Super-low-altitude spraying flight control method for large unmanned aerial vehicle

Abstract

The invention discloses an ultra-low-altitude spraying flight control method for a large unmanned aerial vehicle, and relates to the field of unmanned aerial vehicle control. The method comprises the steps of firstly obtaining unmanned aerial vehicle performance parameters and working environment parameters; then fusion decision planning is carried out based on the parameters, and flight and spraying control instructions are generated; and finally executing the instruction to complete the precise operation. The method is characterized in that through global static path planning and obstacle avoidance, a night sensing mode fusing laser radar, thermal imaging and millimeter wave radar data is combined, and a hierarchical obstacle avoidance decision is executed according to the comprehensive confidence coefficient of an obstacle and predicted collision time; meanwhile, feedforward control based on instantaneous track curvature is introduced into flight control, and flow of nozzles on the inner side and the outer side is distributed according to the same curvature difference in spraying control; the problems of poor sensing reliability of night operation, weak wind disturbance resistance of ultra-low-altitude flight and uneven turning spraying are solved, and safe, efficient and accurate spraying operation of the large unmanned aerial vehicle under the all-weather condition is achieved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control, and more particularly to a method for controlling the ultra-low-altitude spraying flight of a large UAV. Background Technology

[0002] With the development of precision agriculture and intelligent aviation technology, large unmanned aerial vehicles (UAVs) have become core equipment for ultra-low-altitude spraying operations in agriculture and forestry due to their high efficiency, high maneuverability, and low operating costs. Compared with manned aircraft, UAVs can achieve autonomous flight with centimeter-level precision, significantly reducing pesticide use and improving operational quality and economic benefits. Ultra-low-altitude flight can make full use of downdrafts, enhancing pesticide penetration, which is key to achieving efficient and precise spraying. Furthermore, to avoid the adverse effects of high daytime temperatures, rapid evaporation, and gusts, extending spraying operations to nighttime has become an urgent industry need to improve operational efficiency and resource utilization. Therefore, developing reliable ultra-low-altitude flight control technology for large UAVs that can operate stably under all weather conditions is of great significance for ensuring food security and sustainable agricultural development. Currently, traditional UAV ultra-low-altitude spraying control systems typically employ an automated operation mode of preset flight paths and fixed-point execution. This system relies on high-precision GNSS / RTK positioning technology for global navigation, controlling the UAV's flight and simultaneously activating and deactivating the spraying system based on a pre-planned digital flight path. However, this method is essentially an open-loop control based on a static map, lacking the ability to autonomously respond to dynamically appearing obstacles (such as power lines and birds) and real-time changing terrain during the operation, resulting in insufficient intelligence. However, the shortcomings of existing technologies lie in the severe unevenness of their environmental perception systems between day and night and in different weather conditions, resulting in insufficient reliability of perception at night and in complex weather conditions, which cannot support safe closed-loop obstacle avoidance decisions. At the same time, the flight control system lacks an effective wind disturbance compensation mechanism, which leads to a decrease in the accuracy of ultra-low altitude track tracking. Furthermore, the spraying system is decoupled from the flight attitude and cannot adapt to turning conditions, resulting in uneven spraying coverage on curves. Ultimately, this restricts the ability of large UAVs to achieve safe, efficient, and all-weather precision operations. Summary of the Invention

[0003] The objective of this invention can be achieved through the following technical solution: a flight control method for ultra-low-altitude spraying of large unmanned aerial vehicles, comprising the following steps; S100: Acquire the performance parameters of the UAV and the environmental parameters of the task area. S200 performs fusion analysis and decision planning based on the performance parameters of the UAV and the environmental parameters of the operational task area to generate flight control commands and spraying control commands; specific steps include: A path search algorithm is used to generate a global flight path based on the work area information, and static obstacle avoidance optimization is performed on the path based on static obstacle information; During flight, when the ambient light intensity is lower than the preset light threshold, the night perception mode is automatically activated. In the night perception mode, the perception results of lidar, thermal imaging and millimeter-wave radar are fused to calculate the comprehensive confidence level of the obstacle and the estimated collision time, and graded obstacle avoidance decisions are made accordingly. The target flow rate is calculated based on the target usage per acre, real-time flight speed and effective spray width, and the ray method is used to determine the positional relationship between the real-time position of the UAV and the boundary of the work area, thereby generating nozzle start and stop commands; The trajectory tracking command of the generated global flight path and the obstacle avoidance command generated in real time are combined to generate the final flight control command through a command priority arbitration mechanism. The command priority arbitration mechanism means that when an obstacle avoidance command exists, the obstacle avoidance command is used to override the trajectory tracking command and is used as the final flight control command; when no obstacle avoidance command exists, the trajectory tracking command is used as the final flight control command. The S300 executes flight control commands and spray control commands to drive the UAV to complete precise flight and adaptive variable spraying operations. The flight control commands include generating feedforward roll angle commands based on the instantaneous trajectory curvature and adjusting the flight speed according to the instantaneous trajectory curvature. The spray control commands include allocating different flow rates to the nozzle groups on the inside and outside of the curve based on the instantaneous trajectory curvature.

[0004] Preferably, the performance parameters and environmental parameters in S100 are as follows: The performance parameters include flight speed, remaining battery power, and estimated flight time. The environmental parameters include information on the boundaries, area, and static obstacles of the work area obtained through preliminary surveys, as well as wind speed and direction obtained in real time through airborne sensors.

[0005] Preferably, the static obstacle avoidance optimization specifically includes: The pre-stored 3D obstacle model is projected onto a 2D grid map and then expanded. The expansion radius is the sum of the radius of the UAV's circumcircle, the preset safety margin, and the standard deviation of the GNSS horizontal positioning accuracy. The preset safety margin ranges from 2 meters to 5 meters and is used to compensate for unmodeled dynamic uncertainties and control errors.

[0006] Preferably, the preset illumination threshold is as follows: The illumination threshold is set to 10 Lux, based on the fact that this value is determined by the performance critical point of industrial-grade vision sensors. When the ambient light is below this critical point, the recognition reliability of the vision sensor decreases significantly. The system switches the perception mode at this threshold to ensure that LiDAR and thermal imaging are enabled as the main sensing sources before visual information fails.

[0007] Preferably, the comprehensive confidence level is specifically: The confidence scores of LiDAR, thermal imaging, and millimeter-wave radar are combined to obtain the overall confidence score of the obstacle; and in nighttime perception mode, the weight of thermal imaging confidence score is set to be no less than that of LiDAR confidence score.

[0008] Preferably, the graded obstacle avoidance decision includes: If the overall confidence level is greater than the high confidence level threshold and the estimated collision time is less than the emergency time threshold, an emergency climb command is generated. If the overall confidence level is greater than the high confidence level threshold and the estimated collision time is less than the normal time threshold but greater than the emergency time threshold, a yaw detour instruction is generated. If the overall confidence level is greater than the low confidence threshold but less than the high confidence threshold, a cautious passage instruction is generated. If the overall confidence level is less than the low confidence level threshold, a pause autonomous decision-making instruction is generated, and obstacle information is pushed to the ground station for operator decision-making.

[0009] Preferably, the trajectory tracking command is calculated using a feedforward-feedback composite control method; wherein, the feedforward control quantity is used to actively compensate for the heading deviation caused by real-time wind speed and direction.

[0010] Preferably, the method for calculating the instantaneous track curvature is as follows: Extract three consecutive waypoints from the current flight path of the UAV to form a local sequence, calculate the curvature of the osculating circle determined by the three points, and use the curvature as the instantaneous track curvature.

[0011] Preferably, the different flow rates allocated to the nozzle groups on the inner and outer sides of the curve are specifically as follows: Based on the instantaneous trajectory curvature, a flow correction coefficient greater than 1 is assigned to the outer nozzle, while a flow correction coefficient less than 1 is assigned to the inner nozzle; and the difference between the flow correction coefficients of the inner and outer nozzles is monotonically increasing with the absolute value of the instantaneous trajectory curvature.

[0012] Preferably, the method further includes a security monitoring and emergency decision-making step, specifically: Signal interruption response: When the duration of the communication link interruption with the server exceeds the disconnection safety threshold, control the drone to execute the predetermined return-to-home procedure; Tiered decision-making: Continuously monitor real-time wind speed and execute tiered decisions: When the wind speed exceeds the operational wind speed threshold but does not exceed the ultimate safe wind speed threshold, a deceleration command is generated to attempt to maintain the operation while ensuring flight safety. When the wind speed exceeds the aforementioned limit safe wind speed threshold, an instruction is generated to suspend operations and immediately return to base.

[0013] The beneficial effects of this invention are: (1) By creating a nighttime perception mode and dynamically fusing data from lidar, thermal imaging and millimeter-wave radar, this invention constructs a graded obstacle avoidance decision-making mechanism based on comprehensive confidence and estimated collision time. This effectively overcomes the perception limitations of a single sensor in complex environments such as nighttime and fog, significantly improves the reliability of obstacle (especially thin wire) identification, and enables the UAV to have a truly all-weather safe and autonomous operation capability. (2) By introducing feedforward compensation based on real-time wind field and feedforward control command based on instantaneous trajectory curvature into the feedforward-feedback composite control, the system can actively counteract the effects of wind disturbance and turning dynamics, greatly enhancing the trajectory tracking accuracy and flight stability of UAV in ultra-low altitude complex environment, and providing a reliable flight platform for precision spraying. (3) By establishing a linkage mechanism between instantaneous trajectory curvature and nozzle flow correction coefficient, the flow of the inner and outer nozzles can be dynamically adjusted according to the turning amplitude, which fundamentally compensates for the uneven coverage caused by the difference between the inner and outer diameters of the turn. While improving the spraying quality, it effectively reduces pesticide waste and achieves a balance between high efficiency and economy. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a schematic diagram of the principle of the ultra-low-altitude spraying flight control method for large unmanned aerial vehicles (UAVs) of the present invention.

[0016] Figure 2 This is a flowchart of the ultra-low-altitude spraying flight control method for large unmanned aerial vehicles (UAVs) according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention provides a method for controlling ultra-low-altitude spraying flight of large unmanned aerial vehicles (UAVs), which is described below in conjunction with... Figure 1 and Figure 2This embodiment describes the specific implementation method in detail. It uses a large farm and vast forest area in Xinjiang as an example of a practical application scenario. Specifically, Xinjiang, as an important agricultural production base in my country, has a distinctly large-scale and intensive agricultural and forestry industry. Large farms and vast forest areas have an urgent need for efficient and economical plant protection operations. However, the region's high daytime temperatures and strong winds create a significant contradiction with the demand for nighttime operations. Existing UAV perception systems face severe challenges: the vision system completely fails at night, and the performance of lidar drops sharply in field dust and orchard canopies, preventing safe operation during optimal working hours. Simultaneously, small obstacles such as power lines in protective forests pose significant flight safety hazards due to their difficulty in reliable identification. These perception bottlenecks severely restrict operational efficiency and safety, hindering the full realization of the technology's potential.

[0019] Specifically, such as Figure 2 As shown, a method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) includes the following steps: S100: Acquire the performance parameters of the UAV and the environmental parameters of the task area; the performance parameters of the UAV include the UAV's flight speed, effective jet width, remaining battery power, and estimated flight time dynamically calculated based on the current battery power and power consumption; the environmental parameters of the task area include the boundary and area of ​​the task area, information on static obstacles within the area, and real-time wind speed and direction. S100 includes the following steps: S110. Before the start of the UAV operation mission, a high-precision digital map of the operation area is acquired through manned aerial surveying or preliminary reconnaissance flights. This map includes at least the boundary coordinates, area, and static obstacle information of the area. This data is pre-stored on a server, forming a global environmental database; specifically: Using manned or unmanned aerial surveying platforms equipped with lidar or high-resolution optical cameras, remote sensing and mapping are carried out to generate digital elevation models or point cloud maps containing elevation information, from which the boundary coordinates, area, and spatial location and height of obstacles in the work area are extracted. The type of obstacle can be identified and marked by analyzing the image data collected by the optical camera or by combining it with on-site survey records. The system retrieves a pre-stored electronic map of the area and reads the marked boundaries, area, and obstacle information. Static obstacle information refers to information about obstacles that exist before the start of the operation and whose positions remain unchanged during a single operation. This information can be obtained through pre-mapped digital elevation models (DEMs), satellite or aerial imagery, and electronic maps containing geographic elements such as buildings, utility poles, towers, and fixed trees. It also includes the spatial location of static obstacles. S120. During the preparation and execution of the UAV mission, the UAV collects dynamic parameters in real time through onboard sensors and uploads them to the server; specifically: The UAV's flight speed is calculated and acquired in real time using a combination of a Global Navigation Satellite System (GNSS) receiver and an Inertial Measurement Unit (IMU) onboard the UAV; wind speed and direction are measured in real time using an onboard pitot tube or ultrasonic anemometer; the remaining battery power is read in real time through the battery management system connected to the UAV's flight control system, and the flight time is dynamically estimated based on the current power consumption and flight status; the real-time collected flight speed, wind speed, wind direction, remaining battery power, and estimated flight time are uploaded to the server via the onboard data link. In this embodiment, a static environment database is constructed through prior mapping, and dynamic flight and environmental data are collected in real time by airborne sensors and uploaded to the server; this provides a comprehensive and reliable data foundation for subsequent planning and control. S200: Based on the acquired performance and environmental parameters, it performs fusion analysis and decision planning to generate flight control commands and spraying control commands for the UAV. S200 includes the following steps: S210. Based on the obtained boundary coordinates and area of ​​the task area, the boundary coordinates of the task area are used as constraints to construct a two-dimensional raster map. Each raster cell represents a passable or impassable spatial location. The raster resolution is set according to the minimum turning radius and spray width of the UAV, for example, set to 1 / 2 of the spray width to ensure the accuracy of path planning. A path search algorithm (such as the bow-shaped scanning method) is used to calculate the direction of the long side of the minimum bounding rectangle of the working area, which is used as the main direction of the flight path to reduce the number of turns. Then, along the main direction, a series of parallel scan lines are generated with the effective spray width as the interval. The nearest neighbor strategy is then used to connect the endpoints of each scan line to generate a continuous flight path, which consists of a series of ordered waypoints. A complete waypoint includes spatial coordinates, waypoint type, action commands, turning mode, and desired speed. The spatial coordinates of a waypoint represent its horizontal position coordinates in a predefined plane coordinate system (such as the UTM coordinate system), as well as its altitude or relative altitude above the ground, used to determine the UAV's spatial position. The waypoint type is an identifier used to indicate the expected behavior of the UAV at that waypoint; for example, a normal waypoint is where the UAV flies smoothly at normal speed; an action waypoint is where the UAV performs a specific task, such as starting or stopping spraying or acquiring images; and a hovering waypoint is where the UAV hovers around the point to await instructions or perform special operations. The action commands for a waypoint are the specific operational instructions associated with the waypoint type; for example, for a dynamic waypoint... For waypoints, the action command can be to turn spraying on or off. The turning mode of a waypoint defines the turning strategy when the UAV flies from the previous waypoint to the current waypoint and then to the next waypoint. The specific turning modes are: straight flight, flying directly to the next waypoint without turning; coordinated turn, the UAV flies to the next waypoint in a smooth arc trajectory, which is the most commonly used mode; pause turn, the UAV decelerates at the current waypoint and then turns towards the next waypoint, which is suitable for sharp turns or scenarios requiring precise positioning; the expected speed of a waypoint is the target flight speed of the UAV when approaching and flying over this waypoint. Different flight speeds can be set for different segments of the path, such as using the standard speed within the operation boundary and automatically decelerating when turning or approaching obstacles. When performing path tracking, the flight controller processes this series of ordered waypoints in sequence. It calculates the deviation between the UAV's current position and the current target waypoint, and generates real-time control surface and throttle commands through control algorithms (such as PID or L1 navigation laws) based on the spatial position, flight speed, and turning strategy defined in the current waypoint. This guides the UAV to fly smoothly and accurately to the waypoint. After reaching the effective range of the waypoint, the controller checks its waypoint type and action commands and triggers the corresponding actions (such as turning spray on or off). Then, it sets the next waypoint as the new target, thus completing the flight mission for all waypoint targets. The pre-stored 3D obstacle model is projected onto a 2D grid map, and the obstacle is inflated, where the inflated radius R... 膨胀 The calculation formula is: R1 is the radius of the circumcircle of the UAV, R2 is a preset additional safety margin, such as 3 meters, and R3 is the standard deviation of the horizontal positioning accuracy provided by the GNSS receiver. This value can be estimated based on the receiver's nominal circular error or through the real-time positioning quality factor. On the expanded obstacle map, the initially generated bow-shaped path is locally optimized using either the A* algorithm or Dijkstra's algorithm to ensure that the path maintains a minimum R-value with all static obstacles. 膨胀 Safe distance; In practice, when the task area is a plain field crop (such as cotton), the system generates long-distance, high-speed parallel flight paths to maximize efficiency; when the area is a tall tree forest belt (such as a shelterbelt), the system will focus on obstacle information and reserve safe avoidance space for known obstacles such as utility poles during the planning stage. In this embodiment, a grid map is constructed based on the boundary and obstacle information of the work area, a bow-shaped scanning method is used to generate a global path, and flight and operation logic is defined through waypoint attributes. The system uses algorithms such as A* to perform static obstacle avoidance optimization on the path to ensure a safe distance from obstacles. It adaptively generates efficient or safe flight routes for different scenarios such as plains, fields, and forest belts. It achieves fully autonomous operation path planning, effectively improves flight safety in complex terrain, and significantly improves the efficiency and reliability of ultra-low altitude spraying operations through precise waypoint control and obstacle avoidance. S220. After generating the global path for the UAV, handle dynamic issues encountered during flight; specifically: The system dynamically adjusts the weights of data fusion based on sensor characteristics to ensure the reliability of nighttime perception. The UAV's onboard main controller continuously monitors ambient light intensity sensor data. When the light intensity value is lower than the preset light threshold, the system determines that it has entered nighttime or low-light conditions and automatically sends a command to activate the nighttime perception mode. The preset light threshold is set to 10 Lux, which is based on the performance critical point of industrial-grade vision sensors. When the ambient light is lower than this critical point, the recognition reliability of the vision sensor decreases significantly. The system switches perception modes at this threshold to ensure that LiDAR and thermal imaging are activated as the main sensing sources before visual information fails. S221. In nighttime perception mode, each sensor works collaboratively and processes independently. The forward-facing LiDAR on the nose of the UAV becomes the main sensor for environmental geometry perception. Its point cloud data is first filtered by voxel grid downsampling to reduce data volume and noise. Then, the random sampling consensus algorithm or Euclidean clustering algorithm is used to segment independent obstacle point cloud clusters from the downsampled point cloud. For each point cloud cluster, its minimum axial bounding box is calculated to estimate its spatial position and size. To assess whether the obstacle is a real, stable target (rather than transient noise), its stability coefficient C is calculated. 稳定 In consecutive time frames t-1 and t, the 3D set centroid of point cloud cluster C1 is calculated, and the Euclidean distance between the centroids between the two consecutive frames is calculated. The Euclidean distance of the centroids is normalized to the interval [0,1] to obtain the stability coefficient C. 稳定 Based on the combined characteristics of point cloud quantity and stability coefficient, the confidence level of the lidar is calculated using a weighted summation model. The calculation formula is as follows: Where α1 and α2 are weighting coefficients, α1+α2=1, for example α1=0.7, α2=0.3, N1 is the number of points contained in the point cloud cluster, and Nmax is the maximum number of points in the point cloud cluster; S222. The thermal imaging video stream is fed into a lightweight convolutional neural network (e.g., a model with fewer than 5 million parameters, modified from the YOLO or MobileNet architecture) for real-time target detection. This network is trained on a large dataset of infrared images containing targets such as power lines and birds at night, and can output detection boxes with category labels (e.g., power lines, animals) and thermal imaging confidence scores. The thermal imaging confidence score is obtained as follows: Thermal Imaging Confidence Score C 热成像 The native confidence score output by the target detection network for each identified target obstacle j is directly used. This score is generated by the Sigmoid or Softmax function at the end of the network and is a floating-point number between 0 and 1, directly representing the model's confidence level in its recognition result. For example, when the network identifies a wire in an image, it outputs a bounding box and a score, such as 0.92. This 0.92 is directly used as the confidence score. 热成像 The value; S223. Millimeter-wave radar data from the UAV serves as redundancy verification. Since millimeter-wave radar provides target traces unaffected by lighting and weather conditions, including range, azimuth, radial velocity, and signal-to-noise ratio (SNR) information, these target traces are primarily used for moving target detection and as safety redundancy in case of failure of the first two types of sensors. The target traces output by the millimeter-wave radar are clustered and tracked. Signal strength factors and motion consistency factors are extracted and fused to calculate the millimeter-wave radar confidence level C. 毫米波 The signal strength factor is positively correlated with the signal-to-noise ratio (SNR) of the target point trace, and is obtained by normalizing the SNR. The motion consistency factor is used to evaluate the rationality of the target's radial velocity. If the target's radial velocity is within a preset velocity range, a higher confidence level is assigned; otherwise, a lower confidence level is assigned. The millimeter-wave radar confidence level C... 毫米波 The signal strength factor and motion consistency factor are obtained by fusing them using a weighted model; In forest protection operations, when the system simultaneously identifies power lines ahead through lidar point clouds and thermal imaging (high confidence), if there is sufficient time, it generates a yaw detour instruction; if it is an emergency, it immediately generates an emergency climb instruction to use the safe airspace above the power lines to overcome the obstacle. This targeted strategy effectively solves the industry problem of thin power lines. S224. Calculate a comprehensive confidence level C for each detected potential obstacle j. 综合 The calculation formula is as follows: This formula, through dynamic weight allocation, enables the system to automatically rely on the sensor with the best performance in different environments. Its core lies in adaptively adjusting the fusion weights of the three confidence levels of lidar, thermal imaging and millimeter-wave radar according to the ambient lighting conditions, so as to achieve complementary perception advantages. The technical effect of this design is to significantly improve the overall reliability and type recognition accuracy of obstacle recognition at night and in complex environments, and to provide a stable and reliable unified quantitative basis for subsequent decision-making. Where C 激光 The confidence level of the lidar is calculated based on the point density and stability of the point cloud cluster; C 热成像 The thermal imaging confidence score is directly taken from the output score of the target detection network; C 毫米波 The confidence level of the millimeter-wave radar is calculated based on signal strength and radial velocity characteristics. The sum of the weighting coefficients α3, α4, and α5 of each sensor is always 1, and their specific values ​​are preset according to the sensor characteristics. The setting rules are as follows: in daytime mode, the lidar is the main sensor with α3≈0.6 and thermal imaging is the auxiliary sensor with α4≈0.3; for millimeter-wave radar, α5≈0.1. The nighttime enhancement rule is that when the system enters nighttime mode, the preset nighttime weight configuration is automatically called, and the thermal imaging weight α4 is increased to about 0.5, making it and the lidar the main sensors. At the same time, α3 is reduced to 0.4 to make full use of the unique recognition advantage of thermal imaging in the absence of light. The weighting coefficients are configured with preset fixed values, and their setting is based on the performance characteristics and inherent limitations of different sensors in day and night environments. In night mode, the weight (α4) of the thermal imaging confidence level is set to a level no lower than that of the lidar weight (e.g., 0.5), based on the following technical facts: Nighttime environments cause visible light visual sensors to fail, requiring the system to rely on non-visible light sensors to provide semantic information about obstacle types. Thermal imaging sensors are currently the only airborne perception method capable of effectively classifying targets such as power lines and birds in the absence of light. By increasing the weight of thermal imaging to the dominant level, the algorithm can compensate for the shortcomings of lidar in nighttime semantic recognition capabilities, thereby solving the problem of nighttime identification of small, specific obstacles during ultra-low-altitude flight. This fixed weight configuration is a specific technical measure taken to address the problem of missing semantic perception at night. For each obstacle obtained after fusion, its estimated collision time T1 is calculated based on its relative distance and relative speed with the drone. The estimated collision time T1 is then compared with the overall confidence level C. 综合 Implement tiered obstacle avoidance decision-making; If the overall confidence level is C 综合 Greater than the high confidence threshold C 高 And the expected collision time T1 is less than the emergency time threshold T 紧急If so, an emergency climb command is generated; the emergency climb command performs an emergency climb based on the target height, which is equal to the current height plus the emergency obstacle crossing height (e.g., 15 meters). This is the last resort to avoid obstacles such as power lines that cannot be bypassed. If the overall confidence level is C 综合 Greater than the high confidence threshold C 高 Furthermore, the expected collision time T1 is less than the normal time threshold T. 正常 But greater than the emergency time threshold T 紧急 If the obstacle is not found, a yaw detour command is generated. Based on the obstacle's position, the yaw detour command uses geometric calculations to generate a temporary waypoint, guiding the UAV to pass beyond its dynamic obstacle avoidance safety distance. The dynamic obstacle avoidance safety distance is the sum of the UAV's circumcircle radius and a dedicated safety margin for dynamic obstacle avoidance (e.g., 2 meters), where the dedicated safety margin is less than the static safety distance R in the static planning. 膨胀 Because it does not need to compensate for the maximum error of the global positioning system, it only considers the short-term uncertainties of real-time control and perception; If the overall confidence level is C 综合 Greater than the low confidence threshold C 低 Less than the high confidence threshold C 高 If the condition is met, a cautious passage command is generated; the cautious passage command controls the drone to decelerate and slightly increase its flight altitude to increase the safety margin, while preparing to execute a higher level of evasive maneuver. If the overall confidence level is C 综合 Less than the low confidence threshold C 低 If so, a pause command for autonomous decision-making is generated, and obstacle information (such as a highlighted thermal image) is pushed to the ground station for operator decision-making; It should be noted that the threshold values ​​set in the hierarchical obstacle avoidance decision-making process are determined comprehensively based on the dynamic constraints of the UAV's ultra-low-altitude flight, sensor performance, and the safety redundancy requirements for real-time decision-making. The specific value selection rules are as follows: Confidence threshold (C) 高 With C 低 The rules for determining the value of ) are as follows: High confidence threshold C 高 (e.g., 0.8): This threshold is set at a high level to ensure that the obstacle perception results that trigger automatic avoidance actions are highly reliable, so as to minimize unnecessary maneuvers caused by sensor false alarms and ensure operational continuity; a value of 0.8 means that the system needs at least two sensors or one sensor to confirm the existence of the target with a very high degree of certainty. Low confidence threshold C 低(e.g., 0.5): This threshold serves as the trigger for risk warning and manual intervention. When the confidence level is below this value, it indicates that the sensor data is ambiguous or contradictory, and the automatic system is unable to make a reliable decision. Requesting manual intervention at this time returns the decision-making responsibility to the most reliable decision-maker (operator) and is the last line of defense in the system design. Time threshold (T) 紧急 With T 正常 The rules for determining the value of ) are as follows: Emergency time threshold T 紧急 (e.g., 2 seconds): This threshold defines the lower limit of the time during which a drone can perform complex evasive maneuvers (such as detours). When the expected collision time is less than 2 seconds, the system does not have enough space to perform a smooth yaw detour and must activate the ultimate avoidance strategy with the least computational load and the fastest response—emergency vertical climb. Normal time threshold T 正常 (e.g., 5 seconds): This threshold defines the time window for initiating standard obstacle avoidance planning. Five seconds provides the drone with sufficient distance and computational margin to plan and execute a smooth, energy-efficient, and safe detour path, while avoiding premature obstacle avoidance that could affect operational efficiency. By setting the above rules, the obstacle avoidance system of the present invention can achieve the optimal balance between operational efficiency, flight safety and system reliability; S225 receives the reference heading from the global path, real-time wind disturbance information from sensor data, and avoidance instructions from the obstacle avoidance decision layer. Through arbitration and fusion, it generates the final comprehensive flight instructions issued to the flight controller. Reference track tracking command generation: The flight controller first calculates the lateral tracking deviation between the UAV's current position and the preset global path; then, it generates a reference heading command using a feedforward-feedback composite control strategy: based on the lateral tracking deviation and its changing trend, it calculates the heading correction amount, the control logic being that the larger the deviation, the larger the correction amount; when the deviation tends to decrease, the correction amount also decreases, thereby eliminating steady-state deviation and suppressing oscillations; at the same time, based on the real-time measured wind speed and direction, it estimates the steady-state impact of wind on the UAV's heading and calculates a feedforward compensation angle, the purpose of which is to offset wind disturbances in advance, rather than waiting for the deviation to occur before responding; Final command arbitration and output: A multi-level command priority arbitration logic is set to cope with different operating conditions. Under normal circumstances, when the system does not trigger any alarms or obstacle avoidance strategies, the above-mentioned reference track tracking command is output as the final heading angle command. When an obstacle threat is detected, when an obstacle avoidance command (such as a yaw detour command) is received from S224, the arbitration logic will be activated. At this time, the obstacle avoidance command has the highest priority and will override and replace the normal reference track tracking command as the final command output to ensure flight safety. In the event of communication or system failure, when the communication link with the server is detected to be interrupted for more than the set disconnection safety threshold, or when the signal of a key sensor (such as GNSS) fails, the system will switch to a degraded safety mode. In this mode, the track command from the server will be ignored, and flight commands will be generated according to the last valid command or a preset strategy (e.g., deceleration, climb, or return along the original route). It should be noted that if the communication link interruption exceeds the set disconnection safety threshold, it is necessary to balance avoiding malfunctions caused by brief communication jitter and ensuring that the safety policy can be activated in time when communication fails. Based on the above principles, the safety threshold is usually set in the range of 2 to 5 seconds. Preferably, its value can be set to 3 seconds, which can effectively filter communication jitter and reserve sufficient time for the activation of the safety mode. In this embodiment, the system automatically switches to nighttime perception mode based on ambient light intensity, integrates data from lidar, thermal imaging, and millimeter-wave radar, and dynamically allocates fusion weights based on sensor characteristics. At night, the thermal imaging weight is increased to 0.5 to compensate for insufficient semantic recognition by lidar. By calculating the overall obstacle confidence level and estimated collision time, a tiered obstacle avoidance decision is executed, including commands such as emergency climb and yaw detour, ensuring reliable avoidance of small obstacles such as power lines under low-light conditions. This method improves the reliability and accuracy of obstacle recognition for UAVs at night and in complex environments. Through the complementary advantages of multiple sensors and a tiered decision-making mechanism, it effectively solves the problem of perceiving and avoiding small obstacles during ultra-low-altitude flight, enhancing operational safety and system robustness. The S230 flight controller receives the target heading angle command and, in conjunction with the airborne wind speed and direction information, generates control surfaces and power commands through the internal attitude control loop to achieve trajectory tracking. This enables the system to maintain the UAV's stable flight along the predetermined trajectory under crosswind or complex wind field conditions. Based on the target application rate A (liters / hectare), real-time flight speed Vg (meters / second), and effective spray width W (meters), the target flow rate Q is calculated in real time according to the formula Q=(A×Vg×W) / K=, where K is a unit conversion constant with a value of 666.7. The spraying command ensures that the actual flow rate is consistent with the target flow rate by adjusting the pump speed. The ray method is used to determine the positional relationship between the real-time position of the UAV and the boundary of the work area. When the UAV enters the boundary from outside the area, a command to open the nozzle is generated; when the UAV leaves the area and moves to the outside, a command to close the nozzle is generated. To eliminate positioning jitter, a boundary buffer zone with a width of 0.5 meters can be set. Whether spraying high-volume foliar fertilizers or low-drift herbicides, the precise flow control provided by this step is fundamental to achieving uniform application and eliminating waste. S240 analyzes the remaining battery power and estimated range in real time, and assesses the feasibility of the task based on the remaining working distance; when the battery power is lower than the battery safety threshold or the range is insufficient to complete the task, it generates an automatic return command; the battery safety threshold is usually 1.2 times the battery power required for return. Simultaneously, real-time wind speed is continuously monitored. If the wind speed exceeds the operating wind speed threshold for the drone platform or spraying operation (e.g., 8 m / s), a deceleration command is generated. This command will control the drone to decelerate to its safe minimum flight speed—the critical airspeed at which the drone is designed to maintain stable flight and control without entering a stall state. This speed aims to reduce pesticide drift and attempt to maintain the operation while ensuring flight safety. If the wind speed exceeds the extreme safe wind speed threshold for the drone platform or spraying operation (e.g., 15 m / s), an emergency stop operation and return command is generated to ensure platform safety. It should be noted that the operating wind speed threshold is set with the primary goal of ensuring effective deposition of the sprayed liquid and preventing excessive drift of the liquid. It is determined based on agricultural aviation plant protection industry standards and field trial data. When the wind speed exceeds this threshold, the drift distance of the liquid droplets will increase significantly, resulting in insufficient liquid coverage per unit area of ​​the target crop and environmental pollution. 8 m / s (approximately level 5 wind) is a critical wind speed recognized by multiple agricultural aviation operation guidelines that can balance operational efficiency and spraying quality. The ultimate safe wind speed threshold is set to ensure the structural safety and flight control stability of the drone platform. It is determined based on the maximum wind resistance level specified in the drone product's design specifications. For example, many large agricultural drones are designed to withstand winds of level 6-7. The upper limit of level 7 wind speed is approximately 13.9-17.1 m / s. To ensure a safety margin, a value between or slightly lower than this is taken, such as 15 m / s (approximately level 7 wind), as the extreme condition for triggering emergency abort. When the wind speed exceeds this threshold, the drone is at risk of losing control or structural damage. In this embodiment, precise pesticide application is achieved by calculating the target flow rate in real time. Combined with the ray method to control the opening and closing of the nozzles, the boundary is ensured to be accurate. The system monitors the battery level and wind speed in real time. When the battery level is insufficient or the wind speed exceeds the limit, the return command is automatically triggered to ensure the safety of the operation. Precise spraying of pesticides with zero waste is achieved, and pesticide drift and pollution are effectively prevented. Through intelligent safety monitoring, the safety of the drone operation and the stability of the platform are ensured in complex environments, thereby improving the quality and reliability of agricultural aviation operations. S300 sends flight control commands and spraying control commands to the UAV's flight controller and spraying system to perform precision flight and spraying operations; S300 includes the following sub-steps: S310 Flight Command Execution: The flight controller receives roll angle and heading angle commands from S200 and drives the actuators through the following closed-loop control process to achieve precise and stable ultra-low-altitude flight; specifically: S311, Target Heading and Attitude Tracking: The flight controller receives roll angle and heading angle commands from the server; combined with airborne wind speed, wind direction, and flight path information, it generates control surface deflection or motor speed adjustment through an internal high-frequency attitude control loop (such as proportional-integral-derivative PID or adaptive controller) to drive the ailerons, rudder, and elevator to achieve attitude tracking; and dynamically allocates control quantities for the ailerons, rudder, and elevator according to the direction and intensity of wind disturbance to improve response speed and steady-state accuracy. S312, Feedforward Wind Disturbance Compensation: To actively counteract the effects of wind disturbance and improve track tracking accuracy, the controller performs feedforward compensation; the calculated feedforward compensation angle is combined with the heading angle command output by the feedback control to form the final target heading angle; the control weights of the ailerons and rudder are dynamically allocated according to the intensity and direction of the wind disturbance; for example, under continuous crosswinds, the ailerons are used first to generate resistance to the roll moment, supplemented by the rudder for coordination. S313, Adaptive Track Curvature Adjustment: To achieve smooth flight and precise tracking during turns, the flight controller performs the following steps: adaptively adjusting flight attitude and speed based on track geometry; extracting a local ternary sequence from the current flight path, containing the current target waypoint, the previous waypoint, and the next waypoint; estimating the instantaneous track curvature Ki at the current waypoint by calculating the curvature of the circle formed by the three points based on the local ternary waypoint sequence; the greater the instantaneous track curvature, the more the flight controller automatically increases the roll angle to ensure smooth turns and reduces the flight speed to reduce centrifugal deviation; maintaining standard operating speed on straight sections with low instantaneous track curvature. A parallel feedforward roll angle command is generated based on the instantaneous track curvature Ki. This feedforward roll angle command is proportional to the instantaneous track curvature, enabling the UAV to pre-establish a flight bank that matches the curvature of the curve before entering the curve, thereby achieving coordinated and smooth turning maneuvers. At the same time, the desired flight speed command is dynamically adjusted according to the instantaneous track curvature Ki. The control logic is: the greater the instantaneous track curvature, the lower the flight speed, to ensure flight stability and controllability during the turning process and effectively reduce track tracking deviation caused by centrifugal force. S314, Dynamic Obstacle Avoidance: The flight controller receives the location of obstacles in real time and calculates the closest distance between each obstacle and the flight path. When the distance is less than a preset safety threshold, the flight controller adjusts the heading angle to deviate from the obstacle and adjusts the roll angle in combination with the curvature of the flight path. If necessary, it decelerates or climbs / descends to avoid obstacles. Through the coordinated operation of the above steps, the flight controller can comprehensively consider multiple tasks such as trajectory tracking, wind disturbance compensation, smooth turning and emergency obstacle avoidance in complex environments at ultra-low altitudes. Ultimately, it drives the UAV's power system and control surfaces through electronic speed controllers and servos to achieve safe and precise flight. In this embodiment, stable flight in complex environments is achieved through coordinated control of attitude tracking, feedforward wind disturbance compensation, adaptive trajectory curvature adjustment, and dynamic obstacle avoidance. The system dynamically adjusts the roll angle and speed according to the trajectory curvature to ensure smooth turns. Combined with real-time wind disturbance compensation, the trajectory tracking accuracy is effectively improved. This method enhances the flight stability and trajectory tracking accuracy of UAVs in ultra-low-altitude operations, enabling them to actively adapt to wind disturbances and complex trajectories, achieve smooth and coordinated turning maneuvers, and ensure emergency obstacle avoidance capabilities, thereby comprehensively improving the safety, accuracy, and operational efficiency of ultra-low-altitude flight. S320, Spraying Command Execution: The spray controller receives the target flow rate and nozzle switching commands from the decision-making layer, and drives the spraying system to achieve precise and adaptive variable spraying through the following closed-loop control logic; specifically: S321. Precise Closed-Loop Control of Flow Rate: The spray controller receives the target flow rate Q and obtains the actual flow rate value fed back by the flow sensor. It calculates the difference between the target flow rate and the actual flow rate at the current moment, i.e., the flow error. Based on the flow error, a control decision is made: if the actual flow rate is lower than the target flow rate, the controller output is adjusted to increase the PWM duty cycle (or analog voltage) of the pesticide pump; if the actual flow rate is higher than the target flow rate, the controller output is adjusted to decrease the PWM duty cycle (or analog voltage) of the pesticide pump. To ensure the smoothness and accuracy of flow control, the controller sets an error threshold and determines the adjustment range based on the absolute value of the flow error. Fast response adjustment: When the absolute value of the flow error exceeds the error threshold, the controller performs a fast response adjustment with a large adjustment range to quickly reduce the flow error. At this time, the controller output adjustment is adjusted based on the magnitude of the flow error and the fast response proportional coefficient β1, and the adjustment amount is... ; Fine-tuning adjustment: When the absolute value of the flow error is less than the error threshold, the controller performs fine-tuning with a small adjustment range to ensure system accuracy and avoid overshoot. At this time, the controller outputs a small adjustment range based on the fine-tuning proportional coefficient β2. ; The values ​​of the fast response proportional coefficient β1 and the fine-tuning proportional coefficient β2 are determined through system identification and controller tuning. The core principle is that β1 must be sufficient to produce a significant adjustment effect within the system's inertial delay to quickly eliminate large errors; β2 must be small enough to suppress overshoot and stabilize near the target value. Based on extensive practical experience with agricultural spraying systems, the values ​​of these two coefficients typically follow these rules: Value range: The typical value range of β1 is 0.5~2.0; the typical value range of β2 is 0.1~0.5. Within this range, the relationship β1>β2 always holds. Tuning Relationship: β1 and β2 are not set independently; their ratio (β1 / β2) defines the trade-off between fast response and fine stability of the controller. This ratio is usually set in the range of 3 to 10. For a given spraying system, a larger β1 can be initially set based on its response speed, and then a smaller β2 can be determined according to the aforementioned ratio based on the desired steady-state accuracy. For example, in a preferred embodiment, for a spraying system with a response time of 0.8 seconds and a target flow rate of 2 liters / minute, using β1=1.2 and β2=0.3 (ratio of 4), combined with a 5% error threshold, can achieve fast flow tracking without overshoot. It should also be noted that the error threshold can be set according to the accuracy requirements of the application rate per acre; for example, 5% of the target flow rate can be used. If the flow rate error is greater than 5%, the system will perform a rapid response adjustment; if the flow rate error is less than or equal to 5%, a fine-tuning will be performed. The adjustment is always performed in the positive direction of the error increase or decrease, and will not exacerbate the error in the opposite direction, so as to ensure the stability and accuracy of the system. S322. Precise start / stop control of spray boundary: Based on the real-time positioning information of the UAV, the spray controller continuously determines the relative position of the UAV and the boundary of the work area through a geometric relationship judgment algorithm (such as the ray method). When it is determined that the UAV has crossed the work boundary, a corresponding nozzle opening or closing command is generated. To eliminate the influence of positioning signal jitter, the system sets a boundary buffer zone, and the final opening and closing action is only executed after the UAV has completely entered or left this buffer zone. S323, Adaptive jet amplitude control based on flight path curvature: To compensate for the impact of the difference in track length between the inner and outer sides during turns on the uniformity of spraying, the system implements adaptive spray width control. The flight controller sends the instantaneous track curvature information calculated in real time to the spray controller. Based on the instantaneous track curvature, the spray controller calculates different flow correction coefficients for the nozzle groups located on the inner and outer sides of the curve, respectively. The specific implementation method is as follows: The value of the flow correction coefficient is proportional to the magnitude of the instantaneous trajectory curvature. The greater the instantaneous trajectory curvature, the greater the flow output of the outer nozzle and the smaller the flow output of the inner nozzle. The controller adjusts the flow of each nozzle or nozzle group independently to ensure that the liquid can achieve uniform ground coverage even on curved trajectories. At the same time, the system applies preset upper and lower limits to the flow correction coefficient to prevent it from exceeding a reasonable adjustment range; Based on the instantaneous trajectory curvature, the spray controller determines the flow correction coefficient for the nozzle groups on both the inner and outer sides of the curve, and its design rules are as follows: Dynamic compensation rule: The coefficient assigned to the outer nozzle must be greater than 1, and the coefficient assigned to the inner nozzle must be less than 1. The difference between the two increases monotonically with the increase of the absolute value of curvature, so as to dynamically compensate for the difference in ground coverage rate caused by different turning radii. System stability rules: The flow correction coefficient is limited to a range determined based on the physical flow range of the nozzle and the safety margin. The upper limit is set to prevent the flow of the outer nozzle from exceeding its maximum effective spray capacity; the lower limit is set to prevent the flow of the inner nozzle from falling below the minimum flow required to maintain stable atomization. The controller independently adjusts the flow rate of each nozzle according to the coefficients determined by the above rules to ensure that the liquid is evenly covered on the ground along the curved flight path.

[0020] In this embodiment, rapid response and fine-tuning of the target flow rate are achieved through closed-loop flow control. Combined with precise boundary start / stop and adaptive spray width adjustment based on the trajectory curvature, the accuracy and uniformity of spraying operations are ensured. The system dynamically allocates the flow rate of the inner and outer nozzles according to the curvature of the turning curve, effectively compensating for the coverage differences caused by the curvature of the trajectory. This method achieves high-precision control of pesticide flow rate and precise boundary management, significantly improving operational efficiency and eliminating pesticide waste. Through adaptive spray width control, the uniformity of spraying under complex trajectories such as curves is guaranteed, effectively improving the quality of agricultural plant protection and resource utilization.

[0021] The core concept of this invention lies in constructing an integrated perception-decision-control autonomous drone operation system. Addressing the needs for ultra-low-altitude flight and precise spraying in nighttime and complex environments, the system solves the problem of reliable obstacle identification and classification under visual failure conditions through multi-sensor (LiDAR, thermal imaging, millimeter-wave radar) fusion perception and a dynamic weight allocation mechanism. Furthermore, based on a graded obstacle avoidance decision-making system using comprehensive confidence and estimated collision time, it achieves a smooth and safe response from emergency climb to manual intervention. Finally, through feedforward wind disturbance compensation, adaptive flight control with flight path curvature, and precise spraying control combining dynamic adjustment of inner and outer flow rates with curvature, the system collaboratively ensures flight stability and spray uniformity under complex flight paths and wind disturbance conditions, ultimately comprehensively improving the drone's operational capabilities, safety, and economy in demanding agricultural scenarios.

[0022] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV), characterized in that, Includes the following steps; S100: Acquire the performance parameters of the UAV and the environmental parameters of the task area. S200 performs fusion analysis and decision planning based on the performance parameters of the UAV and the environmental parameters of the operational task area to generate flight control commands and spraying control commands; specific steps include: A path search algorithm is used to generate a global flight path based on the task area information, and static obstacle avoidance optimization is performed on the path based on static obstacle information. During flight, when the ambient light intensity is lower than the preset light threshold, the night perception mode is automatically activated. In the night perception mode, the perception results of lidar, thermal imaging and millimeter-wave radar are fused to calculate the comprehensive confidence level of the obstacle and the estimated collision time, and graded obstacle avoidance decisions are made accordingly. The target flow rate is calculated based on the target usage per acre, real-time flight speed and effective spray width, and the ray method is used to determine the positional relationship between the real-time position of the UAV and the boundary of the work area, thereby generating nozzle start and stop commands; The trajectory tracking command of the generated global flight path and the obstacle avoidance command generated in real time are combined to generate the final flight control command through a command priority arbitration mechanism. The command priority arbitration mechanism means that when an obstacle avoidance command exists, the obstacle avoidance command is used to override the trajectory tracking command and is used as the final flight control command; when no obstacle avoidance command exists, the trajectory tracking command is used as the final flight control command. The S300 executes flight control commands and spray control commands to drive the UAV to complete precise flight and adaptive variable spraying operations. The flight control commands include generating feedforward roll angle commands based on the instantaneous trajectory curvature and adjusting the flight speed according to the instantaneous trajectory curvature. The spray control commands include allocating different flow rates to the nozzle groups on the inside and outside of the curve based on the instantaneous trajectory curvature.

2. The method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The specific performance parameters and environmental parameters in S100 are as follows: The performance parameters include flight speed, remaining battery power, and estimated flight time. The environmental parameters include information on the boundaries, area, and static obstacles of the work area obtained through preliminary surveys, as well as wind speed and direction obtained in real time through airborne sensors.

3. The method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The static obstacle avoidance optimization specifically refers to: The pre-stored 3D obstacle model is projected onto a 2D grid map and then expanded. The expansion radius is the sum of the radius of the UAV's circumcircle, the preset safety margin, and the standard deviation of the GNSS horizontal positioning accuracy. The preset safety margin ranges from 2 meters to 5 meters and is used to compensate for unmodeled dynamic uncertainties and control errors.

4. The method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The preset illumination threshold is as follows: The illumination threshold is set to 10 Lux, based on the fact that this value is determined by the performance critical point of industrial-grade vision sensors. When the ambient light is below this critical point, the recognition reliability of the vision sensor decreases significantly. The system switches the perception mode at this threshold to ensure that LiDAR and thermal imaging are enabled as the main sensing sources before visual information fails.

5. The method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The comprehensive confidence level is specifically as follows: The confidence scores of LiDAR, thermal imaging, and millimeter-wave radar are combined to obtain the overall confidence score of the obstacle; and in nighttime perception mode, the weight of thermal imaging confidence score is set to be no less than that of LiDAR confidence score.

6. The method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The graded obstacle avoidance decision-making includes: If the overall confidence level is greater than the high confidence level threshold and the estimated collision time is less than the emergency time threshold, an emergency climb command is generated. If the overall confidence level is greater than the high confidence level threshold and the estimated collision time is less than the normal time threshold but greater than the emergency time threshold, a yaw detour instruction is generated. If the overall confidence level is greater than the low confidence threshold but less than the high confidence threshold, a cautious passage instruction is generated. If the overall confidence level is less than the low confidence level threshold, a pause autonomous decision-making instruction is generated, and obstacle information is pushed to the ground station for operator decision-making.

7. The method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The trajectory tracking command is calculated using a feedforward-feedback composite control method; wherein, the feedforward control quantity is used to actively compensate for the heading deviation caused by real-time wind speed and wind direction.

8. The method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle according to claim 1, characterized in that, The method for calculating the instantaneous track curvature is as follows: Extract three consecutive waypoints from the current flight path of the UAV to form a local sequence, calculate the curvature of the osculating circle determined by the three points, and use the curvature as the instantaneous track curvature.

9. A method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The different flow rates are allocated to the nozzle groups on the inside and outside of the curve, specifically: Based on the instantaneous trajectory curvature, a flow correction factor greater than 1 is assigned to the outer nozzle, while a flow correction factor less than 1 is assigned to the inner nozzle; and the difference between the flow correction factors of the inner and outer nozzles is monotonically increasing with the absolute value of the instantaneous trajectory curvature.

10. A method for controlling ultra-low-altitude spraying flight of a large unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The method also includes security monitoring and tiered decision-making steps: Safety monitoring: When the duration of the communication link interruption with the server exceeds the disconnection safety threshold, control the drone to execute the predetermined return-to-home procedure; Tiered decision-making: Continuously monitor real-time wind speed and execute tiered decisions: When the wind speed exceeds the operational wind speed threshold but does not exceed the ultimate safe wind speed threshold, a deceleration command is generated to attempt to maintain the operation while ensuring flight safety. When the wind speed exceeds the aforementioned limit safe wind speed threshold, an instruction is generated to suspend operations and immediately return to base.

Images