classification results and slope values, and determine the initial allocation scheme of pesticide application amount by combining real-time environmental data, vegetation cover density and terrain factors.

[0049] The real-time adjustment module is used to adjust the spray pump pressure cyclically through feedback control if the amount of pesticide in the uphill area is lower than that in the downhill area in the initial allocation plan. This obtains the amount of pesticide sprayed to match the slope and plant status, and updates the instructions of the UAV flight controller to control the nozzle to perform real-time attitude switching and flow regulation, so as to achieve precise pesticide application that adapts to changes in terrain.

[0050] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0051] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0052] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for precise pesticide application by unmanned aerial vehicles (UAVs) in mountainous areas, characterized in that, The method includes: The terrain of the work area is scanned by a terrain scanning sensor to obtain three-dimensional point cloud data. The three-dimensional point cloud data is then processed into a grid, and the local slope value of each grid point is calculated to generate a slope distribution map on the current flight path. The slope values ​​of the upslope and downslope areas are extracted from the slope distribution map, and the required adjustment angle of the drone nozzle is determined based on the slope values. The adjustment angle is the angle perpendicular to the plant canopy. If the adjustment angle exceeds a preset threshold, the slope values ​​and historical flight data are combined to calculate the compensation coefficient and obtain the nozzle attitude parameters of the upslope and downslope areas. The plant images in the work area are acquired and segmented. Leaf areas are extracted and leaf density and health indicators are analyzed to determine the plant health status and obtain status classification results. A control matrix is ​​generated by combining the status classification results and the slope value. The initial allocation scheme of the pesticide spraying amount is determined by combining real-time environmental data, vegetation cover density and terrain factors. If the amount of pesticide in the uphill area is lower than that in the downhill area in the initial distribution plan, the pump pressure is adjusted cyclically through feedback control to obtain the amount of pesticide sprayed to match the slope and plant condition. The instructions of the UAV flight controller are updated to control the nozzle to perform real-time attitude switching and flow regulation, so as to achieve precise pesticide application that adapts to changes in terrain.

2. The method according to claim 1, characterized in that, Generate a slope distribution map along the current flight path, including: The three-dimensional point cloud data is processed into a grid to generate a regular grid point set. Based on the regular grid point set, the local slope value of each grid point is calculated to obtain a slope value set. If the slope value of any grid point in the slope value set exceeds a preset threshold, it is marked as a high slope area, and a high slope area set is generated. The high slope area set is overlaid with the flight path data to determine the slope distribution characteristics on the path. Using the slope distribution characteristics, a slope distribution map of the current flight path is generated.

3. The method according to claim 1, characterized in that, Determining the required adjustment angle for the drone nozzle based on the slope value includes: The slope values ​​of the upslope and downslope areas are extracted from the slope distribution map to generate a slope value set. A convolutional neural network is used to extract features from the slope value set to obtain a terrain feature set. Based on the terrain feature set, the relative angle between each terrain feature point and the plant canopy is calculated to generate an angle distribution set. If the relative angle of the terrain feature points in the angle distribution set exceeds a preset threshold, it is marked as a first region, and a first region set is generated. The adjustment angle of the drone nozzle is calculated through the first region set to obtain the nozzle adjustment angle set.

4. The method according to claim 1, characterized in that, Obtain nozzle attitude parameters for uphill and downhill areas, including: If the adjustment angle exceeds the preset threshold, the slope value is obtained from the slope distribution map, and the historical flight data of the UAV is obtained. The slope value and the historical flight data are fused using the particle swarm optimization algorithm to obtain the terrain compensation coefficient. Based on the terrain compensation coefficient and the slope value of the uphill area, the downward spray attitude parameters are calculated to obtain the nozzle depression angle set of the uphill area. Based on the terrain compensation coefficient and the slope value of the downhill area, the upward spray attitude parameters are calculated to obtain the nozzle elevation angle set of the downhill area. If the angle values ​​in the nozzle depression angle set or the nozzle elevation angle set exceed the safe range, a preset threshold table is used for correction to obtain the corrected angle set.

5. The method according to claim 1, characterized in that, Assess the plant's health status and obtain status classification results, including: The image acquisition device of the drone is used to capture images of plants in the work area. The plant images are then denoised and enhanced using a preset image preprocessing algorithm to obtain optimized plant images. A semantic segmentation algorithm is used to segment the optimized plant image, extract leaf regions, and obtain a set of leaf regions. Based on the set of leaf regions, the percentage of leaf pixels and the uniformity of distribution are calculated to obtain the leaf density parameter. According to the leaf density parameter, combined with a preset health indicator database, color features and texture features are extracted to obtain a set of health indicators. If the color features or texture features in the health indicator set exceed the preset threshold range, the plant status is classified using a random forest algorithm to obtain the status classification result.

6. The method according to claim 5, characterized in that, Determine the initial allocation plan for the pesticide spraying volume, including: Based on the state classification results and the slope value, an initial control matrix is ​​generated using a grid partitioning method, and the initial control matrix is ​​smoothed using a Kriging interpolation algorithm to obtain a smoothed control matrix. The vegetation cover density is calculated based on the leaf density parameter; real-time environmental data is obtained, and the spatial distribution of pesticide spraying amount is calculated by combining the vegetation cover density and real-time environmental data through the smoothing control matrix to obtain the spraying amount distribution. If a local area in the spraying amount distribution exceeds the preset threshold range, the spraying amount is adjusted by the support vector machine algorithm to obtain the adjusted spraying amount distribution. Based on the adjusted spray volume distribution, and by incorporating terrain factors, an initial allocation scheme for the pesticide spray volume is generated.

7. The method according to claim 6, characterized in that, The initial allocation scheme for pesticide spraying volume, which incorporates terrain factors, includes: The adjusted spray volume distribution is coupled with the topographic factors of the work area through a multi-source process, including the slope change rate, surface undulation amplitude, and soil type. A terrain-sensitive allocation matrix is ​​constructed based on multi-source input, and the allocation matrix is ​​used to represent the influence weight of different terrain units on the amount of pesticide sprayed. In the allocation matrix, the slope change rate is used to correct the migration trend of the pesticide solution along the slope direction, the surface undulation amplitude is used to correct the spraying uniformity, and the soil type is used to correct the permeability and adsorption characteristics of the pesticide solution. The spraying amount of each grid cell is dynamically adjusted according to the influence weight to generate an initial allocation scheme for the spraying amount of pesticide solution that incorporates terrain factors.

8. The method according to claim 1, characterized in that, To obtain the appropriate pesticide spraying volume based on slope and plant condition, including: If the difference in the amount of pesticide sprayed between the uphill area and the downhill area exceeds a preset threshold, real-time slope data and plant status data are obtained through sensors to generate an initial area distribution. Based on the initial regional distribution, a linear regression algorithm is used to fit the relationship between slope data and pesticide spraying volume to obtain a spraying volume adjustment model. Through the spraying volume adjustment model, combined with a feedback control mechanism, the dynamic adjustment value of the spray pump pressure is calculated to generate a pressure control scheme. If the flow rate in a local area of ​​the pressure control scheme deviates from the requirements of the plant status data, the pump pressure is optimized using a gradient descent algorithm to obtain an optimized pesticide flow rate distribution. Based on the optimized pesticide flow rate distribution, a pesticide spraying amount matching the slope and plant status is generated.

9. The method according to claim 1, characterized in that, Controlling the nozzles to perform real-time attitude switching and flow regulation enables precise pesticide application that adapts to changes in terrain, including: The drone applies pesticides based on the matched slope and plant condition. If the spraying angle of the drone deviates from the actual terrain features by more than a preset threshold during the application process, the nozzle attitude is optimized through a gradient descent algorithm to obtain an adjusted spraying angle distribution. Based on the adjusted spraying angle distribution and real-time wind speed data, the drone control commands are updated to generate a spatially adaptive pesticide application command set. This controls the nozzle to perform real-time attitude switching and flow regulation, enabling the drone to obtain a precise pesticide application path that adapts to changes in terrain along its flight path.

10. A mountain drone precision pesticide application system, used to implement the mountain drone precision pesticide application method as described in any one of claims 1-9, characterized in that, The system includes: The terrain scanning module is used to scan the terrain of the work area through the terrain scanning sensor, obtain three-dimensional point cloud data, perform gridding processing on the three-dimensional point cloud data, calculate the local slope value of each grid point, and generate a slope distribution map on the current flight path. Angle decision module is used to extract the slope values ​​of the upslope and downslope areas according to the slope distribution map, and determine the required adjustment angle of the drone nozzle based on the slope values. The adjustment angle is the angle perpendicular to the plant canopy. If the adjustment angle exceeds a preset threshold, the slope value and historical flight data are combined to calculate the compensation coefficient and obtain the nozzle attitude parameters of the upslope and downslope areas. The pesticide application module is used to acquire plant images in the work area and perform image segmentation, extract leaf areas and analyze leaf density and health indicators, determine the health status of the plants, obtain status classification results, generate a control matrix by combining the status classification results and the slope value, and determine the initial distribution scheme of pesticide application amount by combining real-time environmental data, vegetation cover density and terrain factors. The real-time adjustment module is used to adjust the spray pump pressure cyclically through feedback control if the amount of pesticide in the uphill area is lower than that in the downhill area in the initial allocation scheme. This obtains the amount of pesticide sprayed to match the slope and plant status, and updates the instructions of the UAV flight controller to control the nozzle to perform real-time attitude switching and flow regulation, thereby achieving precise pesticide application that adapts to changes in terrain.

Images