Agricultural drone operation path planning method
By constructing a three-dimensional scene of farmland and dividing it into zones for operation, generating differentiated flight path parameters and optimized paths, the problems of uneven pesticide spraying and high energy consumption of plant protection drones in complex farmland have been solved, achieving efficient and precise pesticide spraying.
Patent Information
- Application Number
- CN202611374566.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-07
- Publication Date
- 2026-11-10
AI Technical Summary
Existing agricultural drone path planning technology cannot adapt to complex and ever-changing farmland plots, resulting in uneven spraying, blind spots, high energy consumption and low efficiency. In particular, there is serious problem of missed spraying in densely planted areas and repeated spraying in sparse areas, and it cannot dynamically adapt to changes in the field.
A 3D scene of farmland is constructed by using lidar, RTK differential positioning and high-definition aerial photography equipment. Combined with intelligent clustering algorithm for zoning operations, differentiated flight path parameters are generated, and the flight path is optimized to adapt to irregular boundaries and static obstacles, enabling real-time obstacle avoidance and dynamic obstacle response.
It improves the uniformity and coverage of pesticide spraying, reduces energy consumption, reduces missed spraying and overspraying, enhances the accuracy and efficiency of operations, and adapts to the dynamic changes in farmland.
Smart Images

Figure SMS_80 
Figure QLYQS_11 
Figure QLYQS_14
Abstract
Description
Technical Field
[0001] This application relates to the field of drone path planning technology, specifically to a method for planning the operation path of a plant protection drone. Background Technology
[0002] Agricultural drones are unmanned aerial vehicles used for the protection of agricultural and forestry plants. Their core value lies in their high efficiency, precision, pesticide savings, and labor savings, making them a standard tool in modern agriculture. With the intelligent and precision transformation of modern agriculture, agricultural drones, with their advantages of high operational efficiency, low labor costs, and strong terrain adaptability, have been widely used in the prevention and control of pests and diseases and weeding operations in field crops, cash crops, and forest vegetation. At present, agricultural drone operations have basically replaced traditional manual backpack spraying and ground-based spraying methods, becoming the core operational equipment for agricultural and forestry plant protection. Path planning, as the core key technology for autonomous operation of agricultural drones, directly determines the pesticide spraying coverage, deposition uniformity, pesticide utilization rate, and overall energy consumption. It is a crucial link in ensuring the quality of plant protection operations, reducing pesticide pollution, and saving operating costs.
[0003] Currently, the autonomous operation system of commercial agricultural drones in China has basically matured, and can realize basic functions such as plot identification, automatic route generation, autonomous flight, and automatic obstacle avoidance. It can meet the basic plant protection operation needs of standardized, regular, and uniformly growing fields. However, my country's farmland has significant characteristics: serious plot fragmentation, irregular plot shape, large differences in micro-topography, and crops within the same plot are affected by light, water, fertilizer, and pests and diseases, resulting in uneven growth, large differences in plant height, and inconsistent canopy density. Existing general path planning schemes are difficult to adapt to the complex and ever-changing actual field conditions.
[0004] Considering the complex realities of field operations, existing general path planning technologies suffer from the following unavoidable and substantial defects, all stemming from the inherent principles of the technology and unable to be resolved through conventional parameter fine-tuning: 1. Using uniform flight path spacing, altitude, speed, and width across the entire farmland ignores differences in crop density, growth, and terrain. In densely planted areas with lush crops, the thick canopy obstructs the spread of pesticides. Uniform flight path spacing, altitude, speed, and width cannot guarantee that the pesticides penetrate the canopy, resulting in missed spraying of the lower and middle layers of crops and the undersides of leaves, leading to serious pest and disease residues. In sparsely planted areas with bare ground, uniform standard parameters will result in excessive overlap in spraying between rows, causing large-scale overspraying and excessive pesticide deposition. This not only wastes pesticides and increases operating costs but also easily leads to ecological problems such as pesticide residues in the soil and water pollution. 2. Traditional serpentine right-angle turns cause the drone to tilt rapidly and its flight attitude to change abruptly during the turn. The airflow under the drone is severely turbulent, and the pesticide droplets are prone to drifting, spreading, and settling interruptions. At the same time, the flight speed changes drastically during the turn, and the spraying stability is greatly reduced. This results in spraying blank areas that are common at the edges and corners of the plots and at the junctions of the flight paths. The rate of pesticide leakage at the edges and corners is high, which are the core blind spots where pests and diseases are prone to occur in the field. These dead zones have been present for a long time and cannot be eliminated. 3. The core optimization objectives of existing path planning algorithms are: shortest flight path and highest operation efficiency. However, they do not establish a two-way constraint mechanism between pesticide deposition effect and flight energy consumption. The shortest path often has problems such as many turns, unstable local airflow, and poor spray uniformity. Simply pursuing operation speed will sacrifice the quality of plant protection operations. If the flight speed is reduced and the row spacing is shortened to improve the efficacy, the flight energy consumption will be greatly increased and the operation efficiency will be reduced. This creates a technical contradiction: improved efficacy leads to increased energy consumption, and energy saving leads to decreased operation quality. It is impossible to achieve the optimal balance between the two. 4. The existing path planning is a one-time static planning mode. Once the plot route model is generated, it is permanently fixed and cannot adapt to dynamic changes in the field. For the same plot, there are significant differences in crop canopy density and plant height under different growth cycles, seasons, and water and fertilizer conditions. Fixed routes cannot adapt to dynamic changes in growth. At the same time, problems such as missed spraying, energy consumption deviation, and uniformity error generated by a single operation cannot be fed back and iterated. Similar operation defects recur, plot adaptability and operation accuracy cannot be continuously improved, and the level of intelligence and refinement is low.
[0005] In summary, existing technologies lack a multi-dimensional dynamic path planning scheme that combines crop growth grading, terrain features, and pesticide efficacy requirements. After a detailed search, the applicant did not find any relevant technical solutions. Therefore, a new technical solution is needed to address the aforementioned technical problems. Summary of the Invention
[0006] A method for planning the operational path of agricultural protection drones includes the following steps: S1: Based on data collected by lidar, high-definition aerial photography equipment, and RTK differential positioning module, extract three types of feature information: farmland terrain, static obstacles, and crop growth, and complete the reconstruction of the three-dimensional farmland scene; S2: The crop canopy coverage rate extracted from the reconstructed 3D farmland scene is used as the core indicator for zoning. The plot boundaries are corrected by auxiliary features, and the irregular adaptive zoning of farmland is achieved by combining intelligent clustering algorithm to generate different sub-regions. The auxiliary features include the average plant height, plant density distribution, the location and area ratio of blank bare land in the field, and the density of crop canopy. The sub-regions include: Level 1 densely planted area, with crop canopy coverage > 85%; Level 2 conventional area, with crop canopy coverage 50%–85%; and Level 3 sparsely planted area, with crop canopy coverage < 50%. S3: Based on different sub-regions, perform zone-specific differentiated route parameter matching; S4: Based on the differentiated route parameters of different sub-regions, generate serpentine routes that fit the boundary contours of each sub-region, and integrate the straight edges, corners and arc edges of farmland to form a circular transition route to generate a composite operation path. S5: Under the premise of a complex operation path, perform obstacle avoidance for static obstacles and generate multiple preliminary operation routes; S6: Based on multiple preliminary operation routes, complete the final operation route selection.
[0007] As a preferred option, it also includes S7: real-time obstacle avoidance for dynamic obstacles when operating based on the final operation route.
[0008] As a preferred option, the standard for matching the zone-differentiated route parameters in S3 is as follows: In the first-level dense planting area, based on the standard flight path spacing, the flight path spacing is reduced by 15%–25%; based on the standard operating altitude, the flight altitude is reduced by 20%–40%; and based on the standard operating speed, the flight speed is reduced by 40%–50%. In the Level 2 standard zone, the spacing between flight paths shall be the standard spacing with a fluctuation error of ≤5%; the flight altitude shall be the standard operating altitude with a fluctuation error of ≤5%; and the flight speed shall be the standard flight speed with a fluctuation error of ≤5%. In the Level 3 sparse zone, the standard flight path spacing is used as the benchmark, with the flight path spacing increased by 10%–20%; the flight altitude adopts the standard operating altitude, with a fluctuation error of ≤5%; and the flight speed is increased by 20%–30% based on the standard operating speed.
[0009] As a preferred embodiment, S4 includes: S41: For main operation areas with regular farmland terrain and no obvious irregular turning points, based on the crop canopy coverage and auxiliary characteristics corresponding to the first-level dense planting area, the second-level regular area, and the third-level sparse area, a serpentine flight path that fits the irregular boundary is dynamically generated with varying distances. The density of the flight path is completely matched with the needs of regional crop operations. S42: For the edges of farmland plots, divide them into straight edges, corners, and curved edges, and optimize flight routes accordingly.
[0010] As a preferred embodiment, step S42 includes the following steps: S421: The control system retrieves the reconstructed 3D model of the farmland, traverses the outer contour of the farmland point by point, collects basic information such as the plane coordinates and elevation of all boundary points, and completely restores the true shape of the plot. S422: Based on the basic data such as plane coordinates and elevation of the collected points, a complete closed boundary outline of farmland plots is generated by fitting continuous coordinate points. Based on the generated boundary outline, the vector changes of adjacent coordinate points are analyzed to calculate the outline direction. According to the abrupt change characteristics of the angle between line segments and the deflection amplitude, the boundary outline is automatically divided into three types of boundary structures: straight edge, corner, and arc edge. S423: Extract boundary features for three types of boundary structures: straight edges, corners, and curved edges. The extraction of boundary features includes curvature calculation and bending feature extraction in different regions. S424: Based on the extracted boundary features, generate corresponding operation routes according to type.
[0011] As a preferred embodiment, in S422, the straight edge is a straight segment with three or more consecutive adjacent line segments whose deflection amplitude is ≤8°, no abrupt change in direction, and constant curvature; the corner is the intersection of two contour segments with a single deflection amplitude ≥30° and a single-point abrupt change in direction; the arc edge is a smooth curved contour segment with three or more consecutive adjacent line segments whose deflection amplitude is between 8° and 30°, no single-point abrupt change, consistent deflection direction, and continuously changing curvature.
[0012] As a preferred embodiment, S422 synchronously introduces elevation data to fit the terrain and correct the spatial position of boundary points, thereby eliminating the boundary spatial offset caused by terrain undulations.
[0013] As a preferred embodiment, step S5 includes the following steps: S51: Load the center coordinates, floor area, height, and circumscribed contour of the static obstacle to complete spatial positioning; S52: Divide the complex operation path at the transition points between straight lines, arcs, and curves into straight sections, corner arc transition sections, and arc edge curve sections. S53: According to the flight sequence of the agricultural drone, conduct collision checks on each segment after it is split up, determine whether there is a collision with a static obstacle, and mark the segments with risks. S54: For the marked risky segments, the operation route after detour is replanned based on the composite operation route generated by S424, generating multiple preliminary operation routes.
[0014] As a preferred option, in S54, after the replanning of the detour operation route is completed, it is checked whether the connection between each segment meets the requirements of continuity and smoothness. If so, all segments are integrated to generate a preliminary operation route; otherwise, the replanning of the detour operation route is carried out, and the preliminary operation route is generated after the retest is passed.
[0015] As a preferred option, the continuity detection method is to extract the three-dimensional coordinates of the beginning and end points of adjacent flight segments and calculate the distance. If the distance is within the error range, it is determined that there is no break in the connection. Smoothness is detected by vector angle and curvature calculation. The flight direction angle and curvature change value at the connection point are calculated. The arc transition form of corners and arc areas is the focus of verification. If there are no sudden changes in angle or curvature, it is qualified.
[0016] As a preferred embodiment, step S6 includes the following steps: S61: Energy consumption constraints. Under the premise of ensuring no blind spots in the operation of the entire plot and no safety conflicts in the flight routes, the first round of screening of multiple preliminary operation routes is carried out based on turning frequency and flight mileage, and the preliminary operation routes with redundant turns, invalid detours and repeated segments are eliminated. S62: Efficacy dimension constraint. Based on the differentiated operation parameters of each zone, the spray coverage effect, deposition uniformity and the proportion of missed and re-sprayed areas of each preliminary operation route are quantitatively verified, and preliminary operation routes with substandard efficacy are eliminated. S63: For the remaining preliminary operation routes, a weighted calculation and comprehensive score are performed on multiple parameters such as energy consumption loss, drug efficacy accuracy, route deviation error, and operation adaptability to select a unique final operation route.
[0017] As a preferred option, the quantization verification standard in S62 is as follows: Full-area pesticide spraying coverage ≥98%: First count the total number of grids, then count the number of grids effectively covered by pesticide. Full-area pesticide spraying coverage = number of effectively covered grids / total number of grids in the area × 100%. A result ≥98% is considered qualified. If the full-area pesticide spraying coverage is <98%, delete the preliminary operation route. The percentage of missed spray areas is ≤1.5%: Missed spray percentage = number of missed spray grids / total number of grids in the area × 100%. A result of ≤1.5% is acceptable. If the percentage of missed spray areas is >1.5%, the preliminary operation route is deleted. The percentage of areas with excessive overspraying is ≤2%: Percentage of areas with excessive overspraying = number of grids with excessive overspraying / total number of grids in the area × 100%. If the calculated result is ≤2%, it is considered acceptable. If the percentage of areas with excessive overspraying is >2%, the preliminary operation route is deleted.
[0018] As a preferred embodiment, the calculation method for S63 is as follows: in, This represents the overall score. Indicates energy consumption points. This indicates the weight corresponding to the energy consumption score. Indicates the precision of drug efficacy. This indicates the weight corresponding to the drug efficacy accuracy score. Indicates the deviation of the flight path. This indicates the weight corresponding to the flight path offset. Indicates job suitability score. This indicates the weight corresponding to the job suitability score; Energy consumption E represents actual energy consumption. This represents the theoretical minimum energy consumption across the entire domain. Indicates the maximum allowable energy consumption limit; Drug efficacy precision C represents the actual coverage rate of the pesticide spray. Indicates the minimum acceptable threshold. This indicates the ideal coverage rate; Flight path deviation H represents the average offset of the operation route relative to the boundary of the farmland plot; Maximum allowable offset limit; Job adaptability score Scoring is based on the smoothness of cornering and the degree of conformity to the curve: conformity to the boundary and no sudden braking during cornering. When making a slight, unreasonable turn, The value should be between 0.4 and 0.9; when there are frequent sharp turns and a risk of scraping, <0.4.
[0019] As a preferred embodiment, the At the same time, the corners are transitioned using tangent circular arcs, with the arc radius matching the profile curvature; there are no sudden changes in the turning angle at a single point, the flight path closely follows the boundary, and the average offset H ≤ H allow H allow H is the threshold of the excellent dividing line. allow < There is no short-distance, continuous, or frequent reversal.
[0020] As a preferred option, When the value is between 0.4 and 0.9, the radius of the corner arc deviates from the optimal value, but no sharp right-angle turns occur; local small segments of the route deviate slightly from the boundary H. allow <H≤ = , This indicates the safe critical distance for the drone; there are a few instances of unreasonable local turns, but no close-range continuous sharp turns or risk of collision; the criteria for judging a few instances of unreasonable local turns are as follows: N represents the total number of turning points at the boundary of this segment, and n represents the number of unreasonable turning points.
[0021] As a preferred option, When the radius is less than 0.4, the aircraft flies directly in a right-angled zigzag path without a rounded transition at corners; it makes multiple large changes of direction in a short distance; and its flight path deviates beyond the limit. Being too close to obstacles poses a risk of scraping.
[0022] As a preferred embodiment, in step S7, real-time obstacle avoidance is performed for dynamic obstacles, and a small-scale avoidance path is generated instantly.
[0023] As a preferred embodiment, step S7 includes the following steps: S71, the agricultural drone flies along the final operation route, and the drone’s onboard lidar and high-definition visual recognition module continuously scan the operation area ahead of the route in real time. S72, when detecting dynamic obstacles, combines the current position of the agricultural drone, the coordinates of the forward flight segment, and the spatial data of the obstacle to calculate the shortest distance between the operation route and the obstacle in real time. S73, when the shortest distance is less than the preset safe critical distance for the drone. At that time, local obstacle avoidance is triggered; S74, with the obstacle as the center, the length of the operation route that re-detours near the obstacle shall not exceed 5% of the entire operation route; within the designated 5% local area, the path shall be finely adjusted by either parallel offset or small-radius circular arc detour. S75, the agricultural drone executes an obstacle avoidance operation route; when it flies out of the range of the obstacle's influence, the control system activates the trajectory matching algorithm and coordinate difference calibration module to gradually align with the original operation route. After completing coordinate calibration, the agricultural drone returns to the final operating route of S6 in S76; all sub-area operating parameters are restored to the state of S4.
[0024] In summary, due to the adoption of the above technical solution, this application has the following advantages: 1. By fusing multi-source heterogeneous data from lidar, RTK positioning, and visual images, a three-dimensional farmland scene including terrain, static obstacles, and crop growth is constructed, breaking through the limitations of traditional two-dimensional planar planning and accurately distinguishing densely planted areas, conventional areas, and sparse areas, providing a data foundation for precise zoned operations. 2. Divide the operation area into three levels based on the crop canopy coverage rate as a quantitative standard, and match the corresponding flight path spacing, flight altitude and flight speed according to different growth conditions to solve the core problems of missed spraying in densely planted areas, waste of pesticide in sparsely planted areas and uneven pesticide efficacy in conventional areas from the source. 3. Dynamic pitch generation creates a serpentine flight path that fits irregular boundaries, and combines it with a smooth arc transition structure to generate a composite operation path. This eliminates the traditional 90° right-angle turn, reduces pesticide drift, and, with turn speed reduction and pressure stabilization control, significantly improves the spray coverage in corner areas. The flight is stable throughout and the spray uniformity is significantly improved. 4. Prevent static obstacles in advance and respond quickly to dynamic obstacles in the field during operation. For dynamic obstacles, only make minor adjustments to the flight path in a small area around the obstacle, and accurately return to the original flight path after bypassing the obstacle. There is no global flight path disorder, which takes into account both flight safety and the integrity of the overall operation layout.
[0025] 5. All zoning standards, operation parameters, obstacle avoidance response, and iteration range are quantified using percentage indicators. The system executes automatically without manual intervention. Key indicators such as operation coverage, missed spraying rate, and energy consumption are superior to existing technologies, making it suitable for the needs of large-scale and precise modern agricultural plant protection. Detailed Implementation
[0026] The following provides a detailed description of specific embodiments of the present invention. It should be noted that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the scope of the invention. Example 1
[0027] This embodiment provides a method for planning the operation path of an agricultural drone, including the following steps: S1: Based on multi-source sensor data from LiDAR, high-definition aerial photography equipment, and RTK differential positioning module, three types of feature information—farmland topography, static obstacles, and crop growth—are extracted to complete the reconstruction of a 3D farmland scene. The reconstructed 3D scene integrates multi-source feature information from farmland topography, static obstacles, and crop growth, replicating the real 3D landform and crop distribution status of farmland. This achieves accurate micro-topography restoration and digital archiving of static obstacles, while also completing topographic distortion correction and spatial normalization processing of 2D growth features to eliminate 2D imaging errors.
[0028] When extracting farmland topographic feature information, the plant protection drone is equipped with RTK differential positioning module, lidar and high-definition aerial photography equipment to conduct full-area scanning and mapping of the target farmland, collect the original farmland topographic data in all directions, and obtain farmland topographic feature information through algorithm calculation; the positioning accuracy of RTK differential positioning module is controlled within ±2 cm, and the lidar elevation sampling interval is ≤0.1 m; among them, farmland topographic feature information mainly includes plot boundary coordinates, full-area elevation gradient data, surface slope parameters, distribution of field ridges and ditches, and ground concave and convex micro-topographic undulation information; plot boundary coordinates are obtained through RTK differential positioning module, and full-area elevation gradient data, surface slope parameters, distribution of field ridges and ditches, and ground concave and convex micro-topographic undulation information are all calculated by lidar three-dimensional point cloud data; the specific calculation method is that the lidar obtains three-dimensional point cloud (X, Y, Z), and calculates through point cloud processing algorithm: (1) full-area elevation gradient data, for the elevation Z of each point cloud position, calculate the ratio of the elevation difference of adjacent points to the horizontal distance, ΔZ represents the elevation difference, the vertical elevation difference between the current lidar point cloud location and the adjacent point cloud location, in meters; ΔL represents the horizontal distance difference, the straight-line distance between the current lidar point cloud location and the adjacent point cloud location, in meters; ΔL represents the horizontal spacing, excluding elevation undulations; by using the ratio of elevation difference / horizontal distance difference, the rate of change of the terrain tilt can be quantified, and after point-by-point calculation, a global elevation gradient data distribution map is generated, realizing the accurate depiction of the elevation changes of farmland micro-topography; (2) When converted to angles, ΔZ represents the elevation difference, ΔX represents the difference in X-axis plane coordinates between two adjacent points, and ΔY represents the difference in Y-axis plane coordinates between two adjacent points; the local plane is fitted with the neighborhood point cloud to calculate the plane tilt angle and obtain the surface slope parameters at each location; (3) The distribution location of field ridges and gullies is identified by elevation change. Linear areas where the elevation suddenly rises are field ridges, and linear depressions where the elevation suddenly drops are gullies; based on the elevation change characteristics of farmland topography, a fixed elevation difference judgment threshold is set, and the point cloud is used to determine the elevation difference. Neighborhood elevation difference comparison automatically identifies linear field ridge areas with sudden elevation increases and linear gully areas with sudden elevation decreases, realizing fully automatic extraction and positioning of linear micro-topographic features; (4) Ground concavity and convexity micro-topographic undulation information, ground concavity and convexity micro-topographic undulation is used to depict the fine changes in elevation of a small area of farmland. The algorithm takes a single target point as the center, delineates a local small neighborhood of fixed size, extracts the elevation extreme values of all point clouds in the neighborhood, and calculates the micro-topographic undulation degree through the range formula, undulation degree = , The maximum elevation value within the small neighborhood. The minimum elevation value within a small neighborhood is used. If the calculated undulation exceeds the preset stability threshold, the area is determined to be a micro-topography with uneven ground. The smaller the undulation, the flatter the surface; the larger the undulation, the more significant the surface unevenness deviation. This enables refined identification of micro-topography across the entire area.
[0029] When extracting static obstacle feature information, agricultural drones equipped with RTK differential positioning modules, lidar, and high-definition aerial photography equipment conduct comprehensive screening, precise calibration, and classification of static obstacles in farmland. The identified objects include various operational interference obstacles such as utility poles, trees and shrubs, irrigation wells, fences, ditch embankments, and piles of debris at field edges. Lidar collects 3D point cloud data from the farmland, accurately acquiring the obstacle's 3D height, outline dimensions, and spatial morphological features. The onboard high-definition aerial photography equipment, combined with semantic segmentation and recognition algorithms, extracts obstacle image textures and edge features to classify obstacle types. The RTK differential positioning module uniformly calibrates the precise center coordinates and location information of various obstacles. Through multi-source data fusion, the center coordinates, area, height, and circumscribed contour range of each static obstacle can be accurately extracted, achieving high-precision and comprehensive calibration of static obstacles in farmland. Finally, all static obstacle data is uniformly entered into the system to construct a field static obstacle coordinate database, providing reliable data support for early-stage obstacle avoidance and preventing operational routes from crossing obstacles at the source of route planning.
[0030] High-resolution aerial photography equipment was used to collect high-resolution images of farmland to extract crop growth characteristics. Combined with machine vision semantic segmentation algorithms, refined feature identification and quantitative extraction were performed on crops across the entire farmland area. Several core growth parameters were extracted, including average plant height, canopy foliage density, crop canopy coverage, plant density distribution, the location and area of bare land, and crop canopy closure. Through image analysis, visual states such as crop growth and field vegetation distribution were transformed into standardized and quantifiable numerical feature data, eliminating human judgment errors. This provides accurate and reliable core data for subsequent farmland operation area division, crop growth level classification, and differentiated plant protection parameter matching.
[0031] The process involves registration, fusion, noise reduction, and spatial mapping of farmland topography, static obstacles, and crop growth. This unifies scattered single-point data, image data, and coordinate data into the same spatial coordinate system, ultimately constructing a realistic 3D farmland scene that incorporates topographical features, obstacle boundaries, and crop semantic information. This step, through a progressive technical logic of multi-source data acquisition and 3D reconstruction, overcomes the shortcomings of traditional 2D planar modeling, which ignores differences in field micro-topography, crop growth, and the 3D range of obstacles.
[0032] S2: Based on the crop canopy coverage rate extracted from the reconstructed 3D farmland scene as the core zoning indicator, the plot boundaries are corrected through auxiliary features, and a pixel-level intelligent clustering algorithm is used to achieve irregular adaptive zoning of farmland, generating different sub-regions. The auxiliary features include the average plant height, plant density distribution, the location and area ratio of bare land in the field, and the density of crop canopy. The pixel-level intelligent clustering algorithm is a mature existing technology, based on classic unsupervised clustering methods such as K-means, fuzzy C-means (FCM), and DBSCAN, which are widely used in precision agriculture and UAV plant protection systems. This application does not improve the algorithm itself. For example, when the canopy coverage rate is 87%, it should be classified as a first-level densely planted area. However, there are large areas of bare land and sparse plants in the middle. Based on the auxiliary features, the contiguous bare land is cut out and classified as a third-level sparse area, and the remaining main body is classified as a first-level densely planted area.
[0033] Unevenly growing farmland was divided into three distinct, precisely defined, and non-overlapping sub-regions for operation. For each sub-region, based on crop growth characteristics, canopy penetration difficulty, and pesticide deposition requirements, differentiated flight path parameters were independently matched. This addresses the technical problems of poor adaptability and inconsistent operation quality inherent in traditional uniform operations. The specific sub-regions and their characteristics are as follows: Level 1 Dense Planting Area: This area features densely growing crops with overlapping and thick, tightly packed canopies, minimal gaps between rows, and a high degree of overall canopy closure. Specific quantitative criteria include: crop canopy coverage > 85%, higher average crop height, more layers of branches and leaves, and no obvious bare ground. This area exhibits a strong canopy shading effect, making it difficult for pesticide droplets to penetrate to the lower and middle layers of the crop and the underside of leaves. It is a high-risk area for missed spraying and insufficient efficacy, requiring enhanced penetration-type operation parameters. The formula for calculating crop canopy coverage is: Canopy Coverage = Crop Canopy Projected Area ÷ Total Statistical Area × 100%. Data is collected using airborne high-definition aerial photography equipment or lidar, and the canopy projected area and total area are calculated separately through image segmentation and point cloud analysis. Level II Conventional Zone: The crops in this zone are growing uniformly and regularly, with moderate density of branches and leaves, and a reasonable canopy structure. The uniformity of growth in the field is high, with no obvious problems of excessive density, excessive sparseness, or large areas of bare land. The quantitative judgment standard is: the crop canopy coverage rate is between 50% and 85%. The conditions for pesticide sedimentation and penetration in this zone are moderate. Conventional operating parameters (standard operating parameters) can meet the requirements for full coverage and uniform deposition without the need for significant parameter correction. This zone serves as the benchmark area for operating parameters across the entire region. Level 3 Sparse Zone: The crop plants in this zone are sparsely distributed, with large gaps between plants and weak crop growth. There are many exposed bare lands and blank areas between rows. The quantitative judgment standard is: crop canopy coverage rate <50%. This zone has a high proportion of bare land, thin crop canopy, and low resistance to pesticide penetration. If standard parameters are used in this zone, it is very easy to cause excessive overlap of spraying between rows and excessive accumulation of pesticide solution, which will lead to pesticide waste and soil residue problems. It is necessary to adopt cost-saving and efficiency-enhancing operation parameters.
[0034] S3: Based on different sub-regions, perform zone-specific differentiated route parameter matching; Different flight path parameters (agricultural drone operation parameters) are matched for different sub-regions to solve the problems of over-spraying and under-spraying at the root: For densely planted areas, addressing the challenges of pesticide penetration and the high risk of missed spraying on lower-layer plants, the following adjustments are made: 1. Based on the standard flight path spacing, the flight path spacing in densely planted areas is uniformly reduced by 15%-25%. This eliminates spraying gaps exceeding 8% between crop rows, increasing crop row coverage by over 20% and resolving localized missed spraying defects caused by canopy shading. For example, if the standard flight path spacing is 3.5m-4.0m, the adjusted spacing is 2.8m-3.2m. 2. Based on the standard operating height, the flight height in densely planted areas is reduced by 20%-40%, shortening the pesticide settling distance, reducing pesticide drift loss by over 30%, increasing effective pesticide deposition flux by 25%, and significantly enhancing penetration through thick canopies. For example, if the standard operating height is 2.0m-2.5m, the adjusted operating height is 1.2m-1.8m. 3. Based on the standard operating speed, the flight speed in densely planted areas is reduced by 40%-50%, and the spraying residence time per unit area is increased by 80%-100%, avoiding high-speed airflow from blowing away droplets and bending branches and leaves, and increasing the deposition rate of pesticide in the lower and middle layers of the canopy and the back of leaves by more than 35%; if the conventional operating speed is 4.0m / s-6.0m / s, the adjusted flight speed is 2.0m / s-3.0m / s.
[0035] In the secondary conventional zone, standard parameters are used as the benchmark for all parameters, ensuring a balance between efficacy and operational efficiency. 1. Standard flight path spacing is adopted, with a fluctuation error of ≤5%. For example, a standard flight path spacing of 3.5m-4.0m can be directly adopted. 2. Standard operating altitude is adopted, with a fluctuation error of ≤5%. For example, a standard operating altitude of 2.0m-2.5m. 3. Standard flight speed is adopted, with a fluctuation error of ≤5%. For example, a standard flight speed of 4.0-6.0m / s. The uniformity of spraying in this zone is ≤8%, the coverage rate of a single operation is ≥96%, there are no obvious problems of over-spraying or missed spraying, and it is suitable for most normally growing farmland.
[0036] For the Level 3 sparse area, to address the issues of pesticide waste and excessive pesticide accumulation in the soil, the following adjustments are made: 1. Based on the standard flight path spacing, the flight path spacing is widened by 10%-20%, effectively reducing the area of repeated spraying between rows and lowering the pesticide waste rate by more than 15%; for example, if the standard flight path spacing is 3.5m-4.0m, the adjusted flight path spacing is 3.8m-4.8m; 2. Based on the standard operating speed, the flight speed is increased by 20%-30%, reducing the operating time per unit area by 20% and lowering the overall operating energy consumption by 12%-18%; for example, if the standard operating speed is 4.0m / s-6.0m / s, the adjusted flight speed is 5.0m / s-7.5m / s; 3. The flight altitude adopts the standard operating altitude, with a fluctuation error of ≤5%. Maintaining the standard operating altitude avoids pesticide drift caused by excessive altitude and ensures that the pesticide deposition on sparse plants meets the standard for a single application; for example, the standard operating altitude is 2.0m-2.5m.
[0037] S4: Based on the differentiated route parameters of different sub-regions, generate serpentine routes that conform to the boundary contours of each sub-region, and integrate the straight edges, corners, and curved edges of farmland to form a circular transition route, generating a composite operation path; specifically including the following steps: S41: For main operation areas with regular farmland terrain and no obvious irregular turning points, based on the crop canopy coverage and auxiliary characteristics corresponding to the first-level dense planting area, the second-level regular area, and the third-level sparse area, a serpentine flight path is dynamically generated with varying distances to fit the irregular boundaries, ensuring that the density of the flight path perfectly matches the needs of the crop operation in the area. Specifically: In the first-level dense planting area, the flight path spacing is automatically reduced by 15%-25%, compensating for the lack of penetration of pesticides in the thick canopy by increasing the density of the flight path arrangement and eliminating blind spots in the spraying between crop rows; For the second-level regular area, the flight path spacing, flight altitude, and flight speed all use the baseline values (standard values), with fluctuation errors strictly controlled within 5%, ensuring a balance between pesticide efficacy and efficiency; For the third-level sparse area, the flight path spacing is automatically increased by 10%-20%, reducing overlapping areas of repeated spraying between rows and avoiding the problems of pesticide accumulation on bare ground and waste of pesticides; The system first acquires the vector boundaries of each irregular sub-region and constructs a closed electronic fence for each sub-region. Then, it automatically matches the corresponding flight path spacing parameters according to each sub-region. Using the longest axis of the partition as the operation direction, it generates a parallel serpentine flight path inside the fence. The flight path is strictly constrained within the boundary to ensure no boundary crossings and no blind spots. When the flight path crosses different sub-regions, it automatically switches the corresponding spacing parameters to achieve dynamic spacing across the entire area. Finally, it generates differentiated operation flight paths that fit the irregular boundaries and adapt to the crop growth. For example, when flying to a densely planted area, the flight path automatically becomes denser; when flying to a regular area, the flight path automatically returns to standard; and when flying to a sparse area, the flight path automatically widens. It flies along the edge of the irregular partition throughout the entire process, without crossing the boundary or missing any areas. This process achieves precise on-demand matching of spraying density in different growth areas and adapts to the spatial heterogeneity of crops in farmland from the perspective of flight path layout.
[0038] S42: For the edges of farmland plots, define straight edges, corners, and curved edges, and optimize flight routes accordingly. The specific steps are as follows: S421: The control system retrieves the reconstructed 3D model of the farmland, traverses the outer contour of the farmland point by point, collects basic information such as the plane coordinates and elevation of all boundary points, and completely restores the true shape of the plot. S422: Based on the collected plane coordinates, elevation, and other basic data of the points, a complete closed boundary outline of farmland plots is generated by fitting continuous coordinate points. Based on the generated boundary outline, the vector changes of adjacent coordinate points are analyzed to calculate the outline direction. According to the abrupt change characteristics of the angle between line segments and the deflection amplitude, the boundary outline is automatically divided into three types of boundary structures: straight edge, corner, and arc edge. Specifically, a straight edge is a straight segment with three or more consecutive adjacent line segments whose deflection amplitude is ≤8°, no abrupt change in direction, and constant curvature of 0; a corner is the intersection of two outline segments with a single deflection amplitude ≥30° and a single-point abrupt change in direction; an arc edge is a smooth curved outline segment with three or more consecutive adjacent line segments whose deflection amplitude is between 8° and 30°, no single-point abrupt change, consistent deflection direction, and continuously changing curvature. Simultaneously, elevation data is introduced to fit the terrain and correct the spatial position of the boundary points, eliminating the boundary spatial offset caused by terrain undulations, ensuring that the boundary restoration is true and accurate, and ensuring that subsequent route planning accurately matches the actual terrain. S423: Extract boundary features for three types of boundary structures: straight edges, corners, and curved edges. Boundary feature extraction includes curvature calculation and bending feature extraction by region. Specifically, for straight edges, since the curvature is always 0 and there are no bending features, there is no need to calculate the curvature separately, and the point coordinates and boundary type information are directly retained. For corners and curved edges: calculate the contour curvature point by point, quantify the degree of bending, and identify and store feature parameters such as bending direction, arc range, and maximum curvature value. S424: Based on the extracted boundary features, corresponding operational routes are generated according to type: Specifically, for straight edges: the shape is regular and the curvature is zero, and a straight route is generated directly along the boundary contour; for corner positions: the optimal arc transition radius is calculated by combining parameters such as the maximum curvature of the contour and the turning direction, and a smooth arc transition trajectory tangent to the preceding and following segments is constructed; for arc edges: using the measured curvature data of each point, a smooth curve route matching the curvature of the original boundary is fitted segment by segment; during the route planning process, the route coordinates are continuously constrained to ensure that all segments are accurately aligned with the land boundary, and the entire flight trajectory is continuous and smooth, without any abrupt changes or breaks.
[0039] During operation, the agricultural drone flies along a composite operational path. When encountering corners or curved changes, the drone smoothly and gradually changes direction following a flight path with a circular transition trajectory, thus ensuring stable flight attitude. Once the flight attitude stabilizes, the violent airflow fluctuations in the turning area are completely eliminated, preventing airflow disturbances from blowing away the pesticide. During operation, the nozzles maintain normal spraying status, and combined with the flight path trajectory that adheres to the boundary, it ensures uniform pesticide coverage in corner areas. Based on the above flight path design, it has the following advantages: reduced airflow disturbance, reducing pesticide drift in corner areas by more than 40%; no breaks or missed areas in the flight path, solving the problem of missed spraying in corners; complete pesticide coverage, eliminating the risk of pest and disease residue caused by missed spraying at the path level.
[0040] S5: Based on composite operation paths, it performs obstacle avoidance for static obstacles and generates multiple preliminary operation routes; the coordinates of static obstacles are marked in advance, and the initial routes avoid 100% of fixed obstacles without route conflicts; The specific steps are as follows: S51: Load information such as the center coordinates, floor area, height, and circumscribed contour of the static obstacle to complete spatial positioning; S52: Divide the complex operation path at the transition points between straight lines, arcs, and curves into straight sections, corner arc transition sections, and arc edge curve sections. S53: According to the flight sequence of the agricultural drone, conduct collision checks on each segment after it is split up, determine whether there is a collision with static obstacles such as trees, utility poles, and buildings, and mark the segments with risks. S54: For the marked risky flight segments, the re-planned operation route after detour is based on the composite operation path generated by S424, generating multiple preliminary operation routes; when the detour operation route passes through corners and arc boundaries, the design of smooth arc turning is continued; preferably, after the re-planning of the operation route after detour is completed, the connection of each flight segment is checked to see if it meets the requirements of continuity and smoothness. If it does, all flight segments are integrated to generate a preliminary operation route; otherwise, the re-planned operation route after detour is re-planned, and a preliminary operation route is generated after passing the retest; the above continuity detection method is to extract the three-dimensional coordinates of the beginning and end points of adjacent flight segments and calculate the distance. If the distance is within the error range, it is determined that there is no break in the connection; smoothness is detected by vector angle and curvature calculation. The flight direction angle and curvature change value at the connection are calculated. The arc transition form is checked in the corner and arc area. If there is no sudden change in angle or curvature, it is qualified.
[0041] This step enables full-scene static obstacle prediction before operation. For all static obstacles such as field utility poles, fixed trees, fences, ditches, and wells, avoidance paths that conform to the obstacle contours are planned in advance, adjusting the layout of the operation route from the source. Through full-domain pre-calibration optimization, a 100% static obstacle avoidance rate of the pre-planned route can be achieved, completely eliminating the overlap and conflict between the initial operation route and static obstacles, reducing the obstacle conflict rate of the initial operation route of the whole machine to 0%, effectively avoiding the problem of large-scale obstacle detours and overall route reconstruction during operation, and ensuring the integrity and regularity of the initial operation route.
[0042] S6: Select the final operation route based on multiple preliminary operation routes; Includes the following steps: S61: Energy consumption constraints. Under the premise of ensuring no blind spots in the entire site and no safety conflicts in flight routes, multiple preliminary operation routes are screened in the first round based on turning frequency and flight mileage. Redundant turns, invalid detours, and repeated flight segments are eliminated. Specifically, the processing is carried out in the following order: first, repeating flight segments are removed, then invalid detours are reduced, and finally, redundant turns are simplified. The rules of circular transition, obstacle avoidance, and full coverage are retained throughout the process. Redundant turns are identified as multiple turns over short distances, repeated changes of direction in the same area, and frequent fine-tuning of the course at small angles. Turns that could originally be completed by a single circular arc are broken into multiple turns. Invalid detours are identified as unnecessary detours caused by avoiding static obstacles, with the flight route deviating significantly from the optimal passage and the route being lengthened beyond the premise of sufficient safety distance. Repeated flight segments are identified as two or more flight segments with overlapping coordinates and consistent flight paths, with the UAV flying back and forth in the same area, resulting in repeated operations.
[0043] S62: Efficacy Constraints. Based on differentiated operational parameters for each zone, the spray coverage, deposition uniformity, and proportion of missed or over-sprayed areas of each initial operational route are quantitatively verified, eliminating initial operational routes that fail to meet efficacy standards. Strict efficacy thresholds are set to prevent sacrificing operational quality for the sake of low energy consumption. More specifically, the entire operational area and irregular sub-regions are uniformly divided into dense unit grids. The grid size is set in conjunction with the spray width of the agricultural drone and the crop plant type, using a single grid as the smallest detection and statistical unit to achieve accurate measurement across the entire area. This is while also incorporating differentiated operational parameters for each zone, such as flight altitude, spray flow rate, and flight speed. Different sub-regions use corresponding standards for calculation. Specific efficacy quantification constraints are as follows: The overall pesticide spraying coverage rate is ≥98%, completely eliminating large-scale blind spots. First, the total number of grids is counted, and then the number of grids effectively covered by the pesticide is counted. The overall pesticide spraying coverage rate = number of effectively covered grids / total number of grids in the area × 100%. A result of ≥98% is considered qualified. If it is lower than this value, it indicates the existence of large-scale blind spots, and the flight path is invalid. The number of grids effectively covered by the pesticide is determined by the spray diffusion model and the spatial position of the UAV. The percentage of missed spraying areas is ≤1.5%, addressing the problem of pest and disease residues caused by insufficient canopy penetration in densely planted areas. The percentage of missed spraying is calculated as: number of missed spraying grids / total number of grids in the area × 100%. A result ≤1.5% is considered acceptable, while a result greater than 1.5% results in invalidation of the flight path. For densely planted areas, the canopy penetration model is used for verification to avoid situations where pesticides are present on the surface but missed in the lower layers. The canopy penetration model divides crop plants into upper, middle, and lower layers. It calculates the remaining amount of pesticide after droplets penetrate each layer of leaves based on a mature pesticide attenuation formula, while also incorporating the influence of the downwash airflow from the UAV rotor for correction. This determines whether the amount of pesticide deposited in the lower layer of the crop meets the standard and whether there are any missed spraying issues.
[0044] The proportion of areas exceeding the standard for over-spraying is ≤2%, avoiding the problems of pesticide accumulation and waste in sparse and bare areas and soil residue. Grids with pesticide deposition exceeding the upper limit of the standard dosage are judged as grids exceeding the standard for over-spraying. Multiple overlapping flight segments and excessively small flight path spacing are the main causes of this area. The proportion of over-spraying is calculated as the number of grids exceeding the standard / the total number of grids in the area × 100%. A result of ≤2% is acceptable, reducing pesticide waste and soil pesticide residue. If the result is greater than 2%, the flight path is invalidated. The grid space is matched based on the flight segment trajectory and spray width. The basic pesticide deposition amount under a single flight segment is calculated using a spray diffusion model. At the same time, the rotor downwash airflow and crop canopy parameters are introduced for correction to obtain the single deposition amount of the flight segment in the corresponding grid. If the grid is covered by multiple flight segments, all single deposition amounts are accumulated and compared with the maximum allowable dosage. If the single deposition amount exceeds the standard, the grid is judged as a grid exceeding the standard for over-spraying.
[0045] The efficacy indicators in this step are far superior to those of traditional fixed-route operation schemes, ensuring the quality of plant protection operations in different growth areas in all aspects.
[0046] S63: To achieve the final operation route selection, for the remaining preliminary operation routes, a weighted calculation and comprehensive score are performed on multi-dimensional parameters such as energy consumption loss, drug efficacy accuracy, route deviation error, and operation adaptability. The comprehensive scores of each route are compared, and the scheme with the highest score is selected as the only final operation route. This scheme has precise selection logic and comprehensive constraints, and the scheme selection accuracy rate can reach 100%, providing accurate and standardized optimal route samples for subsequent site operation data review and iteration and long-term optimization.
[0047] The specific calculation method is as follows: , This represents the overall score. Indicates energy consumption points. This indicates the weight corresponding to the energy consumption score. Indicates the precision of drug efficacy. This indicates the weight corresponding to the drug efficacy accuracy score. Indicates the deviation of the flight path. This indicates the weight corresponding to the flight path offset. Indicates job suitability score. This indicates the weight corresponding to the job suitability score; the recommended partition weights are as follows, prioritizing densely planted areas with high efficacy. =0.2、 =0.5、 =0.2、 =0.1; A balanced, normal region. =0.35、 =0.35、 =0.15、 =0.15; sparse regions prioritizing energy conservation. =0.5、 =0.25、 =0.15、 =0.10; Energy consumption E represents actual energy consumption. This represents the theoretical minimum energy consumption across the entire domain. This indicates the maximum allowable energy consumption limit; the higher the energy consumption score, the lower the energy consumption.
[0048] Drug efficacy precision C represents the actual coverage rate of the pesticide spray. Indicates the minimum acceptable threshold. The ideal coverage rate is indicated by the actual pesticide spraying coverage rate as the scoring basis. A minimum qualified threshold and the ideal coverage rate are preset. If the actual pesticide spraying coverage rate is lower than the minimum qualified threshold, a score of zero is given and the operation route is removed. When the actual coverage rate is between the minimum qualified threshold and the ideal coverage rate, the score increases linearly with the actual pesticide spraying coverage rate. The higher the actual pesticide spraying coverage rate reaches or exceeds the ideal coverage rate, the higher the efficacy score is capped at full marks.
[0049] Flight path deviation H represents the average offset of the operation route relative to the boundary of the farmland plot; The maximum allowable offset limit is set; the smaller the deviation from the boundary, the higher the score.
[0050] Job adaptability score Scoring is based on the smoothness of cornering and the degree of conformity to curves: When adhering to the boundary and turning without sudden braking, More specifically, corners utilize tangent circular arc transitions, with the arc radius matching the contour curvature; there are no abrupt changes in the turning angle at any single point, the flight path closely follows the boundary, and the average offset H ≤ H allow H allow H is the threshold of the excellent dividing line. allow < There is no short-distance, continuous, or frequent reversal.
[0051] When making a slight, unreasonable turn The value ranges from 0.4 to 0.9; specifically, the radius of the corner arc deviates from the optimal value, but no sharp right-angle turn occurs; small local segments of the route deviate slightly from the boundary H. allow <H≤ = , This indicates the safe critical distance for the drone; there are a few instances of unreasonable local turns, but no close-range continuous sharp turns or risk of collision; the criteria for judging a few instances of unreasonable local turns are as follows: N represents the total number of turning points at the boundary of this segment, and n represents the number of unreasonable turning points.
[0052] When there is a risk of collision due to frequent sharp turns, <0.4; Specifically, this refers to: no rounded transition at corners, direct right-angle zigzag flight; multiple large-scale changes of direction within a short distance; and flight path deviation exceeding the limit H> Being too close to obstacles poses a risk of scraping.
[0053] S7, when operating based on the final work route, performs real-time obstacle avoidance for dynamic obstacles; accurately avoids sudden interference in the field. The farmland operation environment is highly uncertain. Unpredictable dynamic obstacles frequently appear in the fields, such as birds, temporarily piled farm tools, scattered debris, and sudden moving obstructions. Traditional static obstacle avoidance systems cannot identify such sudden disturbances, easily leading to flight collisions, emergency hovering, and flight path interruptions. This application utilizes an UAV-borne LiDAR and a high-definition visual recognition module to construct a high-frequency real-time obstacle perception mechanism. It continuously scans the work area ahead of the flight path throughout the operation, identifying sudden dynamic obstacles in real time. This dynamic recognition system boasts ultra-high precision and extremely fast response capabilities, with an obstacle recognition accuracy of ≥98% and a single obstacle avoidance trigger response time of ≤0.2s. It can quickly identify various temporary, dynamic, and sudden field disturbances and instantly generate small-scale avoidance paths using parallel offset or small-radius circular detour methods, while maintaining the original smooth flight path design throughout. Without disrupting the overall flight path layout, it achieves safe obstacle avoidance, overcoming the technical shortcomings of traditional obstacle avoidance systems in handling sudden field conditions, and significantly improving the safety and environmental adaptability of complex farmland operations.
[0054] S7 includes the following specific steps: S71, the agricultural drone flies along the final operation route, and the drone’s onboard lidar and high-definition visual recognition module continuously scan the operation area ahead of the route in real time. S72, when detecting dynamic obstacles, combines the current position of the agricultural drone, the coordinates of the forward flight segment, and the spatial data of the obstacle to calculate the shortest distance between the operation route and the obstacle in real time. S73, when the shortest distance is less than the preset safe critical distance for the drone. At that time, local obstacle avoidance is triggered; S74: Centered on the obstacle, the length of the rerouting route near the obstacle should not exceed 5% of the entire operation route. Within the designated 5% local area, the path is fine-tuned using two methods: parallel offset and small-radius circular arc detour. This step ensures that the offset amplitude is controlled within a small range, without large-angle or long-distance rerouting, and retains the original smooth arc turning shape throughout, without adding right-angle turns or redundant segments. Differentiated operation parameters such as plot zoning, flight speed, and spray flow rate remain completely unchanged. The coordinate difference between the fine-tuned avoidance operation route and the original operation route is compared to calculate the overall offset error, ensuring that the global offset error is ≤3%. If the offset error exceeds the threshold, readjustment is required. S75, the agricultural drone executes an obstacle avoidance operation route; once it flies outside the obstacle's influence range, the control system activates the trajectory matching algorithm and coordinate difference calibration module to gradually align with the original operation route; specifically, the control system retrieves the coordinate sequence, heading, and curvature data of the final operation route in S6, performs point-by-point matching between the current flight trajectory and the original trajectory, calculates the coordinate deviation between the current point and the corresponding point on the original trajectory, and corrects the flight attitude and direction of travel point by point; After completing coordinate calibration, the agricultural drone returns to the final operating route of S6; all sub-area operating parameters are restored to the state of S4. This step effectively avoids problems such as large-scale flight path confusion and zoned operation failure, with the increase in the area of repeated operations ≤2%. It retains the optimization advantages of the previous serpentine flight path and smooth arc turns to the greatest extent, ensuring that the uniformity of spraying across the entire area and the accuracy of zoned operations are not compromised. This step achieves a safe operation effect throughout the entire process, avoiding static obstacles before operation, avoiding sudden dynamic obstacles during operation, and being able to restore the final operation flight path after avoiding dynamic obstacles. Under the premise of ensuring the flight safety of the UAV, the final operation flight path in S6 is retained to the greatest extent, without compromising the uniformity of spraying of pesticides across the entire area and the zoned operations. Example 2
[0055] This embodiment provides a specific application scenario: Typical irregular composite farmland was selected as the experimental plot: The farmland plot has a total area of 12.8 mu (approximately 1.8 hectares), and is generally irregularly shaped like a round arc, including field ridges and local uneven micro-topography; There are static obstacles in the farmland: 3 utility poles and 2 scattered trees. There is no large area of water accumulation or extremely steep slopes. The crop is mature corn, and the growth in the field is uneven, with dense areas, normal areas and sparse bare areas, which is fully compatible with the hierarchical planning logic of this plan.
[0056] Set the baseline operating parameters for agricultural drones: baseline row spacing 3.8m, baseline flight altitude 2.2m, baseline flight speed 5.0m / s; preset the qualified operating thresholds: full area spraying coverage ≥96%, missed spraying rate ≤2%, energy consumption deviation ≤10%.
[0057] S1: The RTK differential positioning module, LiDAR and high-definition aerial photography equipment complete the full-domain scanning, collect three types of multi-source data: farmland terrain, static obstacles and crop growth, and complete the reconstruction of the three-dimensional scene of farmland based on the collected multi-source data; S2: Based on the crop canopy foliage coverage extracted from the reconstructed 3D farmland scene, the core zoning indicator is used. Auxiliary features such as average crop height, crop density distribution, the location and area of bare land, and crop canopy density are used to correct plot boundaries. Combined with a pixel-level intelligent clustering algorithm, irregular adaptive zoning of farmland is achieved, generating different sub-regions, as follows: Level 1 dense planting area: The core area in the center of the plot, with a canopy coverage of 88%, accounting for 42% of the total area; the branches and leaves are thick and the canopy is dense, making it difficult for pesticides to penetrate and easy to miss the bottom layer. Second-level conventional area: The upper part of the plot is evenly planted, with a canopy coverage of 72%, accounting for 38% of the total area; the crops are growing evenly, with no obvious defects in density; Level 3 sparse area: the edge of the plot and the area around the field ridge, with a canopy coverage of 41%, including a small amount of bare land, accounting for 20% of the total area; the plants are sparse, and conventional operations are prone to waste of pesticides and repeated spraying.
[0058] S3: Based on different sub-regions, perform zone-specific differentiated route parameter matching; For the first-level densely planted area, gradient optimization adjustments were implemented according to the standard plan, with row spacing reduced by 20%, flight height reduced by 30%, and flight speed reduced by 45%. The adjusted parameters are: row spacing 3.04m, flight height 1.54m, and flight speed 2.75m / s. Based on the above adjustments, the blank spraying zone between rows can be eliminated by 9%, pesticide drift loss can be reduced by 32%, and spraying time per unit area can be increased by 90%, completely solving the problem of missed spraying at the bottom layer of thick canopy. For the secondary conventional areas, the above-mentioned pre-set benchmark operating parameters for agricultural drones are used to control the fluctuation error to ≤5%; specifically, the row spacing is 3.8m, the flight altitude is 2.2m, and the flight speed is 5.0m / s, to ensure a balance between pesticide efficacy and operational efficiency. For the sparse areas of level three, gradient optimization adjustments were implemented according to the standard plan, with row spacing increased by 15%, flight speed increased by 25%, and operating height remaining unchanged. The adjusted parameters are: row spacing 4.37m, flight height 2.2m, and flight speed 6.25m / s. Based on the above adjustments, the area to be sprayed repeatedly can be reduced by 18%, the waste rate of pesticide solution can be reduced by 16%, the overall energy consumption of the operation can be reduced by 15%, and the waste of pesticide accumulation on bare ground can be eliminated.
[0059] S4: Based on the differentiated route parameters of different sub-regions, generate serpentine routes that fit the boundary contours of each sub-region, and integrate the straight edges, corners and arc edges of farmland to form a circular transition route to generate a composite operation path. For the main regular area of the plot, a zoned variable spacing serpentine flight path is adopted to match the differentiated row spacing of the three zones; For the rounded corners and irregular inflection points of the plot, a smooth rounded transition flight path is generated, reducing the abrupt change rate of the turning trajectory to 0% and the drift of the medicine at the corners to 42%; All turning sections have a uniform speed reduction of 35%, the machine tilt error is controlled within ±3°, the spray coverage rate in turning areas is increased by 28%, and blank areas in corners and edges are completely eliminated.
[0060] S5: Based on the composite operation path, it performs obstacle avoidance for static obstacles and generates 4 preliminary operation routes; the coordinates of utility poles and tree obstacles are marked in advance, and the initial routes avoid 100% of fixed obstacles without route conflicts. S6: Based on multiple preliminary operation routes, the final operation route is selected. In this embodiment, 44 preliminary operation routes are selected by weighted calculation and comprehensive scoring of multiple parameters such as energy consumption dimension constraints, efficacy dimension constraints, energy consumption loss value, efficacy attainment accuracy, route deviation error, and operation adaptability to select a unique final operation route: the total mileage of the final route is shortened by 12%, the energy consumption of the whole machine is reduced by 16%, the coverage rate of the whole area spraying is 98.7%, the proportion of missed spraying is 1.1%, the proportion of overspraying exceeding the standard is 1.5%, and the uniformity error of liquid deposition is 5.2%, all of which are better than the preset qualified threshold.
[0061] When operating based on the final operational route, the S7 performs real-time obstacle avoidance for dynamic obstacles and dynamically corrects the local path around obstacles within 5%; during flight, the radar identifies temporary debris scattered in the field and responds quickly to local obstacle avoidance in 0.18 seconds; after obstacle avoidance, it returns to the original route with 100% accuracy and the global trajectory deviation error is only 2.1%.
[0062] The data comparison between this embodiment and the traditional solution is shown in Table 1 below: Table 1:
[0063] Compared to the traditional unified operation mode, this embodiment completely solves industry pain points such as missed spraying in dense areas, waste in sparse areas, failure at corners, high energy consumption, and poor uniformity.
[0064] In summary, due to the adoption of the above technical solution, this application has the following advantages: 1. By fusing multi-source heterogeneous data from lidar, RTK positioning, and visual images, a three-dimensional farmland scene including terrain, static obstacles, and crop growth is constructed, breaking through the limitations of traditional two-dimensional planar planning and accurately distinguishing densely planted areas, conventional areas, and sparse areas, providing a data foundation for precise zoned operations. 2. Divide the operation area into three levels based on the crop canopy coverage rate as a quantitative standard, and match the corresponding flight path spacing, flight altitude and flight speed according to different growth conditions to solve the core problems of missed spraying in densely planted areas, waste of pesticide in sparsely planted areas and uneven pesticide efficacy in conventional areas from the source. 3. Dynamic pitch generation creates a serpentine flight path that fits irregular boundaries, and combines it with a smooth arc transition structure to generate a composite operation path. This eliminates the traditional 90° right-angle turn, reduces pesticide drift, and, with turn speed reduction and pressure stabilization control, significantly improves the spray coverage in corner areas. The flight is stable throughout and the spray uniformity is significantly improved. 4. Prevent static obstacles in advance and respond quickly to dynamic obstacles in the field during operation. For dynamic obstacles, only make minor adjustments to the flight path in a small area around the obstacle, and accurately return to the original flight path after bypassing the obstacle. There is no global flight path disorder, which takes into account both flight safety and the integrity of the overall operation layout. 5. All zoning standards, operation parameters, obstacle avoidance response, and iteration range are quantified using percentage indicators. The system executes automatically without manual intervention. Key indicators such as operation coverage, missed spraying rate, and energy consumption are superior to existing technologies, making it suitable for the needs of large-scale and precise modern agricultural plant protection.
[0065] The devices and connections not specifically described above are all prior art, and will not be described in detail here.
[0066] The preferred embodiments of this application have been described in detail above. However, this application is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this application, various simple modifications can be made to the technical solution of this application, and these simple modifications all fall within the protection scope of this application.
[0067] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable way without contradiction. In order to avoid unnecessary repetition, the various possible combinations in this application will not be described separately.
[0068] Furthermore, various different implementations of this application can be combined in any way, as long as they do not violate the spirit of this application, and such combinations should also be regarded as the content disclosed in this application.
Claims
1. A method for planning the operational path of a plant protection drone, characterized in that, Includes the following steps: S1: Based on data collected by lidar, high-definition aerial photography equipment, and RTK differential positioning module, extract three types of feature information: farmland terrain, static obstacles, and crop growth, and complete the reconstruction of the three-dimensional farmland scene; S2: The crop canopy coverage rate extracted from the reconstructed 3D farmland scene is used as the core indicator for zoning. The plot boundaries are corrected by auxiliary features, and the irregular adaptive zoning of farmland is achieved by combining intelligent clustering algorithm to generate different sub-regions. Auxiliary features include average crop height, plant density distribution, location and area of bare land in the field, and crop canopy density. The sub-regions include: Level 1 dense planting area, with crop canopy coverage > 85%; and Level 2 conventional planting area, with crop canopy coverage 50%–85%. Level III sparse area, crop canopy coverage <50%; S3: Based on different sub-regions, perform zone-specific differentiated route parameter matching; S4: Based on the differentiated route parameters of different sub-regions, generate serpentine routes that fit the boundary contours of each sub-region, and integrate the straight edges, corners and arc edges of farmland to form a circular transition route to generate a composite operation path. S5: Under the premise of a complex operation path, perform obstacle avoidance for static obstacles and generate multiple preliminary operation routes; S6: Based on multiple preliminary operation routes, complete the final operation route selection.
2. The method for planning the operation path of a plant protection drone according to claim 1, characterized in that, It also includes S7: when operating based on the final work route, it performs real-time obstacle avoidance for dynamic obstacles.
3. The method for planning the operation path of a plant protection drone according to claim 1, characterized in that, The standard for matching the zone-differentiated route parameters in S3 is as follows: In the first-level dense planting area, based on the standard flight path spacing, the flight path spacing is reduced by 15%–25%; based on the standard operating altitude, the flight altitude is reduced by 20%–40%; and based on the standard operating speed, the flight speed is reduced by 40%–50%. In the Level 2 standard zone, the spacing between flight paths shall be the standard spacing with a fluctuation error of ≤5%; the flight altitude shall be the standard operating altitude with a fluctuation error of ≤5%; and the flight speed shall be the standard flight speed with a fluctuation error of ≤5%. In the Level 3 sparse zone, the standard flight path spacing is used as the benchmark, with the flight path spacing increased by 10%–20%; the flight altitude adopts the standard operating altitude, with a fluctuation error of ≤5%; and the flight speed is increased by 20%–30% based on the standard operating speed.
4. The method for planning the operation path of a plant protection drone according to claim 1, characterized in that, S4 includes: S41: For the main operation area of farmland, based on the crop canopy coverage and auxiliary characteristics corresponding to the first-level dense planting area, the second-level conventional area, and the third-level sparse area, a serpentine flight path that fits the irregular boundary is dynamically generated by varying the distance. S42: For the edges of farmland plots, divide them into straight edges, corners, and curved edges, and optimize flight routes accordingly.
5. The method for planning the operation path of a plant protection drone according to claim 4, characterized in that, S42 includes the following steps: S421: The control system retrieves the reconstructed 3D model of the farmland, traverses the outer contour of the farmland point by point, collects the plane coordinates and elevation information of all boundary points, and completely restores the true shape of the plot. S422: Based on the collected basic data, a complete closed boundary contour line of farmland plots is generated by fitting continuous coordinate points. Based on the generated boundary contour line, the vector changes of adjacent coordinate points are analyzed to calculate the contour direction. According to the abrupt change characteristics of the angle between line segments and the deflection amplitude, the boundary contour is automatically divided into three types of boundary structures: straight edge, corner, and arc edge. S423: Extract boundary features for three types of boundary structures: straight edges, corners, and curved edges. The extraction of boundary features includes curvature calculation and bending feature extraction in different regions. S424: Based on the extracted boundary features, generate corresponding operation routes according to type.
6. The method for planning the operation path of a plant protection drone according to claim 5, characterized in that, In S422, a straight edge is a straight segment with three or more consecutive adjacent line segments whose deflection amplitude is ≤8°, no abrupt change in direction, and constant curvature; a corner is the intersection of two contour segments with a single deflection amplitude ≥30° and a single-point abrupt change in direction; and an arc edge is a smooth curved contour segment with three or more consecutive adjacent line segments whose deflection amplitude is between 8° and 30°, no single-point abrupt change, consistent deflection direction, and continuously changing curvature.
7. The method for planning the operation path of a plant protection drone according to claim 5, characterized in that, S5 includes: S51: Load the center coordinates, floor area, height, and circumscribed contour of the static obstacle to complete spatial positioning; S52: Divide the complex operation path at the transition points between straight lines, arcs, and curves into straight sections, corner arc transition sections, and arc edge curve sections. S53: According to the flight sequence of the agricultural drone, conduct collision checks on each segment after it is split up, determine whether there is a collision with a static obstacle, and mark the segments with risks. S54: For the marked risky segments, the operation route after detour is replanned based on the composite operation route generated by S424, generating multiple preliminary operation routes.
8. The method for planning the operation path of a plant protection drone according to claim 5, characterized in that, S6 includes the following steps: S61: Energy consumption constraints. Under the premise of ensuring no blind spots in the operation of the entire plot and no safety conflicts in the flight routes, the first round of screening of multiple preliminary operation routes is carried out based on turning frequency and flight mileage, and the preliminary operation routes with redundant turns, invalid detours and repeated segments are eliminated. S62: Efficacy dimension constraint. Based on the differentiated operation parameters of each zone, the spray coverage effect, deposition uniformity and the proportion of missed and re-sprayed areas of each preliminary operation route are quantitatively verified, and preliminary operation routes with substandard efficacy are eliminated. S63: For the remaining preliminary operation routes, a weighted calculation and comprehensive score are performed on multiple parameters such as energy consumption loss, drug efficacy accuracy, route deviation error, and operation adaptability to select a unique final operation route.
9. The method for planning the operation path of a plant protection drone according to claim 8, characterized in that, The standard for quantization verification in S62 is: The overall pesticide spraying coverage rate is ≥98%: First, count the total number of grids, then count the number of grids effectively covered by the pesticide. The overall pesticide spraying coverage rate = number of effectively covered grids / total number of grids in the area × 100%. A result of ≥98% is considered qualified. The percentage of unsprayed areas is ≤1.5%: Unspray percentage = number of unsprayed grids / total number of grids in the area × 100%, and the result is ≤1.5% to be considered acceptable; The percentage of areas with excessive respraying is ≤2%: Percentage of areas with excessive respraying = Number of grids with excessive respraying / Total number of grids in the area × 100%. A result of ≤2% is considered acceptable.
10. The method for planning the operation path of a plant protection drone according to claim 8, characterized in that, The calculation method for S63 is as follows: in, This represents the overall score. Indicates energy consumption points. This indicates the weight corresponding to the energy consumption score. Indicates the precision of drug efficacy. This indicates the weight corresponding to the drug efficacy accuracy score. Indicates the deviation of the flight path. This indicates the weight corresponding to the flight path offset. Indicates job suitability score. This indicates the weight corresponding to the job suitability score; Energy consumption E represents actual energy consumption. This represents the theoretical minimum energy consumption across the entire domain. Indicates the maximum allowable energy consumption limit; Drug efficacy precision C represents the actual coverage rate of the pesticide spray. Indicates the minimum acceptable threshold. This indicates the ideal coverage rate; Flight path deviation H represents the average offset of the operation route relative to the boundary of the farmland plot; Maximum allowable offset limit; Job adaptability score Scoring is based on the smoothness of cornering and the degree of conformity to the curve: conformity to the boundary and no sudden braking during cornering. When making a slight, unreasonable turn, The value should be between 0.4 and 0.9; when there are frequent sharp turns and a risk of scraping, <0.4.
