Tobacco unmanned aerial vehicle precision plant protection operation liquid flow regulation method

Abstract

The present application discloses tobacco unmanned aerial vehicle precision plant protection operation liquid flow regulation, relates to the tobacco operation technical field, and the method comprises the following steps: synchronously collecting three-dimensional point cloud data and multispectral image data of the tobacco canopy; based on the three-dimensional point cloud data, a two-dimensional canopy directed distance field is constructed in real time, and the canopy physiological information and real-time environmental wind speed inverted from the multispectral image data are fused to generate a liquid projection priority map through a demand mapping function; according to the liquid projection priority map, the respective corresponding instantaneous target coverage area of each nozzle in the partition controllable nozzle array of the unmanned aerial vehicle is determined; based on the instantaneous target coverage area and the current environmental parameters, a forward trajectory simulation is carried out by using a fog droplet dynamics agent model, and the optimal fog droplet diameter and liquid flow combination of each nozzle are solved with the goal of maximizing the effective deposition of the target area; and the partition controllable nozzle array is driven to perform the spraying operation according to the optimal combination obtained by solving.

Description

Technical Field

[0001] This application relates to the field of tobacco operation technology, and more specifically to the field of control of unmanned aerial vehicle (UAV) airborne spraying systems. In particular, it relates to a method for adjusting the flow rate of pesticide liquid in precision plant protection operations using tobacco UAVs. Background Technology

[0002] As an important economic crop, pest and disease control during the tobacco growth cycle is crucial for ensuring tobacco yield and quality. With the widespread application of drone technology in agricultural plant protection, drone-based plant protection has gradually become one of the mainstream methods for tobacco plant protection due to its advantages such as high operational efficiency, strong terrain adaptability, and reduction of manual field work intensity. The regulation of pesticide flow rate is a core technical aspect of drone-based tobacco plant protection operations. Its regulation effect directly affects the efficiency of pest and disease control, pesticide usage, and the risk of pesticide residues in tobacco leaves. If the flow rate is too high, it can easily lead to pesticide waste and excessive pesticide residues in some areas of tobacco leaves; if the flow rate is too low, it cannot meet the needs of pest and disease control, thus affecting tobacco growth and final quality. Therefore, achieving precise control of pesticide flow rate is of great significance for green tobacco production and improving economic benefits.

[0003] Currently, there are still significant shortcomings in the methods for adjusting pesticide flow rate in tobacco drone plant protection operations. On the one hand, existing methods mostly rely on preset fixed parameters such as tobacco planting spacing and growth cycle, or only make rough flow rate adjustments based on the drone's flight altitude. These methods fail to fully consider the actual differences in the growth conditions of the tobacco canopy in the field, resulting in a mismatch between the pesticide flow rate and the actual control needs of the canopy in different areas. This may lead to problems of pesticide oversupply or undersupply in some canopy areas.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones, in order to solve the aforementioned technical problems.

[0006] This application provides a method for adjusting pesticide flow rate in precision plant protection operations using tobacco drones, comprising: simultaneously acquiring three-dimensional point cloud data and multispectral image data of the tobacco canopy using a lidar and multispectral imager mounted on the drone; constructing a two-dimensional oriented distance field of the canopy in real time based on the three-dimensional point cloud data, and fusing canopy physiological information retrieved from the multispectral image data with real-time environmental wind speed, generating a pesticide projection priority map through a demand mapping function; determining the corresponding instantaneous target coverage area for each nozzle in the partitioned controllable nozzle array of the drone according to the pesticide projection priority map; performing forward trajectory simulation using a droplet dynamics surrogate model based on the instantaneous target coverage area and current environmental parameters, and solving for the optimal combination of droplet diameter and pesticide flow rate for each nozzle with the goal of maximizing effective deposition in the target area; and driving the partitioned controllable nozzle array to perform spraying operations according to the optimal combination obtained.

[0007] Based on the embodiments provided in this application, by simultaneously acquiring three-dimensional point cloud data and multispectral image data of the tobacco canopy using lidar and multispectral imager, it is possible to obtain real-time information on the true three-dimensional structure and physiological state of the tobacco canopy. Then, by combining the real-time environmental wind speed, a directional distance field of the canopy is constructed and a priority map for pesticide application is generated. This allows the pesticide application demand to be accurately matched with the actual growth status and environmental influence of the canopy in different areas. This effectively avoids the problem of excessive or insufficient pesticide supply caused by existing methods that do not combine the real canopy conditions and real-time environment. This makes the pesticide application of tobacco drone plant protection operations more targeted and meets the differentiated pest and disease control needs of the tobacco canopy. For zoned controllable nozzle arrays, this application first determines the instantaneous target coverage area for each nozzle based on the pesticide projection priority map. Then, based on this area and the current environmental parameters, it simulates the trajectory through a droplet dynamics proxy model and solves for the optimal combination of droplet diameter and pesticide flow rate. This allows the operating parameters of each nozzle to be highly adapted to the needs of the corresponding target canopy area, thereby maximizing the effective deposition of pesticide in the target area. This solves the problems of droplet drift and uneven deposition caused by the lack of precise optimization of individual nozzle operating parameters in existing methods, reduces pesticide waste and environmental pollution risks, and improves the actual control effect and resource utilization efficiency of precision plant protection operations using tobacco drones. Attached Figure Description

[0008] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0009] Figure 1 This is a flowchart of an optional method for adjusting the pesticide flow rate in precision plant protection operations using a tobacco drone, according to an embodiment of this application.

[0010] Figure 2 This is a flowchart of another optional method for adjusting the pesticide flow rate in precision plant protection operations using a tobacco drone according to an embodiment of this application;

[0011] Figure 3 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application.

[0012] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0013] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0014] According to one aspect of the embodiments of this application, such as Figure 1 As shown, this application provides a method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones, including:

[0015] S101 uses a lidar and multispectral imager mounted on a drone to simultaneously collect three-dimensional point cloud data and multispectral image data of the tobacco canopy;

[0016] S102, based on three-dimensional point cloud data, constructs a two-dimensional canopy oriented distance field in real time, and integrates canopy physiological information retrieved from multispectral image data with real-time environmental wind speed, and generates a drug delivery priority map through a demand mapping function;

[0017] After acquiring 3D point cloud data, instead of directly using the raw point cloud, it is converted into an intermediate representation that better reflects "sprayability"—the canopy directed distance field (SDF). Traditional 3D models or 2D orthogonal projection images mainly describe the outline and density of the canopy, but it is difficult to quantify the "accessibility" of a point in space relative to a leaf. This invention introduces the concept of SDF because it can accurately describe the spatial topological relationships inside and outside the canopy in a continuous and quantitative way. The SDF value of a point can characterize two key questions: "how far is it from the nearest leaf surface?" and "on which side of the leaf surface is the point located?", thus providing a direct geometric basis for determining whether the pesticide can reach the location and the ease with which it can reach it, achieving precise quantification of "sprayability".

[0018] S103, based on the liquid projection priority map, determine the instantaneous target coverage area for each nozzle in the UAV's partitioned controllable nozzle array;

[0019] S104, based on the instantaneous target coverage area and current environmental parameters, uses a droplet dynamics surrogate model to simulate the forward trajectory, and with the goal of maximizing the effective deposition in the target area, solves the optimal combination of droplet diameter and liquid flow rate for each nozzle;

[0020] It's important to explain that real-time environmental parameters are a crucial set of inputs, including at least ambient wind speed, ambient temperature, and ambient relative humidity. Ambient wind speed is acquired using an ultrasonic anemometer mounted on the drone, used to assess the primary external forces driving the pesticide spray. Ambient temperature and relative humidity are obtained through temperature and humidity sensors; these two parameters directly affect the evaporation rate of the pesticide droplets and are important factors in correcting droplet trajectories and estimating the actual deposition amount. These environmental parameters, along with the aforementioned canopy data, constitute the input to the decision-making system. They are fused under a unified timestamp to provide a comprehensive description of the on-site conditions for subsequent decision-making.

[0021] The droplet dynamics surrogate model plays a central role as a "digital simulator." It is a lightweight, differentiable set of computational models that integrates wind field forecasting, droplet trajectory simulation, and deposition effect prediction. Its input is connected to the canopy structure, physiological information, and environmental parameters provided by the sensing system, while its output is connected to the optimization controller, providing the expected deposition effect under different spray parameters. It serves as a bridge connecting "sensing" and "decision-making."

[0022] In this invention, "effective deposition" is a comprehensive evaluation index, defined as: the deposition coverage density of pesticide droplets on the target tobacco leaf surface (especially the underside of the leaf) reaches a preset threshold, with uniform spatial distribution, while minimizing droplet deposition in non-target areas (such as soil and inter-row gaps). Therefore, it is not only a requirement for the amount of deposition, but also a comprehensive reflection of the accuracy of deposition location, uniformity of coverage, and environmental friendliness, which is also the ultimate goal pursued by subsequent optimization control.

[0023] S105, drive the partitioned controllable nozzle array to perform spraying operations according to the optimal combination obtained by solving.

[0024] refer to Figure 1 The method for adjusting the liquid flow rate of pesticide spraying in precision plant protection operations using tobacco drones in this application relates to the control field of drone-borne spraying systems.

[0025] Furthermore, based on 3D point cloud data, a 2D canopy oriented distance field is constructed in real time, and canopy physiological information retrieved from multispectral image data is integrated with real-time environmental wind speed. A drug delivery priority map is generated through a demand mapping function, including:

[0026] The 3D point cloud data is projected onto a 2D grid parallel to the ground. For each grid cell, the directed distance from its center point to the nearest tobacco leaf surface along the vertical direction is calculated to construct a canopy directed distance field. When the center point of the grid cell is located in the outer space of the leaf, the directed distance value is positive; when it is located in the inner space of the leaf, the directed distance value is negative; and when it is located on the leaf surface, the directed distance value is zero.

[0027] The demand weight at each location in the directed distance field of the canopy is calculated using a demand mapping function. The calculation logic of the demand mapping function includes the synergistic canopy penetration rule, leaf attachment rule, and wind field pre-compensation rule.

[0028] The leaf area index retrieved from multispectral image data is used as a global adjustment factor and fused with the calculated demand weights to generate a drug projection priority map that reflects spatial structure, physiological state, and wind field environment.

[0029] The canopy physiological information includes leaf area index, chlorophyll content, and leaf water content. Chlorophyll content is inverted using red-edge isoband reflectance from multispectral images to assess the health of tobacco plants. Leaf water content is estimated using near-infrared isoband features, which affects the adhesion, spread, and absorption efficiency of pesticides on the leaf surface. This information collectively constitutes a comprehensive assessment of the tobacco canopy's physiological state. In one embodiment, the specific construction process is as follows: First, the three-dimensional point cloud data is projected onto a two-dimensional grid parallel to the ground. The granularity (resolution) of this grid is a key parameter, typically set to 1 / 5 to 1 / 3 of the diameter of a single nozzle's projection on the ground. This captures sufficient canopy detail while matching the nozzle's control capabilities, avoiding a surge in computational load due to excessively fine grids. The grid resolution Rgrid, the drone's flight speed Vdrone, and the control system's processing cycle Tcycle should satisfy Rgrid ≥ Vdrone × Tcycle to ensure real-time performance.

[0030] Secondly, for each cell in the grid, the directed distance from its center point to the nearest tobacco leaf surface is calculated. In this invention, the "vertical direction" refers to being absolutely perpendicular to the horizontal ground (the direction of gravity). This is because the movement of droplets in the air is mainly affected by gravity and wind, and their falling trajectory is clearly projected in the vertical direction. Using an absolutely vertical direction conforms to their physical motion laws, is simple to calculate, and has clear physical meaning. The calculation of the "nearest tobacco leaf surface" is achieved by constructing a KD-Tree spatial index structure for the point cloud. This algorithm can efficiently organize large amounts of point cloud data and supports fast nearest neighbor lookups for any grid center point, thus meeting the real-time computing needs during UAV flight.

[0031] Based on the embodiments provided in this application, traditional plant protection operations often rely on two-dimensional images or point cloud density to understand canopy structure, making it difficult to accurately distinguish between the front and back surfaces of leaves and their internal space. This application's embodiments construct a directed distance field for the canopy, quantifying three-dimensional spatial relationships into numerical fields with clear physical meaning. Positive, negative, and zero value regions correspond to the exterior, interior, and surface of the leaves, respectively. This representation enables the system to accurately perceive the spatial orientation of leaves, establishing an accurate geometric reference system for subsequent pesticide application mapping and overcoming the problem of uneven pesticide coverage caused by spatial ambiguity in traditional methods.

[0032] Furthermore, the canopy penetration rule includes: for regions where the directed distance value in the canopy directed distance field is negative, the required weight of each grid cell in that region is set to the exponential decay function value of the absolute value of the directed distance value corresponding to that grid cell, so as to achieve key marking of the hidden areas inside the canopy;

[0033] It's important to explain that the interior of the tobacco canopy, especially the middle and lower layers, suffers from poor ventilation and light penetration, making it a high-risk area for pests and diseases. However, as pesticide droplets penetrate the upper canopy, their penetration ability decreases exponentially with canopy depth due to collisions and interception. Therefore, using a negative exponential function to weight the interior of the canopy (regions with negative SDF) is precisely to mathematically characterize this physical law. The exponential form ensures sufficient attention is paid to deep, hidden areas; even if these areas are difficult to reach, their high demand weight will drive the system to attempt to use better parameters for coverage.

[0034] The leaf attachment rules include: for regions in the canopy directional distance field that are within a preset neighborhood where the directional distance value is close to zero, the demand weight of each grid cell in the region is set to the preset maximum demand weight value in order to lock the effective leaf attachment target area.

[0035] The canopy penetration rule is designed based on the physical penetration characteristics of drug droplets in the canopy: the droplet penetration ability decreases exponentially with increasing canopy depth. Applying a negative exponential weighting function to the internal canopy region accurately characterizes this physical law, ensuring sufficient attention is paid to deep, concealed areas.

[0036] The design of the leaf surface adhesion rule is based on the fundamental needs of plant protection operations: droplets can only exert their efficacy when they adhere to the leaf surface. By assigning the maximum weight to the leaf surface area, the most effective target area for pesticide deposition is directly locked.

[0037] In practice, the preset neighborhood value is determined based on the physical thickness of the tobacco leaf itself. For example, for most tobacco varieties, the leaf thickness is typically between 1 and 2 millimeters, so the preset neighborhood value can be set to 2 millimeters. This means that the system identifies the space within 2 millimeters above and below zero of the canopy directional distance field value as the leaf surface area where pesticide application is required.

[0038] The preset maximum weight value is a normalized relative value used to mark the highest demand level on the priority map. It is typically set to a baseline value of 1.0. The demand weights calculated for all other locations on the map are eventually scaled proportionally so that their values ​​fall between 0 and this maximum value of 1.0. Droplets only become effective when they impact and adhere to the leaf surface (where the SDF is near zero). The leaf surface is the only effective "target" for the pesticide. Therefore, assigning the "maximum weight" to the leaf surface area is directly determined by the fundamental purpose of plant protection operations, ensuring that system resources prioritize the most critical objectives.

[0039] In one specific implementation, the above rules are further specified, and the following liquid demand weighting function is proposed to calculate the demand value of each grid cell in the liquid projection priority map. ;

[0040]

[0041] in, Represents grid coordinates The final liquid spray demand weight at each location is a dimensionless relative value used to compare the priority of spray demand across different spatial locations. It represents the leaf area index obtained from multispectral images. As a global adjustment factor, the denser the canopy (the larger the LAI value), the higher the overall demand weight. It is a dimensionless pure ratio. The value represents the oriented distance field of the canopy at the center of the grid cell, with its physical unit being a unit of length (e.g., meter); A, B, C, and D represent normal parameters obtained by fitting historical plant protection experimental data; A represents the basic weight controlling canopy penetration demand, used to adjust the overall level of demand weight in the inner canopy region; B represents the rate of decay of penetration demand with increasing canopy depth; the larger the value of B, the faster the decay of attention to deeper regions, and its value reflects the canopy permeability characteristics of a specific tobacco variety and growth stage, with its dimension being the reciprocal of length [L]. -1[B] (e.g., "per meter"). It determines the rate at which the demand weight decreases with increasing canopy depth. Its value is related to the permeability characteristics of the tobacco canopy at a specific growth stage; the denser the canopy, the larger the B value usually is. C represents the maximum weight value controlling leaf attachment demand, determining the highest demand weight level achievable on the leaf surface area. D represents the effective width controlling the leaf attachment area. It defines the range of the "preset neighborhood near zero value," and its value is related to the average thickness of the tobacco leaves and the droplet size. For example, it can be set to a width of 2D to cover most of the leaf surface area, with the dimension being length [L] (e.g., "meter"). It defines the effective range of the "preset neighborhood near zero value," and its value matches the average thickness of the tobacco leaves and the size of the droplets.

[0042] Represent the Herveside step function, when When <0 (inside the canopy), =1; otherwise =1. Additionally... This corresponds to the canopy penetration rule. It only applies within the canopy (where SDF is negative), and its value decays exponentially as it penetrates deeper into the canopy (|SDF| increases). This corresponds to the leaf surface attachment rule. It is a Gaussian function that reaches its maximum value C when SDF=0 (leaf surface) and maintains a high weight in its neighborhood (the width is controlled by the parameter D), thereby accurately locking the effective attachment target area on the leaf surface. This represents the wind field compensation weight term calculated by the wind field pre-compensation rules.

[0043] Those skilled in the art should understand that calculation The first term in the formula This refers to a specific embodiment of a negative exponential function. The core characteristic of this negative exponential function is that its function value decays exponentially with increasing |SDF|, and the decay direction conforms to the technical objective of the canopy penetration rule. Any equivalent function transformation that satisfies this core characteristic (including but not limited to exponential functions with a base of e, 2, or other constants greater than 1, or other expressions obtained through equivalent mathematical transformations) falls within the protection scope of this invention.

[0044] The specific implementation of the calculation of the wind field compensation weight term is as follows:

[0045] The real-time measured environmental wind speed vector is decomposed into two components on the horizontal plane: one parallel to the UAV's current flight direction and one perpendicular to the flight direction. Based on the direction of the vertical component, two key regions are identified and delineated on a two-dimensional grid: the canopy region upstream of the current wind direction and the canopy region downstream of the current wind direction.

[0046] The wind field compensation weight is calculated for each grid cell. The rule is as follows: for grid cells located upstream in the wind direction, the wind field compensation weight is a positive number, and its magnitude is proportional to the magnitude of the vertical wind speed component. In other words, the greater the wind speed perpendicular to the flight direction, the greater the positive compensation weight assigned to the upstream area. Conversely, for grid cells located downstream in the wind direction, the wind field compensation weight is a negative number, and its absolute value is also proportional to the magnitude of the vertical wind speed component. For grid cells in other areas, the wind field compensation weight is set to zero. The proportional relationship between the positive and negative compensation weight values ​​and the magnitude of the vertical wind speed component is determined by a preset compensation coefficient. The essence of this calculation rule is to proactively offset the effect of crosswinds blowing the liquid away from the target area at the spatial decision-making level by using an asymmetric strategy of "increasing demand" upstream and "reducing demand" downstream.

[0047] Based on the embodiments provided in this application, the accessibility difficulty within the canopy increases exponentially with depth, while the probability of pests and diseases increases accordingly. This rule accurately describes this contradictory relationship through mathematical expression, ensuring priority coverage of high-value concealed areas. The leaf surface attachment rule assigns the highest weight to the zero-value neighborhood, based on the physical characteristic that the deposition efficiency of droplets on the leaf surface is much higher than in other areas. The two rules work together through different mathematical expressions to achieve a balance between "focused penetration" and "precise attachment".

[0048] Furthermore, the wind field pre-compensation rules include: based on the direction and magnitude of the real-time wind speed vector, positively strengthening the demand weight of the canopy area located upstream of the current wind direction, and negatively weakening the demand weight of the canopy area located downstream of the current wind direction, in order to pre-compensate for the spatial drift of the liquid.

[0049] Traditional methods adjust flow rates to compensate for wind field impacts during the execution phase, which suffers from a lag. This invention performs pre-compensation during the priority map generation phase of the decision-making process. By proactively adjusting spatial demand weights, it guides the system to target the corrected wind field location from the command source, thereby achieving a more stable and accurate countermeasure against wind field impacts at the physical level. The execution of the wind field pre-compensation rules includes:

[0050] The real-time wind speed vector is decomposed into two components on the horizontal plane: a first component parallel to the current flight direction of the drone, and a second component perpendicular to the current flight direction of the drone.

[0051] It should be noted that the first component primarily affects the drift timing characteristics of the liquid in the flight path (i.e., the lag or lead of the droplet landing position relative to the nozzle trigger). Its compensation mechanism is automatically implemented through the forward trajectory simulation step of the droplet dynamics proxy model. That is, the droplet dynamics proxy model dynamically adjusts the optimal spraying timing of each nozzle based on the value of the first component (e.g., triggering earlier in headwinds and delaying triggering in tailwinds), rather than through spatial correction via a demand weight map. Therefore, the asymmetric spatial weight compensation strategy specifically refers to the processing of the second component. This is determined by the difference in the physical action mechanisms of the two components. That is, the second component produces lateral offset (requiring map pre-compensation), while the first component produces longitudinal temporal offset (requiring nozzle trigger timing compensation).

[0052] Those skilled in the art should understand that this divide-and-conquer strategy decouples spatial weight compensation from temporal control, thereby improving the flexibility and computational efficiency of the system response.

[0053] For the second component, an asymmetric spatial weight compensation strategy is implemented, including: in the canopy area upstream of the current wind direction, the demand weight of the second component is increased according to the magnitude of the second component by a first preset ratio; in the canopy area downstream of the current wind direction, the demand weight of the second component is decreased according to the magnitude of the second component by a second preset ratio.

[0054] Among them, the absolute values ​​of the first preset ratio and the second preset ratio are both positively correlated with the magnitude of the second component, and the first preset ratio is positive and the second preset ratio is negative.

[0055] In practice, the real-time wind speed vector is first decomposed into components parallel and perpendicular to the flight direction. For the vertical component, the demand weight is linearly increased in the upstream canopy region according to its magnitude, while the demand weight is linearly decreased in the downstream region. The absolute values ​​of the first and second preset ratios are both positively correlated with the magnitude of the vertical wind speed component. A benchmark compensation intensity can be set, for example, defined as "every 1 m / s vertical wind speed causes a weight change of 0.15". When the vertical wind speed is 1 m / s, the demand weight of all grids on the upstream side increases by 0.15, and the demand weight of all grids on the downstream side decreases by 0.15. When the vertical wind speed increases to 2 m / s, the weight increase on the upstream side increases to 0.30, and the weight decrease on the downstream side increases to 0.30. This process uses a moving average filter to process the raw wind speed data to smooth short-term fluctuations and ensure the stability of the compensation strategy.

[0056] Based on the embodiments provided in this application, traditional wind field compensation is mostly achieved by adjusting the flow rate during the execution phase, which has a lag. The embodiments of this application construct a spatial pre-compensation mechanism during the decision-making phase through wind speed vector decomposition and asymmetric weight adjustment. After decomposing the wind speed into parallel and vertical components, the focus is on strengthening the vertical component upstream and weakening it downstream, which physically corresponds to the overall translational effect of the pesticide solution caused by crosswinds. This forward-looking spatial weight adjustment, compared to reactive flow rate adjustment, can more effectively offset the systematic influence of the wind field on the spatial distribution of the pesticide solution.

[0057] Furthermore, based on the liquid projection priority map, the instantaneous target coverage area corresponding to each nozzle in the UAV's partitioned controllable nozzle array is determined, including:

[0058] Establish the mapping relationship between the physical position of each nozzle in the partitioned controllable nozzle array and the spatial coordinates of the two-dimensional grid;

[0059] Based on the theoretical coverage of each nozzle, the corresponding candidate area is delineated on the liquid projection priority map;

[0060] The theoretical coverage area refers to the region under ideal static conditions of no wind, standard atmospheric pressure, and constant temperature, where the theoretical deposition probability of a group of liquid droplets with optimal droplet diameter generated by a single nozzle at its rated operating pressure is greater than or equal to a set threshold (e.g., 80%) on the target height plane of the canopy. This region is determined by the nozzle atomization angle, nozzle installation height, droplet diameter, and the nozzle's own flow characteristics, and its boundary is determined by forward trajectory simulation under zero wind speed conditions using a droplet dynamics surrogate model.

[0061] Perform dynamic region allocation optimization, including: for each nozzle, identify the spatial distribution of demand weights within its candidate region, and prioritize the allocation of the sub-region with the highest demand weight and spatial contiguousness as the instantaneous target coverage area of ​​that nozzle;

[0062] When the candidate areas of multiple nozzles overlap, the coverage area is arbitrated and redistributed according to the demand weight of each grid cell in the overlapping area by comparing weight attribution or proportional allocation, so as to ensure the optimal overall coverage efficiency.

[0063] Fixed zoning prevents nozzle resources from adapting to dynamic changes in canopy demand. This application establishes a mapping relationship between nozzle locations and grids, defining candidate regions for each nozzle within its theoretical coverage area (a circular area projected onto the ground considering atomization angle and height). A connected component analysis algorithm is used to identify the spatially continuous sub-regions within the candidate regions that have the highest demand weight as the instantaneous target coverage area.

[0064] When multiple sprinkler regions overlap, arbitration is conducted based on the weight distribution of each grid cell within the overlapping area: if the weight distribution is uneven, the area is assigned to a sprinkler with a higher historical average weight; if the distribution is uniform, the responsibility area is allocated according to the weight ratio of each sprinkler in the overlapping area. This dynamic allocation mechanism significantly improves resource utilization efficiency and coverage uniformity.

[0065] Based on the embodiments provided in this application, the core challenge faced by multi-nozzle systems is how to avoid duplicate coverage and missed spraying. This application achieves an upgrade from "fixed partitioning" to "dynamic partitioning" by establishing a precise mapping between nozzle positions and the grid, defining candidate regions for each nozzle, and then optimizing the allocation based on weight distribution. When regions overlap, the arbitration mechanism of weight comparison or proportional allocation is essentially solving the task allocation problem of a multi-agent system. This design ensures the optimal match between nozzle resources and space requirements.

[0066] Furthermore, such as Figure 2 As shown, based on the instantaneous target coverage area and current environmental parameters, a droplet dynamics surrogate model is used to simulate the forward trajectory. With the objective of maximizing effective deposition in the target area, the optimal combination of droplet diameter and liquid flow rate for each nozzle is solved, including:

[0067] S201, Construct a physical information neural network wind field forecaster; The physical information neural network wind field forecaster takes the canopy directional distance field, UAV flight status and real-time environmental wind speed as inputs, and through coupling the fluid dynamic physical constraints represented by the Navier-Stokes equations during its training and inference process, it deduces and outputs the three-dimensional dynamic disturbance wind field of the tobacco canopy region within a future preset time window.

[0068] Purely data-driven models exhibit weak generalization ability outside the training data range. The physical information neural network wind field forecaster of this invention incorporates the residuals of the Navier-Stokes equations as part of the loss function, ensuring that the model inference conforms to both statistical data patterns and strictly adheres to the conservation laws of fluid dynamics, significantly improving prediction accuracy and generalization ability in complex canopy environments. The fluid dynamic physical constraints represented by the Navier-Stokes equations are well-known technologies and will not be elaborated upon in this embodiment.

[0069] S202 utilizes a droplet dynamics surrogate model to simulate the forward trajectory, specifically through a differentiable spatiotemporal deposition potential field predictor. The spatiotemporal deposition potential field predictor takes the droplet diameter, liquid flow rate, spatial location, and three-dimensional dynamic disturbance wind field of the nozzle as inputs, and outputs the predicted deposition distribution map and the corresponding prediction confidence variance map of the nozzle within its instantaneous target coverage area.

[0070] The differentiable spatiotemporal sedimentary potential predictor employs a fully connected neural network structure. It takes nozzle parameters, spatial location, and dynamic wind field as input and outputs a predicted sedimentary distribution map and a confidence variance map. Its differentiability allows the optimization algorithm to calculate gradients through backpropagation, providing the necessary conditions for subsequent efficient gradient optimization.

[0071] S203, establish an optimization problem with the droplet diameter and liquid flow rate of all nozzles as decision variables;

[0072] Wherein, objective function Defined as:

[0073] ;

[0074] in, For the first Predicted sediment distribution map for each nozzle. This represents the predicted deposition distribution map for all nozzles. The total sediment distribution map obtained after spatial overlay; A map showing the priority of drug delivery; Map showing the total calculated sediment distribution Priority map of drug delivery Spatial Pearson correlation coefficient between them; Predict its confidence variance plot; This represents the variance plot of the prediction confidence for all nozzles. The total variance plot obtained after spatial overlay; The first weighting coefficient; These are the second weighting coefficients. The optimization problem aims to maximize... For example, if the primary objective is to ensure that the drug deposition distribution closely matches the canopy requirements, and the secondary objective is to reduce system uncertainty, then the first weighting factor can be set to 0.7, and the second weighting factor to 0.3. This means that in the optimization process, the accuracy of the deposition effect accounts for about 70% of the importance, while the system stability accounts for about 30%.

[0075] S204 employs a sequential quadratic programming algorithm to solve the optimization problem. During the solution process, the value of the canopy directed distance field is used as the basis for the spatial penalty term. This includes: for a spatial location, the corresponding penalty intensity is positively correlated with the value of the canopy directed distance field at that location. This results in the contribution of the predicted deposition amount in the air region where the value of the canopy directed distance field is greater than a preset positive threshold being attenuated in the objective function. This guides the optimization parameters to generate a drug deposition distribution that is highly matched with the spatial morphology of the canopy leaves.

[0076] The preset positive threshold is used to effectively distinguish between "acceptable adjacent leaf areas" and "ineffective pesticide spraying areas that need to be avoided" during the optimization process. The specific value of this threshold is determined based on the typical leaf distribution density and droplet dispersion characteristics of the tobacco canopy. In tobacco plant protection operations, the typical range of this preset positive threshold is 2 cm to 5 cm.

[0077] When the directional distance field value of the canopy is less than or equal to this threshold, it means that the spatial location is close to the tobacco leaf surface, and the distribution of pesticide droplets in this area is still considered to be within the range that may be effective for leaf surface adhesion.

[0078] When the directional distance field value of the canopy exceeds this threshold, it indicates that the location is far from the leaf surface and in an open area between the canopies. In this area, pesticide deposition mainly originates from ineffective drift, causing not only waste but also potential environmental pollution. Therefore, it is necessary to suppress the predicted deposition amount in such areas in the optimization objective.

[0079] In practical applications, the threshold can be adjusted within this range according to the canopy structure characteristics of the specific tobacco variety: for varieties with spread leaves and open canopies, a larger threshold (such as 4-5 cm) can be used; for varieties with compact leaves and dense canopies, a smaller threshold (such as 2-3 cm) can be used. This value ensures that the system can minimize the loss of pesticide spraying while ensuring effective leaf coverage.

[0080] Based on the embodiments provided in this application, the Navier-Stokes equations are embedded as physical constraints into the network training. This makes wind field prediction not only rely on statistical data patterns but also follow the basic principles of fluid mechanics, significantly improving the extrapolation capability in complex canopy environments. The differentiable deposition potential field predictor transforms time-consuming trajectory simulation into instantaneous inference, providing the necessary conditions for subsequent gradient-based optimization algorithms. The correlation term in the objective function ensures spraying accuracy, while the variance term controls system stability. This multi-objective design improves the overall performance of the system.

[0081] Furthermore, a sequential quadratic programming algorithm is used to solve the optimization problem, including:

[0082] Initialize the droplet diameter and liquid flow rate parameter combination for all nozzles;

[0083] At the current parameter point, based on the gradient information provided by the spatiotemporal sedimentary potential field predictor, the canopy oriented distance field, and the physical information neural network wind field predictor, the gradient of the objective function and the constraint conditions is calculated, and a local quadratic programming subproblem is constructed accordingly.

[0084] The constraints mainly fall into two categories:

[0085] Equipment physical constraints: The droplet diameter and pesticide flow rate of each nozzle must be within the operating range allowed by its hardware. For example, the droplet diameter of a certain nozzle may be adjustable from 50 to 150 micrometers, and the pesticide flow rate may be adjustable from 100 to 500 ml / min. The parameters obtained from the optimization solution must fall within this range.

[0086] System quality conservation constraint: The total output flow of all nozzles at the same time needs to match the current flight speed of the drone to ensure that the amount of pesticide applied per unit area meets agronomic requirements and avoids local over- or under-application.

[0087] Solve the local quadratic programming subproblem to obtain the parameter correction direction and step size that optimizes the objective of the local quadratic programming subproblem;

[0088] Update the droplet diameter and liquid flow rate parameters of all nozzles according to the parameter correction direction and step size;

[0089] The above steps are repeated iteratively until convergence. Convergence is the criterion for determining when the iterative solution process of the optimization algorithm can terminate. It typically includes two conditions, either of which must be satisfied:

[0090] Parameter variation convergence: The changes in all nozzle parameters (droplet diameter and flow rate) calculated in two consecutive iterations are very small, for example, the average change is less than one-thousandth.

[0091] Objective function improvement convergence: In two consecutive iterations, the numerical improvement of the comprehensive objective function is very slight, for example, the improvement is less than one ten-thousandth.

[0092] Throughout the iteration process, the Jacobian matrix of the global deposition distribution on each nozzle parameter is calculated and analyzed to characterize the parameter coupling effect among multiple nozzles. This Jacobian matrix is ​​then used to decouple the parameter coupling effect during optimization, thereby allocating droplet diameter and liquid flow rate to each nozzle when their coverage areas overlap to achieve globally optimal coverage. A data structure for calculating a Jacobian matrix is ​​used to quantify the mutual influence between nozzles. Each row of this matrix represents a nozzle parameter, and each column represents a spatial location. Each specific value in the matrix characterizes "how much a fine-tuning of the parameters of one nozzle will affect the deposition effect in the area covered by another nozzle." By analyzing this matrix, the coupling and interference between nozzle parameters can be clearly observed.

[0093] During the iterative process of the optimization algorithm, the system uses this Jacobian matrix to guide the direction of parameter updates. Specifically, the algorithm seeks an update direction that minimizes the negative interference (i.e., coupling effect) caused by a nozzle optimizing its own objective on other nozzles. This is equivalent to mathematically solving a multi-objective coordination problem, thereby achieving cooperation among nozzles rather than them operating independently.

[0094] A sequential quadratic programming algorithm integrates multi-source gradient information from sediment potential field, SDF field, and wind field forecaster to construct a local quadratic programming subproblem. The Hessian matrix approximation is updated using the BFGS method, and the Jacobian matrix is ​​calculated to characterize the parameter coupling effect between nozzles. During optimization, matrix information is used to actively decouple mutual interference, achieving true collaborative optimization. The convergence criterion is set as the parameter change or the improvement in the objective function being less than a preset threshold.

[0095] Based on the embodiments provided in this application, the sequential quadratic programming algorithm is applied to the fusion and utilization of multi-source gradient information. Gradient information from the sedimentation potential field, the directed distance field, and the wind field forecaster jointly guides the optimization direction, enabling the sequential quadratic programming algorithm to simultaneously consider sedimentation effects, spatial geometry, and environmental influences. The coupling effect between nozzles is explicitly characterized by the Jacobian matrix. Its technical essence is to decompose the high-dimensional optimization problem into interrelated sub-problems, revealing the intrinsic relationship between parameters through matrix operations, thereby achieving true global collaborative optimization.

[0096] Furthermore, the method also includes a closed-loop step of deposition effect feedback and online model parameter calibration, specifically including:

[0097] After the controllable nozzle array performs spraying operations and a preset deposition time has elapsed, the treated canopy leaf surface is scanned by a multispectral imager mounted on a drone to obtain leaf surface reflectance data in the near-infrared band.

[0098] The preset deposition time refers to the waiting time from the completion of spraying by the nozzle to the start of deposition effect testing. It is set based on the physical time required for pesticide droplets to impact, spread, and reach a stable adhesion state on the tobacco leaf surface. This time is usually determined through experimental observation, generally between 10 and 30 seconds, to ensure that the measured reflectance data accurately reflects a stable pesticide deposition situation, rather than a dynamically changing liquid film.

[0099] Based on leaf surface reflectance data and a pre-established reflectance and deposition calibration model, the actual drug deposition distribution map was calculated.

[0100] The specific steps for establishing the reflectivity and sedimentation calibration model include:

[0101] Sample preparation: On clean tobacco leaves, different known volumes of drug solution are sprayed quantitatively using a precision instrument to create a series of samples with precise gradient deposition amounts.

[0102] Data collection: After the drug solution stabilizes, use the same type of multispectral imager as the drone to scan these samples and obtain their reflectance data in the near-infrared and other bands.

[0103] Mathematical fitting: The measured reflectance data is correlated with the known sedimentation data. Mathematical methods such as linear regression or polynomial fitting are usually used to establish a conversion model from "reflectance value" to "sedimentation amount".

[0104] During flight, after spraying and waiting for the preset deposition time, the drone activates its onboard multispectral imager to scan the area that has just been treated. The system inputs the near-infrared reflectance data of each pixel obtained from the scan into a pre-established "reflectance-deposition amount" calibration model to calculate the actual amount of pesticide deposition corresponding to each pixel, thereby generating an "actual pesticide deposition amount distribution map".

[0105] Spatial difference comparison is performed between the actual drug solution deposition distribution map and the expected deposition distribution derived from the drug solution projection priority map to generate a deposition residual map;

[0106] The "actual drug solution deposition distribution map" is spatially compared pixel-by-pixel with the previously generated "expected deposition distribution map" (derived from the drug solution projection priority map). Specifically, the expected value is subtracted from the actual value to generate a new "deposition residual map". Positive values ​​in this map indicate that the actual deposition exceeds the expectation (over-deposition), while negative values ​​indicate that the actual deposition does not meet the expectation (under-deposition).

[0107] Based on the spatial distribution pattern of the sedimentation residual map, at least one key physical parameter in the spatiotemporal sedimentation potential field predictor or the physical information neural network wind field predictor is adjusted.

[0108] Analyze the spatial distribution patterns of the depositional residual map. For example, if the residual map shows large areas of negative values ​​in the internal region of the canopy (systematic insufficient deposition), it is determined that the current model's simulation of canopy penetration is too optimistic. Therefore, the air drag coefficient in the droplet dynamics model is reduced, which weakens the droplet penetration ability in subsequent trajectory predictions and makes it more consistent with reality.

[0109] Based on the unique absorption characteristics of water in the near-infrared band of the pesticide solution, a linear model of reflectance-deposition amount was established through laboratory calibration. After a preset deposition stabilization time following spraying, the near-infrared reflectance of the leaf surface was acquired using a multispectral imager to invert the actual deposition distribution map and generate a residual map by comparing it with the expected distribution.

[0110] Based on the embodiments provided in this application, traditional plant protection operations lack an effect feedback mechanism, making continuous optimization difficult. This application obtains leaf surface reflectance through multispectral imaging and inverts the actual deposition volume based on a calibration model, establishing a closed-loop channel from the physical world to the digital model. The deposition residual map not only reflects the absolute error of the spraying effect, but more importantly, it reveals the spatial distribution pattern of the error. This pattern contains rich information about the sources of system deviations, providing a precise decision-making basis for model parameter correction.

[0111] Furthermore, key physical parameters are adjusted based on the spatial distribution pattern of the sedimentary residual map, including:

[0112] If the deposition residual map shows a systematic negative bias in the internal region of the canopy, it is determined that the predicted value of the droplet penetration ability in the canopy in the spatiotemporal deposition potential field predictor is higher than the actual value, and the air drag coefficient used for its internal modeling is reduced accordingly.

[0113] If the deposition residual plot shows a systematic positive bias in the top canopy region, it is determined that the predicted value of the evaporation loss of droplets before reaching the leaf surface is lower than the actual value, and the evaporation rate parameter used in its internal modeling is increased accordingly.

[0114] A recursive least squares method with a forgetting factor is used to perform the adjustment process of key physical parameters, so as to achieve robust online updates of model parameters.

[0115] The physical correlation between system bias and model parameters is addressed: insufficient deposition within the canopy is associated with an underestimation of the drag coefficient, while excessive deposition at the canopy top is associated with an underestimation of the evaporation rate. A parameter update method with a forgetting factor is employed.

[0116]

[0117] in, and These represent the values ​​of a key physical parameter (such as air drag coefficient or evaporation rate) before and after this update. θ is the learning rate, a dimensionless gain coefficient, typically set to a small positive number (e.g., 0.01) to control the step size of each parameter update and prevent over-correction. J represents the sensitivity vector, whose elements are dimensionless. It characterizes how sensitive the model's predicted deposition rate is to changes in the parameter θ under the current operating environment. Higher sensitivity results in a larger correction magnitude. This represents the normalized residual, which is dimensionless. It is the average value of the residuals in the relevant region (such as the entire canopy interior) of the sedimentary residual map, after normalization, and represents the systematic bias of the current model prediction.

[0118] Based on the embodiments provided in this application, parameter orientation correction based on residual graph spatial patterns reflects a profound understanding of the physical mechanisms of droplet motion. The correlation between insufficient deposition within the canopy and the drag coefficient is based on the physical principle that "excessive drag leads to insufficient penetration"; the correlation between excessive deposition at the top of the canopy and the evaporation rate is based on the law of conservation of mass that "insufficient evaporation leads to excessive deposition". The recursive least squares method with a forgetting factor ensures that the system can adapt to environmental changes during continuous learning without over-adjusting due to single anomalous data.

[0119] Furthermore, the optimal combination obtained from the solution is used to drive the zoned controllable nozzle array to perform spraying operations, including:

[0120] The optimal droplet diameter obtained by the solution is combined with the liquid flow rate, along with its corresponding nozzle identifier, and packaged into a control instruction set.

[0121] The control command set is distributed to each corresponding nozzle controller in the partitioned controllable nozzle array via a real-time bus.

[0122] The nozzle controller synchronously and independently adjusts the oscillation frequency or pulse width modulation duty cycle of its piezoelectric actuator according to the received instructions to generate the target droplet diameter, and coordinates the power of the micro pump to output the target liquid flow rate.

[0123] The spraying operation is synchronized with the drone's flight speed to ensure that the total amount of pesticide sprayed per unit area meets agronomic requirements. The "synchronicity" and "independence" of control commands are guaranteed through the EtherCAT real-time industrial Ethernet bus. This bus features a precise distributed clock mechanism, ensuring that commands are delivered to all nozzle controllers simultaneously; at the same time, its message structure allows for independent commands from multiple nozzles, enabling independent parsing and execution under synchronized reception.

[0124] Based on the embodiments provided in this application, the transformation from optimization results to physical execution is the final crucial step in achieving precision plant protection. This solution ensures the synchronization and accuracy of multi-nozzle control through instruction packaging and real-time bus distribution. The design of synchronously adjusting atomization and flow parameters in the nozzle controller solves the problem of unstable droplet spectrum distribution caused by asynchronous atomization and flow control in traditional systems. This hardware-level collaborative control provides a reliable execution guarantee for the optimization results at the algorithm level.

[0125] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones is also provided. This electronic device may be... Figure 3 The terminal device or server shown. This embodiment uses this electronic device as an example of a server. Figure 3As shown, the electronic device includes a memory 402, a processor 404, and a transmission device 406. The memory 402 stores a computer program, and the processor 404 is configured to execute the steps in any of the above method embodiments through the computer program.

[0126] Optionally, in this embodiment, the aforementioned electronic device may be located in at least one of a plurality of network devices in a computer network.

[0127] Optionally, the transmission device 406 is used to receive or send data via a network. Specific examples of the network described above may include wired and wireless networks. In one example, the transmission device 406 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 406 is a Radio Frequency (RF) module used to communicate with the Internet wirelessly. Furthermore, the electronic device also includes a display 408 and a connection bus 410, which connects the various module components within the electronic device.

[0128] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for adjusting pesticide flow rate in precision plant protection operations using tobacco drones, characterized in that, include: The drone is equipped with a lidar and a multispectral imager to simultaneously collect three-dimensional point cloud data and multispectral image data of the tobacco canopy. Based on the three-dimensional point cloud data, a two-dimensional canopy oriented distance field is constructed in real time, and the canopy physiological information retrieved from the multispectral image data and the real-time environmental wind speed are integrated to generate a drug delivery priority map through a demand mapping function. Based on the liquid projection priority map, the instantaneous target coverage area corresponding to each nozzle in the UAV's partitioned controllable nozzle array is determined. Based on the instantaneous target coverage area and current environmental parameters, a forward trajectory simulation is performed using a droplet dynamics proxy model. With the goal of maximizing effective deposition in the target area, the optimal combination of droplet diameter and liquid flow rate for each nozzle is solved. The optimal combination obtained from the solution drives the partitioned controllable nozzle array to perform spraying operations.

2. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 1, characterized in that, Based on the three-dimensional point cloud data, a two-dimensional canopy oriented distance field is constructed in real time, and canopy physiological information retrieved from the multispectral image data is fused with real-time environmental wind speed. A drug delivery priority map is generated through a demand mapping function, including: The three-dimensional point cloud data is projected onto a two-dimensional grid parallel to the ground. For each grid cell, the directed distance from its center point to the nearest tobacco leaf surface along the vertical direction is calculated to construct the canopy directed distance field. When the center point of the grid cell is located in the outer space of the leaf, the directed distance value is positive; when it is located in the inner space of the leaf, the directed distance value is negative; and when it is located on the leaf surface, the directed distance value is zero. The demand weight at each location in the canopy oriented distance field is calculated using the demand mapping function; wherein, the calculation logic of the demand mapping function includes the synergistic canopy penetration rule, leaf attachment rule, and wind field pre-compensation rule. The leaf area index retrieved from the multispectral image data is used as a global adjustment factor and fused with the calculated demand weights to generate the drug projection priority map that reflects spatial structure, physiological state, and wind field environment.

3. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 2, characterized in that, The canopy penetration rule includes: for regions where the directed distance value in the canopy's directed distance field is negative, the required weight of each grid cell in that region is set to the exponential decay function value of the absolute value of the directed distance corresponding to that grid cell, so as to achieve key marking of the hidden areas inside the canopy; The leaf attachment rule includes: for the region in the canopy directional distance field that is within a preset neighborhood where the directional distance value is near zero, the demand weight of each grid cell in the region is set to a preset maximum demand weight value in order to lock the effective leaf attachment target area.

4. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 2, characterized in that, The wind field pre-compensation rule includes: based on the direction and magnitude of the real-time wind speed vector, positively strengthening the demand weight of the canopy area located upstream of the current wind direction, and negatively weakening the demand weight of the canopy area located downstream of the current wind direction, in order to pre-compensate for the spatial drift of the liquid. The execution of the wind field pre-compensation rules includes: The real-time wind speed vector is decomposed into two components on the horizontal plane: a first component parallel to the current flight direction of the UAV, and a second component perpendicular to the current flight direction of the UAV. For the second component, an asymmetric spatial weight compensation strategy is implemented, including: in the canopy region upstream of the current wind direction, the demand weight of the second component is increased according to the magnitude of the second component by a first preset ratio; in the canopy region downstream of the current wind direction, the demand weight of the second component is decreased according to the magnitude of the second component by a second preset ratio. Wherein, the absolute values ​​of the first preset ratio and the second preset ratio are both positively correlated with the magnitude of the second component; and the first preset ratio is a positive value, while the second preset ratio is a negative value.

5. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 2, characterized in that, The step of determining the instantaneous target coverage area for each nozzle in the UAV's partitioned controllable nozzle array based on the liquid projection priority map includes: Establish the mapping relationship between the physical position of each nozzle in the partitioned controllable nozzle array and the spatial coordinates of the two-dimensional grid; Based on the theoretical coverage of each nozzle, the corresponding candidate area is delineated on the liquid projection priority map; Perform dynamic region allocation optimization, including: for each nozzle, identify the spatial distribution of demand weights within its candidate region, and prioritize the allocation of the sub-region with the highest demand weight and spatial contiguousness as the instantaneous target coverage area of ​​that nozzle; When the candidate areas of multiple nozzles overlap, the coverage area is arbitrated and redistributed based on the demand weight of each grid cell within the overlapping area, either by comparing weight allocation or by proportional allocation.

6. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 5, characterized in that, Based on the instantaneous target coverage area and current environmental parameters, a forward trajectory simulation is performed using a droplet dynamics surrogate model. With the objective of maximizing effective deposition in the target area, the optimal combination of droplet diameter and liquid flow rate for each nozzle is solved, including: A physical information neural network wind field forecaster is constructed. The physical information neural network wind field forecaster takes the canopy directional distance field, UAV flight status and real-time environmental wind speed as inputs. By coupling the fluid dynamic physical constraints represented by the Navier-Stokes equations during its training and inference process, it deduces and outputs the three-dimensional dynamic disturbance wind field of the tobacco canopy region within a future preset time window. The forward trajectory simulation using a droplet dynamics proxy model is specifically implemented through a differentiable spatiotemporal deposition potential field predictor. The spatiotemporal deposition potential field predictor takes the droplet diameter, liquid flow rate, spatial position of the nozzle, and the three-dimensional dynamic disturbance wind field as inputs, and outputs the predicted deposition distribution map and the corresponding prediction confidence variance map of the nozzle in its instantaneous target coverage area. An optimization problem is established with the droplet diameter and liquid flow rate of all nozzles as decision variables, and the objective function is... Defined as: ； in, For the first Predicted sediment distribution map for each nozzle. This represents the predicted deposition distribution map for all nozzles. The total sediment distribution map obtained after spatial overlay; A map showing the priority of drug delivery; Map showing the total calculated sediment distribution Priority map of drug delivery Spatial Pearson correlation coefficient between them; Predict its confidence variance plot; This represents the variance plot of the prediction confidence for all nozzles. The total variance plot obtained after spatial overlay; The first weighting coefficient; This is the second weighting coefficient; The optimization problem is solved using a sequential quadratic programming algorithm. In the solution process, the value of the canopy directed distance field is used as the basis for the spatial penalty term. This includes: for a spatial location, the corresponding penalty intensity is positively correlated with the value of the canopy directed distance field at that location, so that the contribution of the predicted deposition amount in the air region where the value of the canopy directed distance field is greater than a preset positive threshold is attenuated in the objective function.

7. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 6, characterized in that, The method of solving the optimization problem using a sequential quadratic programming algorithm includes: Initialize the droplet diameter and liquid flow rate parameter combination for all nozzles; At the current parameter point, based on the gradient information provided by the spatiotemporal deposition potential field predictor, the canopy oriented distance field, and the physical information neural network wind field predictor, the gradient of the objective function and the constraint conditions is calculated, and a local quadratic programming subproblem is constructed accordingly. Solve the local quadratic programming subproblem to obtain the parameter correction direction and step size that optimize the objective of the local quadratic programming subproblem; The droplet diameter and liquid flow rate parameters of all nozzles are updated based on the correction direction and step size described above. Repeat the above steps iteratively until convergence; Throughout the iteration process, the Jacobian matrix of the global deposition distribution on each nozzle parameter is calculated and analyzed to characterize the parameter coupling effect between multiple nozzles. The Jacobian matrix is ​​then used to decouple the parameter coupling effect during the optimization process, thereby allocating the droplet diameter and liquid flow rate of each nozzle to achieve global optimal coverage when the nozzle coverage areas overlap.

8. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 6 or 7, characterized in that, The method also includes a closed-loop step of deposition effect feedback and online model parameter correction, specifically including: After the controllable nozzle array performs spraying operations and a preset deposition time has elapsed, the treated canopy leaf surface is scanned by a multispectral imager mounted on a drone to obtain leaf surface reflectance data in the near-infrared band. Based on the leaf surface reflectance data and the pre-established reflectance and deposition calibration model, the actual drug deposition distribution map was calculated. The actual drug solution deposition distribution map is spatially compared with the expected deposition distribution derived from the drug solution projection priority map to generate a deposition residual map. Based on the spatial distribution pattern of the sedimentation residual map, at least one key physical parameter in the spatiotemporal sedimentation potential field predictor or the physical information neural network wind field predictor is adjusted.

9. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 8, characterized in that, The key physical parameters are adjusted based on the spatial distribution pattern of the sedimentation residual map, including: If the deposition residual map shows a systematic negative bias in the internal region of the canopy, it is determined that the predicted value of the droplet penetration ability in the canopy in the spatiotemporal deposition potential field predictor is higher than the actual value, and the air drag coefficient used for its internal modeling is reduced accordingly. If the deposition residual plot shows a systematic positive deviation in the top canopy region, it is determined that the predicted value of the evaporation loss of droplets before reaching the leaf surface is lower than the actual value, and the evaporation rate parameter used in its internal modeling is increased accordingly. The adjustment process of key physical parameters is carried out using a recursive least squares method with a forgetting factor.

10. The method for adjusting the pesticide flow rate in precision plant protection operations using tobacco drones according to claim 1, characterized in that, The step of driving the partitioned controllable nozzle array to perform spraying operations based on the optimal combination obtained from the solution includes: The optimal droplet diameter obtained by the solution is combined with the liquid flow rate, along with its corresponding nozzle identifier, and packaged into a control instruction set. The control instruction set is distributed to each corresponding nozzle controller in the partitioned controllable nozzle array via a real-time bus. The nozzle controller synchronously and independently adjusts the oscillation frequency or pulse width modulation duty cycle of its piezoelectric actuator according to the received instructions to generate the target droplet diameter, and coordinates the power of the micro pump to output the target liquid flow rate.

Images