A Precision Application Method and System for Litchi Orchards Based on the Combined Use of Agricultural Drones and Adjuvants

Abstract

This invention discloses a method and system for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants, belonging to the field of agricultural plant protection technology. The method includes: constructing a dynamic topological deformation model characterizing the spatial heterogeneity index of the canopy; dividing the leaf canopy into virtual pesticide-receiving interfaces according to the pesticide response potential and generating a pulsed hovering phase sequence based on the model; calculating the droplet-wind interface interaction potential energy based on this sequence, and using this to introduce an asymmetric disturbance operator to reconstruct a spatially discrete droplet group adapted to the curvature of the virtual pesticide-receiving interface; using the spatial distribution parameters of the spatially discrete droplet group to modulate the charge state of the charged adjuvant in the pesticide solution, forming a targeted deposition field matching the curvature characteristics of the concealed leaf canopy unit; and using a hidden Markov chain to invert the deviation into boundary constraints for correcting the dynamic topological deformation model. This invention achieves deep synergy between agricultural drones and adjuvants in the spatiotemporal dimensions, improving the accuracy of pesticide application under complex canopy conditions in litchi orchards.

Description

Technical Field

[0001] This invention relates to the field of agricultural plant protection technology, and in particular to a method and system for precision pesticide application in litchi orchards based on the combined use of plant protection drones and adjuvants. Background Technology

[0002] In litchi orchard plant protection operations, due to the complex canopy structure and strong spatial heterogeneity of litchi trees, traditional plant protection drone spraying methods typically employ uniform or globally planned flight paths for continuous spraying. This makes it difficult to achieve precise targeted deposition on hidden leaf canopy units and leaf canopy surfaces with different curvatures within the canopy, resulting in poor droplet penetration, significant pesticide waste, and unstable control effects. While existing technologies have attempted canopy modeling based on remote sensing images or lidar, the constructed models are mostly static geometric models, failing to reflect the spatial differences in the canopy's pesticide response potential. Furthermore, they lack a closed-loop control mechanism that links the model with drone flight control, pesticide atomization, and adjuvant regulation. In addition, current spraying methods often only focus on macroscopic adjustments to atomized particle size or surface tension when mixing pesticides and adjuvants, failing to fully integrate the dynamic characteristics of the drone rotor downwash flow field and the geometric features of the canopy target interface to finely reshape droplet trajectories. Moreover, they lack non-contact online sensing and feedback correction capabilities for post-application deposition effects. Therefore, how to achieve dynamic matching and closed-loop optimization between the flight control of plant protection drones, the regulation of adjuvant properties, and the geometric characteristics of canopy targets, based on fully considering the spatial heterogeneity of litchi tree canopies, is an urgent problem to be solved. Summary of the Invention

[0003] This invention overcomes the shortcomings of the prior art and provides a method and system for precise pesticide application in litchi orchards based on the combined use of plant protection drones and adjuvants.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention discloses a method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants, comprising the following steps: A dynamic topological deformation model characterizing the spatial heterogeneity index of tree canopy was constructed based on multispectral images of litchi orchards and point clouds of single-tree lidar. Based on the dynamic topological deformation model, the inner canopy is divided into multiple virtual drug-receiving interfaces according to the drug-receiving response potential, and a pulsed hovering phase sequence matching the spatiotemporal phase of each virtual drug-receiving interface is generated. Based on the pulsed hovering phase sequence, the potential energy of the droplet-wind field interface interaction between the UAV rotor downwash flow field and the droplet group after mixing with the drug and additives is calculated. Using the potential energy of the droplet-wind field interface as a constraint, an asymmetric turbulence operator is introduced to reconstruct a spatial discrete droplet swarm that matches the curvature of the virtual drug-receiving interface. By utilizing the spatial distribution parameters of the spatially discrete droplet group, the charge state of the charged additive in the drug solution is modulated through magnetohydrodynamics to form a targeted deposition field that matches the curvature characteristics of each concealed leaf canopy unit in the dynamic topological deformation model. Acoustic echo signals from different depths within the canopy after drug application were collected. The actual coverage boundary of the targeted deposition field was analyzed, and the deviation between the actual coverage boundary and the virtual drug-receiving interface was inverted into the boundary constraints of the modified dynamic topological deformation model based on a hidden Markov chain.

[0005] Furthermore, a dynamic topological deformation model characterizing the spatial heterogeneity index of the tree canopy is constructed based on multispectral images of the litchi orchard and point clouds of single-tree lidar, specifically as follows: Acquire multispectral images of litchi orchards and point clouds from single-tree lidar, and register the two to the same spatial coordinate system to form a spatially aligned multimodal data volume; From the spatially aligned multimodal data volume, a spatial heterogeneity index is extracted to characterize the heterogeneity of leaf canopy distribution in the vertical and horizontal directions within each individual tree canopy. The spatial heterogeneity index is obtained by weighted fusion of the texture gradient reflecting chlorophyll content in the multispectral image and the structural entropy reflecting the gaps between branches and leaves in the lidar point cloud. Using the spatial heterogeneity index as the initial boundary condition, an initial mesh model covering the entire canopy spatial domain is constructed. Based on the single-tree skeleton topology extracted from the lidar point cloud, Laplacian smoothing is applied to the initial mesh model to generate a static canopy model with continuous surface features. Using the spatial heterogeneity index as the deformation driving field, non-uniform normal perturbations are applied to each grid node of the static canopy model to form a dynamic topological deformation model.

[0006] The dynamic topological deformation model ensures that the displacement of the grid nodes is consistent with the spatial gradient direction of the spatial heterogeneity index by iteratively minimizing the deformation energy function.

[0007] Furthermore, based on the dynamic topological deformation model, the inner canopy is divided into multiple virtual drug-receiving interfaces according to the drug-receiving response potential, and a pulsed hovering phase sequence matching the spatiotemporal phase of each virtual drug-receiving interface is generated, specifically: Based on the dynamic topological deformation model, the spatial coordinates, spatial heterogeneity index, and normal curvature of each grid node are extracted, and the canopy foliage is discretized into multiple foliage micro-units with independent geometric features and heterogeneous properties according to the topological connection relationship between each grid node. Based on the spatial heterogeneity index and normal curvature of each foliation microunit, the drug-receiving response potential energy of each foliation microunit to droplet deposition is determined. The drug-receiving response potential energy is positively correlated with the spatial heterogeneity index and negatively correlated with the normal curvature. Adjacent leaf micro-units with drug-receiving response potential energy higher than a preset threshold are spatially clustered to form multiple virtual drug-receiving interfaces, and each virtual drug-receiving interface is assigned a spatiotemporal phase identifier that represents its spatial position and orientation in three-dimensional space. Based on the spatiotemporal phase identifiers of each virtual drug receiving interface, the hovering point coordinates and hovering time windows required by the UAV when traversing each virtual drug receiving interface are calculated, and a pulsed hovering phase sequence corresponding to each spatiotemporal phase identifier is generated. Each pulsed hovering phase sequence includes the time when the UAV arrives at the hovering point and the duration of pulsed spraying at the hovering point.

[0008] Furthermore, based on the pulsed hovering phase sequence, the droplet-wind interface interaction potential energy of the UAV rotor downwash flow field and the droplet swarm after mixing with the drug and additives is calculated, specifically as follows: Based on the pulsed hovering phase sequence, the rotor speed and blade pitch angle corresponding to each hovering point are extracted, and the three-dimensional velocity vector distribution and turbulence intensity distribution of the downwash flow field generated by the UAV rotor at each hovering point are calculated to form dynamic rotor aerodynamic parameters that correspond one-to-one with the hovering point. The physical property parameters of the mixture of drug solution and adjuvant are obtained, including the surface tension and viscosity of the mixture and the dynamic surface tension relaxation time caused by the adjuvant. The dynamic rotor aerodynamic parameters and the physical property parameters are input into the droplet breakup and motion model to solve the particle size distribution and initial velocity vector of the droplet swarm formed under the action of the downwash flow field at different hovering points after being ejected from the nozzle, and the initial dynamic state of the droplet swarm associated with the hovering point is generated. The initial velocity vector of each droplet in the initial dynamic state of the droplet swarm is coupled point by point with the three-dimensional velocity vector distribution of the downwash flow field to determine the momentum exchange intensity between the droplet surface and the flow field when each droplet moves in the downwash flow field. Based on the dynamic surface tension relaxation time correction of the droplet deformation response hysteresis effect during the motion process, the droplet-wind field interface interaction potential energy characterizing the energy transfer efficiency between the droplet swarm and the downwash flow field is generated.

[0009] Furthermore, using the potential energy of the droplet-wind field interface interaction as a constraint, an asymmetric disturbance operator is introduced to reconstruct a spatially discrete droplet swarm that adapts to the curvature of the virtual drug-receiving interface, specifically: Based on the potential energy of the droplet-wind field interface interaction, the spatial distribution gradient of the energy transfer efficiency between the droplet group and the downwash flow field at each hovering point is extracted, and the spatial distribution gradient is used as the weighting coefficient to construct the initial configuration of the asymmetric turbulence operator. The asymmetric turbulence operator consists of multiple sets of turbulence vector fields that are non-uniformly distributed along the circumferential direction of the rotor rotation. Obtain the curvature distribution data of the virtual drug receiving interface, map the principal curvature direction and curvature radius of each virtual drug receiving interface to the turbulence vector field of the asymmetric turbulence operator, and adjust the amplitude and direction of each turbulence vector field to make the asymmetric turbulence operator and the surface geometric features of the virtual drug receiving interface form a spatial correspondence. The asymmetric turbulence operator is superimposed on the downwash flow field to apply an asymmetric deflection effect to the trajectory of the droplet group in the downwash flow field, so that the droplet group is spatially deaggregated into multiple mutually separated droplet clusters that conform to the curved contours of each virtual drug-receiving interface. Based on the degree of conformity between each droplet cluster and the corresponding virtual drug-receiving interface, the spatial distribution parameters of the asymmetric perturbation operator are iteratively optimized until the deviation between the spatial distribution morphology of each droplet cluster and the curvature characteristics of the virtual drug-receiving interface converges, forming a spatially discrete droplet swarm.

[0010] Furthermore, utilizing the spatial distribution parameters of the spatially discrete droplet swarm, the charge state of the charged additive in the drug solution is modulated via magnetohydrodynamics to form a targeted deposition field that matches the curvature characteristics of each concealed foliage unit in the dynamic topological deformation model. Specifically: Based on the spatial distribution parameters of the spatial discrete droplet group, the spatial coordinates, spatial orientation, and relative positional relationship between each droplet cluster and the corresponding virtual drug receiving interface in three-dimensional space are extracted. The spatial distribution parameters are then mapped to the modulation space of the charged adjuvant at the nozzle to generate a spatial charge density distribution corresponding to each droplet cluster. Obtain the curvature characteristics of each concealed leaf curtain unit in the dynamic topological deformation model, including the spatial position, normal curvature, and concealment angle formed by the concealed leaf curtain unit being blocked by adjacent leaf curtains; Establish a mapping relationship between the space charge density distribution and the curvature characteristics, and calculate the target charge state required for the fog droplets to overcome flow field disturbances and spread along the curvature surface during their flight toward the corresponding concealed leaf curtain unit, based on the mapping relationship. The target charge state includes charge magnitude and charge polarity. Based on the target charge state, the electric field strength and direction of the charged additive applied to the nozzle are dynamically adjusted so that the charged additive forms a charge state spatial modulation pattern in the liquid that is compatible with the spatial charge density distribution. The liquid medicine with the aforementioned charge state spatial modulation pattern is sprayed out in a sequence corresponding to the droplet clusters, so that each droplet cluster maintains a charge state consistent with the target charge state during flight, and spreads directionally along the curved surface with the aid of charge-induced mirror force when it reaches the concealed leaf canopy unit, forming a targeted deposition field.

[0011] Furthermore, acoustic echo signals from different depths within the canopy after drug application were collected to analyze the actual coverage boundary of the targeted deposition field. Based on a hidden Markov chain, the deviation between the actual coverage boundary and the virtual drug-receiving interface was inverted into boundary constraints for correcting the dynamic topological deformation model. Specifically: Based on the preset spatial distribution characteristics of the targeted deposition field, multiple acoustic signature collection points located at different depths within the canopy are set. Broadband acoustic pulses are emitted at each acoustic signature collection point, and echo signals reflected by the leaf canopy surface and scattered by the droplet layer are received, forming echo timing characteristics that include time delay and spectral attenuation information. The acoustic impedance variation characteristics corresponding to droplet deposition at each depth location are extracted from the echo timing characteristics, and the deposition boundary profile formed by the targeted deposition field in actual operation is analyzed based on the spatial continuity of the acoustic impedance variation characteristics, which serves as the actual coverage boundary. The actual coverage boundary and the spatial boundary of the virtual drug-receiving interface are compared point by point, and the deviation sequence between them in terms of spatial position, spatial orientation and boundary curvature is calculated. The deviation sequence is then used as the observation state sequence of the hidden Markov chain. Based on the forward-backward algorithm of hidden Markov chains, the most likely state transition path leading to the deviation is inverted from the observed state sequence. The state transition path corresponds to the distribution of the deviation of each grid node in the dynamic topological deformation model under the boundary constraints. The deviation distribution is superimposed as a boundary constraint on the initial boundary of the dynamic topological deformation model to generate the corrected boundary constraint.

[0012] The acoustic fingerprint collection points are non-uniformly distributed according to the spatial distribution density of the virtual drug-receiving interface and the concealment angle of each concealed leaf curtain unit, so that the spatial sampling density of the acoustic fingerprint collection points is higher in areas with larger concealment angles.

[0013] The second aspect of this invention discloses a precision pesticide application system for litchi orchards based on the combined use of plant protection drones and adjuvants. The system includes a memory and a processor. The memory stores a program for a precision pesticide application method for litchi orchards based on the combined use of plant protection drones and adjuvants. When the program for the precision pesticide application method for litchi orchards based on the combined use of plant protection drones and adjuvants is executed by the processor, the steps of the precision pesticide application method for litchi orchards based on the combined use of plant protection drones and adjuvants as described in any one of the claims are implemented.

[0014] This invention addresses the technical deficiencies in the prior art and offers the following advantages: Through a comprehensive chain of "target modeling - flight path planning - flow field modulation - charge control - effect feedback," this invention achieves deep spatiotemporal synergy between agricultural drones and adjuvants, effectively improving the accuracy of pesticide application under complex canopy conditions in litchi orchards. It enables pesticide droplets to actively adapt to the spatial heterogeneity of the canopy and the geometric characteristics of the leaf surface, enhancing the penetration ability and deposition uniformity of droplets within the canopy and in concealed areas. Simultaneously, a closed-loop feedback mechanism continuously inverts and corrects the model's boundary conditions based on the application effect, progressively optimizing operational accuracy over multiple application processes, providing an adaptive and intelligent solution for plant protection operations in litchi orchards. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0016] Figure 1 A flowchart of a precision pesticide application method for litchi orchards based on the combined use of agricultural drones and adjuvants; Figure 2 This is a framework diagram of a precision pesticide application system for litchi orchards based on the combined use of agricultural drones and adjuvants. Detailed Implementation

[0017] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0019] like Figure 1 As shown, the first aspect of this invention discloses a method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants, comprising the following steps: A dynamic topological deformation model characterizing the spatial heterogeneity index of tree canopy was constructed based on multispectral images of litchi orchards and point clouds of single-tree lidar. Based on the dynamic topological deformation model, the inner canopy is divided into multiple virtual drug-receiving interfaces according to the drug-receiving response potential, and a pulsed hovering phase sequence matching the spatiotemporal phase of each virtual drug-receiving interface is generated. Based on the pulsed hovering phase sequence, the potential energy of the droplet-wind field interface interaction between the UAV rotor downwash flow field and the droplet group after mixing with the drug and additives is calculated. Using the potential energy of the droplet-wind field interface as a constraint, an asymmetric turbulence operator is introduced to reconstruct a spatial discrete droplet swarm that matches the curvature of the virtual drug-receiving interface. By utilizing the spatial distribution parameters of the spatially discrete droplet group, the charge state of the charged additive in the drug solution is modulated through magnetohydrodynamics to form a targeted deposition field that matches the curvature characteristics of each concealed leaf canopy unit in the dynamic topological deformation model. Acoustic echo signals from different depths within the canopy after drug application were collected. The actual coverage boundary of the targeted deposition field was analyzed, and the deviation between the actual coverage boundary and the virtual drug-receiving interface was inverted into the boundary constraints of the modified dynamic topological deformation model based on a hidden Markov chain.

[0020] Furthermore, a dynamic topological deformation model characterizing the spatial heterogeneity index of the tree canopy is constructed based on multispectral images of the litchi orchard and point clouds of single-tree lidar, specifically as follows: Acquire multispectral images of litchi orchards and point clouds from single-tree lidar, and register the two to the same spatial coordinate system to form a spatially aligned multimodal data volume; From the spatially aligned multimodal data volume, a spatial heterogeneity index is extracted to characterize the heterogeneity of leaf canopy distribution in the vertical and horizontal directions within each individual tree canopy. The spatial heterogeneity index is obtained by weighted fusion of the texture gradient reflecting chlorophyll content in the multispectral image and the structural entropy reflecting the gaps between branches and leaves in the lidar point cloud. Using the spatial heterogeneity index as the initial boundary condition, an initial mesh model covering the entire canopy spatial domain is constructed. Based on the single-tree skeleton topology extracted from the lidar point cloud, Laplacian smoothing is applied to the initial mesh model to generate a static canopy model with continuous surface features. Using the spatial heterogeneity index as the deformation driving field, non-uniform normal perturbations are applied to each grid node of the static canopy model to form a dynamic topological deformation model.

[0021] The dynamic topological deformation model ensures that the displacement of the grid nodes is consistent with the spatial gradient direction of the spatial heterogeneity index by iteratively minimizing the deformation energy function.

[0022] In practical implementation, to construct a dynamic topological deformation model that accurately reflects the structural differences of the litchi tree canopy, the collected multispectral images of the litchi orchard were first spatially registered with the point clouds of individual trees from a lidar system, forming a multimodal data volume. Based on this, a spatial heterogeneity index was extracted from the data volume. This index is obtained by weighted fusion of the texture gradient reflecting chlorophyll content in the multispectral images and the structural entropy reflecting the gaps between branches and leaves in the lidar point cloud. Its physical significance lies in quantifying the heterogeneity of the leaf canopy distribution in the vertical and horizontal directions within the canopy. When constructing the initial mesh model, the spatial heterogeneity index was used as the initial boundary condition, and Laplacian smoothing was applied to the initial mesh based on the topological structure of the individual tree skeleton extracted from the lidar point cloud. This process ensured both the topological consistency between the mesh model and the real canopy skeleton and avoided surface discontinuities caused by point cloud discrete noise. Finally, the spatial heterogeneity index is used as the deformation driving field to apply non-uniform normal perturbation to each grid node of the static canopy model. By iteratively minimizing the deformation energy function, the displacement of each node is kept consistent with the spatial gradient direction of the spatial heterogeneity index. This ensures that the model deformation strictly follows the distribution law of canopy heterogeneity, ultimately forming a dynamic topological deformation model. The constructed model not only reflects the static geometry of the canopy but also internalizes the spatial differences in the leaf canopy's drug-receiving response potential into the model's computable deformation characteristics. This provides a dynamically adaptable target basis for the division of virtual drug-receiving interfaces and precise drug delivery.

[0023] Furthermore, based on the dynamic topological deformation model, the inner canopy is divided into multiple virtual drug-receiving interfaces according to the drug-receiving response potential, and a pulsed hovering phase sequence matching the spatiotemporal phase of each virtual drug-receiving interface is generated, specifically: Based on the dynamic topological deformation model, the spatial coordinates, spatial heterogeneity index, and normal curvature of each grid node are extracted, and the canopy foliage is discretized into multiple foliage micro-units with independent geometric features and heterogeneous properties according to the topological connection relationship between each grid node. Based on the spatial heterogeneity index and normal curvature of each foliation microunit, the drug-receiving response potential energy of each foliation microunit to droplet deposition is determined. The drug-receiving response potential energy is positively correlated with the spatial heterogeneity index and negatively correlated with the normal curvature. Adjacent leaf micro-units with drug-receiving response potential energy higher than a preset threshold are spatially clustered to form multiple virtual drug-receiving interfaces, and each virtual drug-receiving interface is assigned a spatiotemporal phase identifier that represents its spatial position and orientation in three-dimensional space. Based on the spatiotemporal phase identifiers of each virtual drug receiving interface, the hovering point coordinates and hovering time windows required by the UAV when traversing each virtual drug receiving interface are calculated, and a pulsed hovering phase sequence corresponding to each spatiotemporal phase identifier is generated. Each pulsed hovering phase sequence includes the time when the UAV arrives at the hovering point and the duration of pulsed spraying at the hovering point.

[0024] It should be noted that each micro-unit carries independent information on its geometric location, degree of heterogeneity, and surface curvature. For each canopy micro-unit, the drug reception response potential energy is calculated based on its spatial heterogeneity index and normal curvature. Specifically, the spatial heterogeneity index reflects the heterogeneity of the canopy distribution in the vertical and horizontal directions within the micro-unit's region. A higher index indicates a more complex canopy structure and greater difficulty for droplet penetration, thus requiring a higher drug reception priority. The normal curvature reflects the degree of curvature of the canopy surface. A greater curvature means a more convex or concave surface, making droplets more prone to slippage or collision loss during deposition, thus reducing the drug reception priority. Based on this, the drug reception response potential energy is defined as the weighted combination of the normalized values ​​of the spatial heterogeneity index and the normalized values ​​of the normal curvature. The spatial heterogeneity index is positively correlated with the potential energy, while the normal curvature is negatively correlated. By setting different weighting coefficients, the proportion of their contribution to the potential energy can be adjusted according to the actual canopy characteristics of the litchi orchard. For example, for litchi tree canopies with dense foliage and clearly interwoven branches, the weight of the spatial heterogeneity index can be appropriately increased to allow micro-units in structurally complex regions to obtain higher potential energy values.

[0025] Then, the calculated drug-receiving response potential energy of each leaf micro-unit is normalized. Adjacent micro-units with normalized potential energy values ​​higher than a preset threshold are spatially clustered. This preset threshold can be flexibly set according to actual operational requirements. For example, when the normalized potential energy value is greater than 0.6, the micro-unit is included in the cluster candidate; if the difference between this value and the potential energy value of adjacent micro-units does not exceed 0.15, they are merged into the same virtual drug-receiving interface. After clustering, a spatiotemporal phase identifier is assigned to each virtual drug-receiving interface. This identifier includes the centroid coordinates of the interface in three-dimensional space, the interface normal orientation (obtained by a weighted average of the normal curvatures of each micro-unit within the cluster), and the expected arrival order of the interface in the UAV operation sequence. Its function is to abstract the irregular leaf canopy set into executable operational units with clear spatiotemporal attributes. Based on the spatiotemporal phase identifiers of each virtual drug-receiving interface, the hovering point coordinates and hovering time window required for the UAV to traverse each interface are calculated. The hovering point coordinates are determined as follows: using the centroid coordinates of the interface as a reference point, a preset safe operating distance is offset in the opposite direction of the interface's normal orientation (i.e., towards the direction the drone is approaching). This distance is 1.2 to 1.5 times the rotor diameter to ensure that the rotor downwash flow field can effectively cover the entire interface area and avoid collisions between the drone and the canopy. The hovering time window is determined by considering three factors: first, the projected area of ​​the interface in the normal orientation, which reflects the actual target area that needs to be covered by the droplets; second, the spray flow rate of the drone at the hovering point, which is determined by the nozzle type, spray pressure, and adjuvant properties; and third, the desired deposition amount per unit area, which is preset based on the requirements for litchi pest and disease control and the type of adjuvant. Dividing the product of the projected area and the desired deposition amount by the spray flow rate estimates the duration required for pulsed spraying at the hovering point. Meanwhile, by combining the spatial distribution order of each virtual drug receiving interface, the spatial distance between interfaces, and the drone's cruising speed, the time when the drone arrives at each hovering point is calculated in sequence, thus forming a pulse-type hovering phase sequence that corresponds one-to-one with each spatiotemporal phase marker.

[0026] Furthermore, based on the pulsed hovering phase sequence, the droplet-wind interface interaction potential energy of the UAV rotor downwash flow field and the droplet swarm after mixing with the drug and additives is calculated, specifically as follows: Based on the pulsed hovering phase sequence, the rotor speed and blade pitch angle corresponding to each hovering point are extracted, and the three-dimensional velocity vector distribution and turbulence intensity distribution of the downwash flow field generated by the UAV rotor at each hovering point are calculated to form dynamic rotor aerodynamic parameters that correspond one-to-one with the hovering point. The physical property parameters of the mixture of drug solution and adjuvant are obtained, including the surface tension and viscosity of the mixture and the dynamic surface tension relaxation time caused by the adjuvant. The dynamic rotor aerodynamic parameters and the physical property parameters are input into the droplet breakup and motion model to solve the particle size distribution and initial velocity vector of the droplet swarm formed under the action of the downwash flow field at different hovering points after being ejected from the nozzle, and the initial dynamic state of the droplet swarm associated with the hovering point is generated. The initial velocity vector of each droplet in the initial dynamic state of the droplet swarm is coupled point by point with the three-dimensional velocity vector distribution of the downwash flow field to determine the momentum exchange intensity between the droplet surface and the flow field when each droplet moves in the downwash flow field. Based on the dynamic surface tension relaxation time correction of the droplet deformation response hysteresis effect during the motion process, the droplet-wind field interface interaction potential energy characterizing the energy transfer efficiency between the droplet swarm and the downwash flow field is generated.

[0027] It should be noted that, in order to quantify the energy coupling relationship between the downwash flow field of the UAV rotor and the droplet swarm formed after mixing with the liquid and additives, the rotor speed and blade pitch angle are extracted from the UAV flight control parameters corresponding to each hovering point based on the pulsed hovering phase sequence to determine the aerodynamic load generated by the rotor. Then, using eddy current theory or computational fluid dynamics methods in rotor aerodynamics, the three-dimensional velocity vector distribution and turbulence intensity distribution of the downwash flow field at each hovering point are calculated. The three-dimensional velocity vector distribution describes the magnitude and direction of the flow velocity at each point in space, while the turbulence intensity distribution reflects the influence of the degree of flow field pulsation on the stability of droplet motion, thereby forming dynamic rotor aerodynamic parameters corresponding one-to-one with the hovering point.

[0028] To obtain the initial dynamic state of a droplet swarm under a specific downwash flow field, this embodiment introduces a droplet breakup and motion model for calculation. The model takes nozzle structural parameters (such as nozzle diameter, number of nozzles, and spray angle), spray pressure, and the aforementioned physical properties as input. First, it analyzes the initial breakup process of the liquid film or liquid column at the nozzle exit based on jet breakup theory or linear instability analysis methods to obtain the initial particle size distribution range. Then, it uses the three-dimensional velocity vector of the downwash flow field as the background flow field, superimposed with the initial jet velocity generated by the nozzle spray. By solving the force balance in the droplet motion equations (including aerodynamic drag, gravity, pressure gradient force, etc.), it obtains the initial velocity vector of droplets in each size range at the instant they leave the nozzle. The magnitude and direction of the initial velocity vector comprehensively reflect the combined effect of the spray pressure and the downwash flow field. For each hovering point, the above calculation process is repeated to generate the droplet swarm particle size distribution and initial velocity vector associated with the downwash flow field characteristics at that point, constituting the initial dynamic state of the droplet swarm.

[0029] The initial velocity vector of each droplet in the initial dynamic state is spatially coupled point by point with the three-dimensional velocity vector distribution of the downwash flow field. Specifically, the difference between the velocity vector of the flow field at the location of the droplet and the velocity vector of the droplet itself is taken as the relative velocity. The momentum exchange intensity between the droplet surface and the flow field can be determined by combining the droplet size, gas-liquid density ratio and drag coefficient.

[0030] Furthermore, considering the dynamic relaxation characteristics of droplet surface tension after the introduction of additives, the change in surface tension during droplet deformation under flow field shear is not instantaneous but exhibits a certain lag. This embodiment modifies the momentum exchange intensity based on the dynamic surface tension relaxation time, dynamically updating the surface tension value during droplet deformation. This influences the droplet breakup critical conditions and motion stability, ultimately generating the droplet-wind field interface interaction potential energy characterizing the energy transfer efficiency between the droplet swarm and the downwash flow field. This interaction potential energy encompasses both the transport effect of the flow field on the droplet swarm and the dynamic adjustment effect of additive properties on droplet deformation and breakup behavior, providing a clearly defined physical constraint for the introduction of the asymmetric disturbance operator.

[0031] Furthermore, using the potential energy of the droplet-wind field interface interaction as a constraint, an asymmetric disturbance operator is introduced to reconstruct a spatially discrete droplet swarm that adapts to the curvature of the virtual drug-receiving interface, specifically: Based on the potential energy of the droplet-wind field interface interaction, the spatial distribution gradient of the energy transfer efficiency between the droplet group and the downwash flow field at each hovering point is extracted, and the spatial distribution gradient is used as the weighting coefficient to construct the initial configuration of the asymmetric turbulence operator. The asymmetric turbulence operator consists of multiple sets of turbulence vector fields that are non-uniformly distributed along the circumferential direction of the rotor rotation. Obtain the curvature distribution data of the virtual drug receiving interface, map the principal curvature direction and curvature radius of each virtual drug receiving interface to the turbulence vector field of the asymmetric turbulence operator, and adjust the amplitude and direction of each turbulence vector field to make the asymmetric turbulence operator and the surface geometric features of the virtual drug receiving interface form a spatial correspondence. The asymmetric turbulence operator is superimposed on the downwash flow field to apply an asymmetric deflection effect to the trajectory of the droplet group in the downwash flow field, so that the droplet group is spatially deaggregated into multiple mutually separated droplet clusters that conform to the curved contours of each virtual drug-receiving interface. Based on the degree of conformity between each droplet cluster and the corresponding virtual drug-receiving interface, the spatial distribution parameters of the asymmetric perturbation operator are iteratively optimized until the deviation between the spatial distribution morphology of each droplet cluster and the curvature characteristics of the virtual drug-receiving interface converges, forming a spatially discrete droplet swarm.

[0032] It should be noted that the spatial distribution gradient (reflecting the difference in transport efficiency of the flow field to the droplets in different spatial directions at the current hovering point) is first extracted from the previously calculated potential energy of the droplet-wind interface interaction. Using this gradient as weights, a set of asymmetric turbulence vector fields is constructed. This set of vector fields is non-uniformly distributed along the rotor rotation circumference; that is, depending on the energy transfer efficiency, the turbulence intensity is greater at some circumferential angles and less at others. Then, the curvature distribution data of each previously defined virtual drug-receiving interface is read, focusing on the principal curvature direction and radius of curvature of each interface. These two parameters together describe the bending orientation and degree of the blade surface. The curvature features are mapped to the turbulence vector field. Specifically, the principal curvature direction is used as the dominant direction of the corresponding turbulence vector field, and the radius of curvature is used as a reference for adjusting the range of action of the turbulence vector field. By adjusting the amplitude and direction of each turbulence vector field, the spatial configuration of the entire asymmetric turbulence operator is made to correspond to the surface geometry of the virtual drug-receiving interface.

[0033] When the curvature-adapted turbulence operator is superimposed on the downwash flow field, it is equivalent to adding an additional turbulence component that matches the curvature characteristics of the leaf canopy to the original downwash airflow. As the droplet swarm moves in this composite flow field, its spatial distribution changes. The originally continuous droplet swarm gradually separates into several independent droplet clusters, each of which gradually aligns with the surface contour of its corresponding virtual drug-receiving interface during its movement. Finally, the matching effect is verified: by calculating the deviation between the actual spatial distribution of each droplet cluster and the curvature of the corresponding virtual drug-receiving interface, if the deviation exceeds a preset deviation threshold, the spatial distribution parameters of the turbulence vector field are adjusted according to the direction and magnitude of the deviation, and a new droplet cluster distribution is regenerated. This process is repeated until the deviation converges to an acceptable range. Through this iterative optimization, the resulting spatially discrete droplet swarm is spatially separated, and the spatial distribution morphology of each droplet cluster matches the curvature characteristics of its corresponding virtual drug-receiving interface.

[0034] Furthermore, utilizing the spatial distribution parameters of the spatially discrete droplet swarm, the charge state of the charged additive in the drug solution is modulated via magnetohydrodynamics to form a targeted deposition field that matches the curvature characteristics of each concealed foliage unit in the dynamic topological deformation model. Specifically: Based on the spatial distribution parameters of the spatial discrete droplet group, the spatial coordinates, spatial orientation, and relative positional relationship between each droplet cluster and the corresponding virtual drug receiving interface in three-dimensional space are extracted. The spatial distribution parameters are then mapped to the modulation space of the charged adjuvant at the nozzle to generate a spatial charge density distribution corresponding to each droplet cluster. Obtain the curvature characteristics of each concealed leaf curtain unit in the dynamic topological deformation model, including the spatial position, normal curvature, and concealment angle formed by the concealed leaf curtain unit being blocked by adjacent leaf curtains; Establish a mapping relationship between the space charge density distribution and the curvature characteristics, and calculate the target charge state required for the fog droplets to overcome flow field disturbances and spread along the curvature surface during their flight toward the corresponding concealed leaf curtain unit, based on the mapping relationship. The target charge state includes charge magnitude and charge polarity. Based on the target charge state, the electric field strength and direction of the charged additive applied to the nozzle are dynamically adjusted so that the charged additive forms a charge state spatial modulation pattern in the liquid that is compatible with the spatial charge density distribution. The liquid medicine with the aforementioned charge state spatial modulation pattern is sprayed out in a sequence corresponding to the droplet clusters, so that each droplet cluster maintains a charge state consistent with the target charge state during flight, and spreads directionally along the curved surface with the aid of charge-induced mirror force when it reaches the concealed leaf canopy unit, forming a targeted deposition field.

[0035] It should be noted that a practical challenge in achieving targeted deposition of droplets into concealed canopy units within the canopy is that even if the spatial distribution of droplet clusters roughly matches the canopy surface profile, the downwash flow field often deflects and attenuates as it passes through the obstructed areas because the concealed canopy units are often blocked by upper or outer branches and leaves. This causes droplets to easily deviate from their intended trajectory as they approach the target, and they also struggle to overcome the rebound or slippage caused by the curvature of the canopy surface after reaching it. To address this issue, this embodiment utilizes magnetohydrodynamics to finely modulate the charge state of the charged agent, enabling the droplets to carry charge information matching the target canopy surface features during flight. This allows electrostatic forces to guide the droplets to complete the precise deposition of the "final leg" of the process. Specifically, the spatial distribution parameters of each droplet cluster are extracted from the previously formed spatially discrete droplet swarm, including the spatial coordinates, orientation, and relative position of each cluster to the corresponding virtual drug-receiving interface. These parameters determine the approximate flight path and spatial attitude of each droplet cluster after it is ejected from the nozzle. Then, these spatial distribution parameters are mapped to the modulation space of the charged additive at the nozzle, generating a set of spatial charge density distributions that correspond one-to-one with each droplet cluster. In other words, a correspondence is established: how much charge each droplet cluster should carry and how the charge is distributed within it are all determined by the cluster's position, orientation, and relative relationship with the target interface in three-dimensional space.

[0036] Establishing a mapping relationship between the space charge density distribution and the aforementioned curvature characteristics is a key aspect of this embodiment. The physical logic of this mapping relationship is as follows: For foliage units with a large concealment angle, droplets need to overcome stronger flow field disturbances and more complex flow paths during their approach. Therefore, it is necessary to assign a higher charge level to the droplet subclusters to enhance the electrostatic force's resistance to flow field disturbances. For foliage surfaces with a large normal curvature, droplets tend to slide along the curved surface after deposition. Therefore, it is necessary to assign a specific charge polarity to the droplet subclusters to generate a mirror force that attracts each other between the droplets and the foliage surface. This mirror force gradually increases as the droplets approach the foliage surface, which can "pull" the droplets towards and "press" them onto the curved surface, thereby suppressing slippage. Based on this logic, a positive correlation is established between the concealment angle of each concealed leaf unit and the charge magnitude, and a correlation is established between the direction of the normal curvature and the charge polarity. The influence of the spatial position of the leaf canopy on the spatial distribution of charge is also considered. In this way, the target charge state required by each droplet cluster in the process of flying to the corresponding concealed leaf unit can be calculated, that is, how much charge the cluster should carry, whether it is positive or negative.

[0037] After obtaining the target charge state, the electric field strength and direction applied to the charged agent at the nozzle are dynamically adjusted. Since the charge state of the charged agent at the nozzle can be controlled in real time via the electric field, a spatial modulation pattern of the charge state, adapted to the spatial charge density distribution, is formed inside the nozzle according to the spatial distribution requirements of the target charge state. It is important to note that this modulation is not uniform but changes over time: when the nozzle is ready to spray a specific droplet cluster, the electric field strength and direction are adjusted to match the target charge state corresponding to that cluster, ensuring that the sprayed liquid carries the required charge level and polarity the instant it atomizes to form the cluster. Each droplet cluster is sprayed sequentially according to a one-to-one correspondence with the virtual drug-receiving interface, and each cluster maintains a charge state consistent with the target charge state throughout its flight. When charged droplet clusters approach their corresponding concealed foliage units, the charge-induced mirror force between the foliage surface and the charged droplets causes them to be pulled by an electrostatic force pointing towards the curved surface before reaching the foliage surface. This helps compensate for trajectory deviations caused by flow field disturbances. After contacting the foliage surface, the mirror force continues to act, spreading the droplets along the curvature direction and reducing liquid loss due to surface curvature. In this way, the resulting targeted deposition field not only matches the spatial distribution of the concealed foliage units but also achieves directional spreading along the curved surface in terms of deposition morphology, thus solving the problem of effective droplet deposition in concealed areas.

[0038] Furthermore, acoustic echo signals from different depths within the canopy after drug application were collected to analyze the actual coverage boundary of the targeted deposition field. Based on a hidden Markov chain, the deviation between the actual coverage boundary and the virtual drug-receiving interface was inverted into boundary constraints for correcting the dynamic topological deformation model. Specifically: Based on the preset spatial distribution characteristics of the targeted deposition field, multiple acoustic signature collection points located at different depths within the canopy are set. Broadband acoustic pulses are emitted at each acoustic signature collection point, and echo signals reflected by the leaf canopy surface and scattered by the droplet layer are received, forming echo timing characteristics that include time delay and spectral attenuation information. The acoustic impedance variation characteristics corresponding to droplet deposition at each depth location are extracted from the echo timing characteristics, and the deposition boundary profile formed by the targeted deposition field in actual operation is analyzed based on the spatial continuity of the acoustic impedance variation characteristics, which serves as the actual coverage boundary. The actual coverage boundary and the spatial boundary of the virtual drug-receiving interface are compared point by point, and the deviation sequence between them in terms of spatial position, spatial orientation and boundary curvature is calculated. The deviation sequence is then used as the observation state sequence of the hidden Markov chain. Based on the forward-backward algorithm of hidden Markov chains, the most likely state transition path leading to the deviation is inverted from the observed state sequence. The state transition path corresponds to the distribution of the deviation of each grid node in the dynamic topological deformation model under the boundary constraints. The deviation distribution is superimposed as a boundary constraint on the initial boundary of the dynamic topological deformation model to generate the corrected boundary constraint.

[0039] The acoustic fingerprint collection points are non-uniformly distributed according to the spatial distribution density of the virtual drug-receiving interface and the concealment angle of each concealed leaf curtain unit, so that the spatial sampling density of the acoustic fingerprint collection points is higher in areas with larger concealment angles.

[0040] It should be noted that, based on the previously defined spatial distribution of the virtual drug-receiving interface, acoustic signature sampling points are pre-set at different depths within the canopy. The density of these points is related to the spatial distribution density of the virtual drug-receiving interface and the concealment angle of each concealed leaf canopy unit. Areas with larger concealment angles have higher spatial sampling densities of acoustic signature sampling points to ensure sufficient feedback information is obtained even in severely obstructed areas. At each sampling point, a broadband acoustic pulse is emitted, and the echo signal reflected from the leaf canopy surface and scattered by the droplet layer is received. Because the droplet deposition layer and the exposed leaf canopy surface have different acoustic impedance characteristics, the echo signal from the deposition area exhibits specific characteristics in terms of time delay and spectral attenuation. These acoustic impedance changes are extracted from the echo timing characteristics, thereby resolving the depositional boundary profile formed by the targeted deposition field in actual operations, i.e., the actual coverage boundary.

[0041] After obtaining the actual coverage boundary, it is compared point-by-point with the spatial boundary of the previously defined virtual drug-receiving interface to calculate the deviation sequence in spatial location, spatial orientation, and boundary curvature. This deviation sequence reflects the gap between the actual deposition effect and the expectation. However, due to the complexity of the canopy structure, a deviation at a certain point may be caused by multiple factors, and it is difficult to directly determine which part of the model should be corrected simply by relying on the deviation value itself. Therefore, this embodiment introduces a Hidden Markov Chain (HMM) to solve this inversion problem. The deviation sequence is used as the observation state sequence of the HMM, while the distribution of the deviation of each grid node in the dynamic topological deformation model under the boundary constraints is used as the hidden state sequence. The application logic of the forward-backward algorithm of the HMM in this embodiment is as follows: the forward algorithm starts from the initial time and recursively calculates the probability of being in each hidden state given a partial observation value in the observation sequence; the backward algorithm recursively calculates the probability of being in each hidden state given a future observation value from the final time. By combining the forward probability and the backward probability, the most likely hidden state of the system at each time step under the complete observation sequence can be obtained. In this specific scenario, the forward-backward algorithm actually finds the path most likely to produce the observed deviation sequence among all possible state transition paths. The state sequence on this path corresponds to the distribution of deviations of each grid node in the dynamic topological deformation model under the boundary constraints, that is, which boundary regions of the model deviate from the reality and by how much.

[0042] In this way, the "phenomenon" of depositional boundary deviation is inverted into the "cause" of the deviation distribution of the model's boundary constraints. Finally, these deviations are superimposed as boundary constraints onto the initial boundary of the dynamic topological deformation model to generate corrected boundary constraints, which are used for the re-division of the virtual drug-receiving interface and the replanning of operational parameters in the next operational cycle. In this way, each operation is not isolated, but rather the model is corrected based on the actual feedback from the previous operation, so that the drug application accuracy is gradually improved over multiple operations.

[0043] In addition, this method also includes: Based on the dynamic topological deformation model, the historical canopy topological deformation model under the same phenological calendar as the target litchi orchard is retrieved. The grid node coordinates of the current dynamic topological deformation model are matched with the grid node coordinates of the historical canopy topological deformation model. By constructing a nonlinear spatial mapping relationship between the two, the temporal deformation residual tensor field of the current canopy with nonlinear spatial morphology drift relative to the historical canopy is calculated. Extract the divergence and curl components of the residual vector at each grid node in the temporal deformation residual tensor field, mark the region where the divergence component exceeds the preset deformation threshold as the expansion deformation region, mark the region where the curl component exceeds the preset torsional threshold as the torsional deformation region, and take the spatial union of the expansion deformation region and the torsional deformation region as the allometric growth anomaly region. The spectral angular distance of each grid node in the abnormal allometric growth region is obtained in the multispectral image. Grid nodes with spectral angular distance greater than a preset spectral threshold are identified as physiological stress-related deformation nodes, and the remaining grid nodes are identified as physiological growth deformation nodes. A set of intrinsic modes of physiological growth is constructed using the spatial heterogeneity index at the physiological growth deformation node as the data source, and a set of intrinsic modes of stress is constructed using the spatial heterogeneity index at the physiological stress-related deformation node as the data source. The set of intrinsic modes of physiological growth and the set of intrinsic modes of stress are used as the basis function set for allometric growth empirical mode decomposition. The original spatial heterogeneity index is then decomposed by projection to obtain the physiological growth component and the stress-related component corresponding to the original spatial heterogeneity index. The stress-related component is removed from the original spatial heterogeneity index, and the removed physiological growth component is used as the corrected spatial heterogeneity index. It is then reloaded into the deformation energy function of the dynamic topological deformation model as the constraint boundary of the normal perturbation, thus generating the dynamic topological deformation model after removing physiological stress interference.

[0044] It should be noted that this embodiment considers one issue when constructing the dynamic topological deformation model: the spatial heterogeneity index of the litchi canopy is not entirely determined by the geometric structure of the canopy itself, but is mixed with abnormal deformations caused by physiological changes during the growth cycle and stress factors such as pests and diseases. If all these mixed factors are directly included in the model, the subsequent virtual treatment interface division and operation parameter planning may deviate from the actual treatment needs. For example, leaf curling caused by pests and diseases may be misjudged as the curvature characteristics of a normal leaf canopy, or sparse branches caused by malnutrition may be mistakenly identified as areas requiring focused treatment. To address this issue, this embodiment first retrieves the historical canopy topological deformation model of the same phenological period as the target litchi orchard, and then spatially matches the grid node coordinates of the current dynamic topological deformation model with the grid node coordinates of the historical model. The key to this step is that the two data collection periods are separated by one growth cycle. If only normal physiological growth exists, the two models should show overall uniform expansion or local regular changes in spatial morphology. However, if stress factors such as pests and diseases exist, irregular local abnormal deformations will appear. By constructing a nonlinear spatial mapping relationship between the two, the temporal deformation residual tensor field of the current canopy relative to the historical canopy at the same time is calculated. This tensor field essentially decomposes the difference between the two models into displacement vectors at each grid node.

[0045] To identify anomalous deformation regions from this residual field, the divergence and curl components of the residual vector at each grid node are extracted. Divergence reflects the degree of expansion or contraction of a local region. If the divergence of the residual vector in a certain region is large, it indicates that the region has significantly expanded compared to the same historical period, belonging to expansionary deformation. Curl reflects the degree of torsion in a local region. If the curl is large, it indicates that the region has twisted or rotated, belonging to torsional deformation. Regions with divergence components exceeding a preset deformation threshold are marked as expansionary deformation regions, and regions with curl components exceeding a preset torsional threshold are marked as torsional deformation regions. The spatial union of the two is the allometric growth anomalous region. This region may be caused by abnormal growth factors such as pests and diseases, nutrient stress, or mechanical damage, and requires further identification.

[0046] For grid nodes within the abnormal allometric growth zone, further verification was performed using multispectral images. The spectral features of each pixel in the multispectral image can reflect the physiological state of the leaf, with significant differences in the spectral curves between healthy and stressed leaves. The system obtains the spectral angular distance of each grid node within the abnormal zone at its corresponding position in the multispectral image, i.e., the angle between the spectral curve at that node and the reference spectrum of a healthy leaf; the larger the angle, the more abnormal the physiological state. Grid nodes with a spectral angular distance greater than a preset spectral threshold are identified as physiological stress-related deformation nodes, which are likely caused by abnormal deformation due to pests, diseases, or nutrient stress; the remaining grid nodes are classified as physiological growth deformation nodes, whose deformation is a natural change during normal growth.

[0047] After obtaining the classifications of the two types of nodes, a set of intrinsic mode functions for physiological growth is constructed using the spatial heterogeneity index at nodes exhibiting physiological growth deformation as the data source, and a set of intrinsic mode functions for stress-related deformation is constructed using the spatial heterogeneity index at nodes exhibiting physiological stress-related deformation as the data source. These two function sets are equivalent to two sets of basis functions, describing the heterogeneity index variation patterns under normal growth and stress conditions, respectively. Using these two function sets as the basis function set for allometric growth empirical mode decomposition, the original spatial heterogeneity index can be decomposed by projection decomposition, thus decomposing the original spatial heterogeneity index into two parts: one part is related to the intrinsic mode function set for physiological growth, representing the normal physiological growth component; the other part is related to the intrinsic mode function set for stress-related deformation, representing the abnormal component caused by stress.

[0048] Finally, the stress-related components were removed from the original spatial heterogeneity index, retaining only the physiological growth component as the corrected spatial heterogeneity index. This component was then reloaded into the deformation energy function of the dynamic topological deformation model as the constraint boundary for normal perturbations. This process removed localized abnormal deformations caused by pests, diseases, or nutrient stress from the model, resulting in a new model that eliminated physiological stress interference and more realistically reflected the physiological geometry of the tree canopy. Consequently, subsequent virtual pesticide application interface segmentation, operational parameter planning, and final precise pesticide application will all be based on this "pure" model, avoiding the bias of misjudging stressed areas as application priorities or omitting truly needed areas. This represents a qualitative leap from "deformation modeling" to "removing false information and retaining the true essence."

[0049] In addition, this method also includes: The micro-roughness texture of the leaf canopy epidermis is extracted from the virtual drug-receiving interface, and the anisotropic distribution characteristics of the micro-roughness texture under different spatial orientations are analyzed to form an anisotropic texture tensor field bound to the spatial position of the virtual drug-receiving interface. The impact terminal velocity vector field of each droplet sub-cluster at the moment of arrival at the virtual drug-receiving interface is extracted from the spatial discrete droplet group, and the impact terminal velocity vector field is decomposed along the normal and tangential directions of the virtual drug-receiving interface to obtain the normal impact kinetic energy component and the tangential sliding kinetic energy component. The anisotropic texture tensor field is spatially convolved with the normal impact kinetic energy component and the tangential sliding kinetic energy component to generate a contact angle dynamic response field that characterizes the dynamic wetting behavior between the droplet and the leaf canopy interface. This contact angle dynamic response field includes the dynamic contact angle change range and relaxation time window of the droplet at the moment of impact. Based on the ratio of the dynamic contact angle to the inherent critical contact angle of the blade surface in the dynamic response field of the contact angle, the landing behavior of each droplet cluster is divided into bouncing state, spreading state or rolling state, and the probability distribution corresponding to each landing behavior is extracted as the landing behavior probability field. The sum of the probabilities of the bouncing state and the rolling state in the landing behavior probability field is used as the droplet deposition failure risk coefficient. With the droplet deposition failure risk coefficient as a constraint, the iterative convergence condition of the asymmetric disturbance operator is modified in reverse. Specifically, when the droplet deposition failure risk coefficient corresponding to a certain virtual drug-receiving interface exceeds a preset threshold, the disturbance energy allocation weight in the iteration process of the asymmetric disturbance operator corresponding to the virtual drug-receiving interface is increased, so that the disturbance vector field applies an incremental deflection effect to the spatial region where the virtual drug-receiving interface is located, until the proportion of the spreading state in the landing behavior probability field converges to a preset range.

[0050] It should be noted that in the initial construction of spatially discrete droplet clusters, the spatial matching between droplet clusters and the virtual drug-receiving interface surface contour was mainly achieved on a macroscopic scale. However, an easily overlooked problem in actual operation is that even if the droplet clusters are aligned with the leaf canopy interface, the microscopic interface behavior between the droplets and the leaf canopy epidermis can still cause deposition failure. Specifically, the surface of litchi leaves often has texture features such as microscopic waxy layers, villous structures, or uneven stomatal distribution. These microstructures exhibit anisotropy under different spatial orientations, meaning that droplets impacting from different directions in the same leaf canopy area may exhibit drastically different wetting, spreading, or rebound behaviors. If this factor is ignored, even if the spatial distribution of the droplet clusters is highly matched with the interface curvature, droplets may still bounce and roll off after impact, failing to spread effectively, resulting in a significant reduction in the actual deposition effect. Therefore, this embodiment extracts the microscopic roughness texture of the leaf canopy epidermis from the virtual drug-receiving interface and analyzes the anisotropic distribution characteristics of these textures under different spatial orientations to form an anisotropic texture tensor field bound to the spatial position of the virtual drug-receiving interface. This tensor field describes the response tendency of each interface region to droplet wetting behavior under a specific spatial orientation. For example, when the droplet impact direction is parallel or perpendicular to the leaf vein direction, its spreading behavior may differ significantly. Simultaneously, the impact terminal velocity vector field of each droplet sub-cluster at the instant of arrival at the virtual drug-receiving interface is extracted from the spatially discrete droplet swarm. This vector field is then decomposed along the normal and tangential directions of the interface to obtain the normal impact kinetic energy component and the tangential sliding kinetic energy component. The former determines the intensity of the vertical impact on the leaf canopy surface during droplet impact, while the latter determines the tendency of the droplet to slide along the leaf canopy surface after impact.

[0051] By spatially convolving the anisotropic texture tensor field with the normal impact kinetic energy component and the tangential sliding kinetic energy component, the system essentially couples the droplet impact conditions with the microscopic response characteristics of the blade surface. This generates a dynamic contact angle response field characterizing the dynamic wetting behavior between the droplet and the blade interface. This response field is not a static contact angle value, but rather includes the range of dynamic contact angle changes at the moment of impact and the relaxation time window—that is, the dynamic process of contact angle change over time from the start of impact to the final stable spreading or retraction of the droplet. With this dynamic response field, the system can determine whether the landing behavior of each droplet sub-cluster is more inclined towards a bouncing, spreading, or rolling state: specifically, by comparing the ratio of the dynamic contact angle to the inherent critical contact angle of the blade surface, if the dynamic contact angle is much greater than the critical contact angle and the relaxation time is short, the droplet tends to bounce away; if the dynamic contact angle rapidly decreases to below the critical contact angle, the droplet tends to spread and adhere; if the tangential sliding kinetic energy is too large and the normal impact kinetic energy is insufficient, the droplet tends to roll and slide. By extracting the probability distribution corresponding to each landing behavior, a landing behavior probability field is formed.

[0052] Based on this, a key indicator is obtained: the sum of the probabilities of the bouncing and rolling states in the landing behavior probability field, defined as the droplet deposition failure risk coefficient. This directly reflects the proportion of droplets that may not be effectively deposited on the leaf canopy surface under the current impact conditions of the droplet clusters. If the deposition failure risk coefficient corresponding to a certain virtual drug-receiving interface exceeds a preset threshold, it indicates that even if the spatial distribution of the droplet clusters is aligned with the interface, the actual deposition effect is still unsatisfactory due to the mismatch between impact dynamics and microtexture. To address this, the failure risk coefficient is used as a constraint to reversely correct the iterative convergence condition of the asymmetric perturbation operator. Specifically, the correction method is to increase the perturbation energy allocation weight in the iteration process of the asymmetric perturbation operator corresponding to the virtual drug-receiving interface, so that the perturbation vector field applies an incremental deflection effect to the spatial region where the interface is located, thereby adjusting the impact velocity vector when the droplet clusters arrive at the interface. For example, by changing the incident angle of the droplets to increase the proportion of normal impact kinetic energy and decrease the proportion of tangential sliding kinetic energy, or by adjusting the spatial attitude of the droplet clusters to match the anisotropic direction of the texture. After this correction, the system iterates the asymmetric perturbation operator again until the proportion of the spread state in the landing behavior probability field corresponding to the interface converges to the preset range.

[0053] Through the above methods, this embodiment extends the control of pesticide application precision from macroscopic spatial alignment to microscopic interface landing, constructing a complete deposition control chain from surface matching to texture adaptation. This allows for the systematic identification and correction of deposition failure problems such as droplet bouncing and rolling caused by microscopic textures, improving the actual adhesion efficiency of droplets on the surface of litchi leaf canopy, and providing a more fundamental technical guarantee for precise pesticide application under complex canopy conditions.

[0054] like Figure 2 As shown, the second aspect of the present invention discloses a precision pesticide application system for litchi orchards based on the combined use of plant protection drones and adjuvants. The system includes a memory and a processor. The memory stores a program for a precision pesticide application method for litchi orchards based on the combined use of plant protection drones and adjuvants. When the program for the precision pesticide application method for litchi orchards based on the combined use of plant protection drones and adjuvants is executed by the processor, the steps of the precision pesticide application method for litchi orchards based on the combined use of plant protection drones and adjuvants as described in any one of the claims are implemented.

[0055] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants, characterized in that, Includes the following steps: A dynamic topological deformation model characterizing the spatial heterogeneity index of tree canopy was constructed based on multispectral images of litchi orchards and point clouds of single-tree lidar. Based on the dynamic topological deformation model, the inner canopy is divided into multiple virtual drug-receiving interfaces according to the drug-receiving response potential, and a pulsed hovering phase sequence matching the spatiotemporal phase of each virtual drug-receiving interface is generated. Based on the pulsed hovering phase sequence, the potential energy of the droplet-wind field interface interaction between the UAV rotor downwash flow field and the droplet group after mixing with the drug and additives is calculated. Using the potential energy of the droplet-wind field interface as a constraint, an asymmetric turbulence operator is introduced to reconstruct a spatial discrete droplet swarm that matches the curvature of the virtual drug-receiving interface. By utilizing the spatial distribution parameters of the spatially discrete droplet group, the charge state of the charged additive in the drug solution is modulated through magnetohydrodynamics to form a targeted deposition field that matches the curvature characteristics of each concealed leaf canopy unit in the dynamic topological deformation model. Acoustic echo signals from different depths within the canopy after drug application were collected. The actual coverage boundary of the targeted deposition field was analyzed, and the deviation between the actual coverage boundary and the virtual drug-receiving interface was inverted into the boundary constraints of the modified dynamic topological deformation model based on a hidden Markov chain.

2. The method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants as described in claim 1, characterized in that, A dynamic topological deformation model characterizing the spatial heterogeneity index of tree canopy is constructed based on multispectral images of litchi orchards and point clouds of single-tree lidar sensors. Specifically: Acquire multispectral images of litchi orchards and point clouds from single-tree lidar, and register the two to the same spatial coordinate system to form a spatially aligned multimodal data volume; From the spatially aligned multimodal data volume, a spatial heterogeneity index is extracted to characterize the heterogeneity of leaf canopy distribution in the vertical and horizontal directions within each individual tree canopy. The spatial heterogeneity index is obtained by weighted fusion of the texture gradient reflecting chlorophyll content in the multispectral image and the structural entropy reflecting the gaps between branches and leaves in the lidar point cloud. Using the spatial heterogeneity index as the initial boundary condition, an initial mesh model covering the entire canopy spatial domain is constructed. Based on the single-tree skeleton topology extracted from the lidar point cloud, Laplacian smoothing is applied to the initial mesh model to generate a static canopy model with continuous surface features. Using the spatial heterogeneity index as the deformation driving field, non-uniform normal perturbations are applied to each grid node of the static canopy model to form a dynamic topological deformation model.

3. The method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants as described in claim 1, characterized in that, The dynamic topological deformation model ensures that the displacement of the grid nodes is consistent with the spatial gradient direction of the spatial heterogeneity index by iteratively minimizing the deformation energy function.

4. The method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants as described in claim 1, characterized in that, Based on the dynamic topological deformation model, the inner canopy is divided into multiple virtual drug-receiving interfaces according to the drug-receiving response potential, and a pulsed hovering phase sequence matching the spatiotemporal phase of each virtual drug-receiving interface is generated, specifically: Based on the dynamic topological deformation model, the spatial coordinates, spatial heterogeneity index, and normal curvature of each grid node are extracted, and the canopy foliage is discretized into multiple foliage micro-units with independent geometric features and heterogeneous properties according to the topological connection relationship between each grid node. Based on the spatial heterogeneity index and normal curvature of each foliation microunit, the drug-receiving response potential energy of each foliation microunit to droplet deposition is determined. The drug-receiving response potential energy is positively correlated with the spatial heterogeneity index and negatively correlated with the normal curvature. Adjacent leaf micro-units with drug-receiving response potential energy higher than a preset threshold are spatially clustered to form multiple virtual drug-receiving interfaces, and each virtual drug-receiving interface is assigned a spatiotemporal phase identifier that represents its spatial position and orientation in three-dimensional space. Based on the spatiotemporal phase identifiers of each virtual drug receiving interface, the hovering point coordinates and hovering time windows required by the UAV when traversing each virtual drug receiving interface are calculated, and a pulsed hovering phase sequence corresponding to each spatiotemporal phase identifier is generated. Each pulsed hovering phase sequence includes the time when the UAV arrives at the hovering point and the duration of pulsed spraying at the hovering point.

5. The method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants as described in claim 1, characterized in that, Based on the pulsed hovering phase sequence, the potential energy of the droplet-wind interface interaction between the UAV rotor downwash flow field and the droplet group after mixing with the liquid and additives is calculated, specifically as follows: Based on the pulsed hovering phase sequence, the rotor speed and blade pitch angle corresponding to each hovering point are extracted, and the three-dimensional velocity vector distribution and turbulence intensity distribution of the downwash flow field generated by the UAV rotor at each hovering point are calculated to form dynamic rotor aerodynamic parameters that correspond one-to-one with the hovering point. The physical property parameters of the mixture of drug solution and adjuvant are obtained, including the surface tension and viscosity of the mixture and the dynamic surface tension relaxation time caused by the adjuvant. The dynamic rotor aerodynamic parameters and the physical property parameters are input into the droplet breakup and motion model to solve the particle size distribution and initial velocity vector of the droplet swarm formed under the action of the downwash flow field at different hovering points after being ejected from the nozzle, and the initial dynamic state of the droplet swarm associated with the hovering point is generated. The initial velocity vector of each droplet in the initial dynamic state of the droplet swarm is coupled point by point with the three-dimensional velocity vector distribution of the downwash flow field to determine the momentum exchange intensity between the droplet surface and the flow field when each droplet moves in the downwash flow field. Based on the dynamic surface tension relaxation time correction of the droplet deformation response hysteresis effect during the motion process, the droplet-wind field interface interaction potential energy characterizing the energy transfer efficiency between the droplet swarm and the downwash flow field is generated.

6. The method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants as described in claim 1, characterized in that, Using the potential energy of the droplet-wind interface interaction as a constraint, an asymmetric disturbance operator is introduced to reconstruct a spatially discrete droplet swarm that adapts to the curvature of the virtual drug-receiving interface, specifically: Based on the potential energy of the droplet-wind field interface interaction, the spatial distribution gradient of the energy transfer efficiency between the droplet group and the downwash flow field at each hovering point is extracted, and the spatial distribution gradient is used as the weighting coefficient to construct the initial configuration of the asymmetric turbulence operator. The asymmetric turbulence operator consists of multiple sets of turbulence vector fields that are non-uniformly distributed along the circumferential direction of the rotor rotation. Obtain the curvature distribution data of the virtual drug receiving interface, map the principal curvature direction and curvature radius of each virtual drug receiving interface to the turbulence vector field of the asymmetric turbulence operator, and adjust the amplitude and direction of each turbulence vector field to make the asymmetric turbulence operator and the surface geometric features of the virtual drug receiving interface form a spatial correspondence. The asymmetric turbulence operator is superimposed on the downwash flow field to apply an asymmetric deflection effect to the trajectory of the droplet group in the downwash flow field, so that the droplet group is spatially deaggregated into multiple mutually separated droplet clusters that conform to the curved contours of each virtual drug-receiving interface. Based on the degree of conformity between each droplet cluster and the corresponding virtual drug-receiving interface, the spatial distribution parameters of the asymmetric perturbation operator are iteratively optimized until the deviation between the spatial distribution morphology of each droplet cluster and the curvature characteristics of the virtual drug-receiving interface converges, forming a spatially discrete droplet swarm.

7. The method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants as described in claim 1, characterized in that, By utilizing the spatial distribution parameters of the spatially discrete droplet swarm, the charge state of the charged additive in the drug solution is modulated through magnetohydrodynamics to form a targeted deposition field that matches the curvature characteristics of each concealed foliage unit in the dynamic topological deformation model. Specifically: Based on the spatial distribution parameters of the spatial discrete droplet group, the spatial coordinates, spatial orientation, and relative positional relationship between each droplet cluster and the corresponding virtual drug receiving interface in three-dimensional space are extracted. The spatial distribution parameters are then mapped to the modulation space of the charged adjuvant at the nozzle to generate a spatial charge density distribution corresponding to each droplet cluster. Obtain the curvature characteristics of each concealed leaf curtain unit in the dynamic topological deformation model, including the spatial position, normal curvature, and concealment angle formed by the concealed leaf curtain unit being blocked by adjacent leaf curtains; Establish a mapping relationship between the space charge density distribution and the curvature characteristics, and calculate the target charge state required for the fog droplets to overcome flow field disturbances and spread along the curvature surface during their flight toward the corresponding concealed leaf curtain unit, based on the mapping relationship. The target charge state includes charge magnitude and charge polarity. Based on the target charge state, the electric field strength and direction of the charged additive applied to the nozzle are dynamically adjusted so that the charged additive forms a charge state spatial modulation pattern in the liquid that is compatible with the spatial charge density distribution. The liquid medicine with the aforementioned charge state spatial modulation pattern is sprayed out in a sequence corresponding to the droplet clusters, so that each droplet cluster maintains a charge state consistent with the target charge state during flight, and spreads directionally along the curved surface with the aid of charge-induced mirror force when it reaches the concealed leaf canopy unit, forming a targeted deposition field.

8. The method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants as described in claim 1, characterized in that, Acoustic echo signals from different depths within the canopy after drug application were collected. The actual coverage boundary of the targeted deposition field was analyzed, and the deviation between the actual coverage boundary and the virtual drug-receiving interface was inverted into boundary constraints for correcting the dynamic topological deformation model based on a hidden Markov chain. Specifically: Based on the preset spatial distribution characteristics of the targeted deposition field, multiple acoustic signature collection points located at different depths within the canopy are set. Broadband acoustic pulses are emitted at each acoustic signature collection point, and echo signals reflected by the leaf canopy surface and scattered by the droplet layer are received, forming echo timing characteristics that include time delay and spectral attenuation information. The acoustic impedance variation characteristics corresponding to droplet deposition at each depth location are extracted from the echo timing characteristics, and the deposition boundary profile formed by the targeted deposition field in actual operation is analyzed based on the spatial continuity of the acoustic impedance variation characteristics, which serves as the actual coverage boundary. The actual coverage boundary and the spatial boundary of the virtual drug-receiving interface are compared point by point, and the deviation sequence between them in terms of spatial position, spatial orientation and boundary curvature is calculated. The deviation sequence is then used as the observation state sequence of the hidden Markov chain. Based on the forward-backward algorithm of hidden Markov chains, the most likely state transition path leading to the deviation is inverted from the observed state sequence. The state transition path corresponds to the distribution of the deviation of each grid node in the dynamic topological deformation model under the boundary constraints. The deviation distribution is superimposed as a boundary constraint on the initial boundary of the dynamic topological deformation model to generate the corrected boundary constraint.

9. The method for precise pesticide application in litchi orchards based on the combined use of agricultural drones and adjuvants as described in claim 1, characterized in that, The acoustic fingerprint collection points are non-uniformly distributed according to the spatial distribution density of the virtual drug-receiving interface and the concealment angle of each concealed leaf curtain unit, so that the spatial sampling density of the acoustic fingerprint collection points is higher in areas with larger concealment angles.

10. A precision pesticide application system for litchi orchards based on the combined use of agricultural drones and adjuvants, characterized in that: The system includes a memory and a processor. The memory stores a program for a precise application method for lychee orchards based on the combined use of plant protection drones and adjuvants. When the processor executes the program for the precise application method for lychee orchards based on the combined use of plant protection drones and adjuvants, it implements the steps of the precise application method for lychee orchards based on the combined use of plant protection drones and adjuvants as described in any one of claims 1 to 9.

Images