Unmanned aerial vehicle precision plant protection operation system based on variable spray technology

Abstract

This invention discloses a precision plant protection system for drones based on variable spraying technology, belonging to the field of plant protection drones. The system includes a drone flight platform, a variable spraying execution unit, a multi-source sensing unit, an airborne edge computing unit, and a ground control unit. The multi-source sensing unit collects real-time data on crop growth, pests and diseases, and field topography and transmits it to the airborne edge computing unit. The unit integrates plant protection prescription maps to dynamically generate instantaneous spray flow commands, driving the variable spraying execution unit to complete differentiated pesticide application for different areas. The ground control unit realizes operation planning, data feedback, and remote equipment control. This invention relies on real-time edge computing power to replace the traditional offline uniform spraying mode. It can precisely adjust droplet size and spray rate according to the differentiated crop conditions in the field, significantly reducing pesticide application, mitigating non-point source pollution of water and soil, improving pest and disease control effectiveness and farmland operation efficiency, and is suitable for various plant protection scenarios such as large-scale field and facility agriculture.

Description

Technical Field

[0001] This invention belongs to the field of agricultural drone technology, and particularly relates to a drone precision agricultural operation system based on variable spraying technology. Background Technology

[0002] With the rapid development of agricultural modernization and low-altitude plant protection technology, plant protection drones have become an important piece of equipment for pest and disease control in my country due to their advantages such as high operating efficiency, wide applicability, and the ability for sprayers to stay away from pesticide spray. They are widely used in plant protection operations for field crops such as wheat, rice, and corn, as well as cash crops such as citrus and apples.

[0003] Most mainstream agricultural drones currently use a constant spraying mode, meaning they maintain a fixed spray flow and flight parameters throughout the operation. This makes it impossible to dynamically adjust the spray based on differences in crop growth and uneven distribution of pests and diseases. This can easily lead to insufficient pesticide application and poor control in areas with severe pests and diseases, while excessive pesticide application occurs in areas with weaker crop growth or no pests or diseases. This not only wastes pesticides and increases the cost of agricultural operations, but also has negative impacts such as excessive pesticide residues and pollution of the farmland environment, which does not meet the development requirements of green agriculture and precision agriculture.

[0004] To address the shortcomings of constant-volume spraying, some existing technologies have proposed variable-volume spraying solutions for plant protection. By pre-setting the application rate parameters for the work area, drones can execute different spraying flows in different plots, which improves pesticide utilization to some extent. However, these solutions are mostly preset variables that rely on pre-mapped plot data and manually set parameters. They cannot respond to real-time changes in the field during the operation and are difficult to achieve precise zoning application down to the centimeter level. The granularity of variable adjustment is relatively coarse, and the accuracy is limited.

[0005] Other solutions use airborne cameras combined with image recognition technology to identify pests and diseases, and adjust the spray flow rate according to the identification results. However, most existing identification solutions use monocular visible light images, and the identification accuracy is greatly affected by lighting, crop shading, and shooting angle. The accuracy of identifying early-stage pests and diseases and hidden pests and diseases is insufficient. Moreover, most solutions deploy the identification calculation on the ground and rely on wireless transmission of image data. Due to the limitations of field signals, transmission delays and data loss are easy to occur, resulting in a lag in the response of variable spraying. It is difficult to match the high-speed flight operation rhythm of drones, and the variable control effect in actual operation is greatly reduced.

[0006] Furthermore, existing variable spray systems mostly only adjust the overall spray flow rate, failing to achieve independent control of individual nozzles. They also lack adaptive adjustment functions for droplet size, and cannot adjust droplet parameters according to crop canopy structure and the location of pests and diseases. This makes it difficult to effectively deposit pests and diseases inside the canopy, resulting in limited improvement in overall plant protection effectiveness. At the same time, the lack of a quantitative evaluation mechanism for operational effectiveness makes it impossible to intuitively verify the actual effectiveness of plant protection operations, which is not conducive to the optimization and iteration of operational processes. Summary of the Invention

[0007] The purpose of this invention is to provide a drone-based precision plant protection system that achieves on-demand pesticide application and precision plant protection through multi-source sensing fusion, edge-end intelligent decision-making, and zoned variable spray control, thereby improving pesticide utilization and operational effectiveness, and reducing pesticide usage and environmental pollution.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] The drone-based precision plant protection system includes a drone flight platform, a variable spray execution unit, a multi-source sensing unit, an airborne edge computing unit, and a ground control unit.

[0010] The multi-source sensing unit is mounted on the bottom of the UAV flight platform and includes a hyperspectral imaging module, a binocular vision module, and a weather sensing module.

[0011] The airborne edge computing unit is fixed inside the fuselage of the UAV flight platform and is electrically connected to the multi-source sensing unit, the variable spray execution unit, and the flight control module of the UAV flight platform. It includes a crop growth recognition module, a pest and disease classification module, and a spray decision model module.

[0012] The variable spray actuator is mounted under the fuselage of the UAV flight platform and includes a medicine tank, an electronically controlled nozzle array, a flow regulating valve group, and a liquid level sensor.

[0013] The ground control unit communicates bidirectionally with the airborne edge computing unit and the UAV flight control module via a wireless communication link for operation path planning, real-time monitoring of operation status, and operation data backtracking.

[0014] During the operation, the multi-source sensing unit collects real-time crop spectral images, visual images, and field meteorological data of the operation area and transmits them to the airborne edge computing unit. The crop growth recognition module extracts multi-dimensional vegetation indices such as normalized vegetation index, photochemical reflectance index, and anthocyanin reflectance index based on hyperspectral data. Combined with crop canopy height and canopy coverage data obtained from binocular vision, it completes a quantitative assessment of crop growth. The pest and disease classification module completes the identification of pest and disease types and the classification of damage levels through a lightweight convolutional neural network. The spray decision model module combines crop growth level, pest and disease damage level, and real-time meteorological data to generate spray flow rate, droplet size, and spray pressure parameters for the corresponding operation area in real time. It outputs control commands to the variable spray execution unit. The flow regulating valve group and the electronically controlled nozzle array dynamically adjust the output parameters of each nozzle according to the commands to realize on-demand variable spraying by region and gradient.

[0015] Furthermore, the hyperspectral imaging module has a band coverage range of 400nm to 1000nm, a spectral resolution of 2.5nm, a spatial resolution of 1024×768, an acquisition frame rate of 15fps, an acquisition angle perpendicular to the direction of the drone's flight, and a ground resolution of 1cm to 3cm per pixel corresponding to the flight altitude. It can accurately capture the differences in spectral characteristics of crop leaves and identify early pests, diseases, and crop stress conditions.

[0016] Furthermore, the binocular vision module has a baseline length of 120mm, a monocular resolution of 1920×1080, and a frame rate of 30fps. It simultaneously performs three functions: crop canopy structure measurement, field obstacle detection, and flight obstacle avoidance. The output depth data can assist in calculating crop canopy height and biomass, improving the accuracy of growth assessment. At the same time, it can detect obstacles on the flight path in real time, ensuring the safety of operational flights.

[0017] Furthermore, the meteorological sensing module includes a wind speed sensor, a wind direction sensor, a temperature sensor, and a humidity sensor. The data sampling frequency is 1Hz. The collected meteorological data is input into the droplet drift prediction model, which can adjust the spray pressure and droplet size parameters according to the real-time wind speed and direction. In windy conditions, it can increase the droplet size, reduce the spray height, reduce pesticide drift loss, and improve pesticide deposition efficiency.

[0018] Furthermore, the electronically controlled nozzle array includes 4 to 8 independent electronically controlled nozzles, arranged at equal intervals along the transverse side of the UAV fuselage. Each nozzle corresponds to an independent micro flow regulating valve. The flow rate regulation range of a single nozzle is 0.5L / min to 3L / min, the droplet size regulation range is 50μm to 300μm, and the response delay of parameter adjustment is less than 200ms. It can achieve centimeter-level width zoned variable spraying to match the application needs of different areas.

[0019] Furthermore, the spray decision model module adopts an improved lightweight YOLOv8 target detection network combined with a random forest regression algorithm. First, the target detection network identifies the location of pest and disease areas and the types of pests. Then, the random forest regression model outputs the optimal spray flow rate, droplet size, and spray pressure parameters for the corresponding area. After structured pruning and INT8 quantization, the model is deployed on the airborne edge computing unit. The inference time for a single frame image is less than 50ms, which can meet the real-time decision-making needs of UAVs flying at high speed.

[0020] Furthermore, the airborne edge computing unit adopts a high-performance edge computing chip based on the ARM architecture, paired with an NPU unit with 2 TOPS computing power and a local storage capacity of no less than 64GB. It supports local caching of all sensing data, control data and status data during the operation, and then batch synchronization to the ground control unit after the operation is completed, thereby reducing the amount of wireless data transmission during the operation and reducing the dependence on field communication signals.

[0021] Furthermore, the ground control unit includes an operation planning module, a real-time monitoring module, and an effect evaluation module.

[0022] The operation planning module supports importing remote sensing images of the plot, automatically identifying operation boundaries and obstacles, generating the optimal operation path and breakpoint resume planning, and the real-time monitoring module synchronously displays the drone flight status, spraying operation status, pest and disease identification results and remaining pesticide information. The effect evaluation module can compare hyperspectral data before and after the operation, quantitatively calculate the pest and disease control effect and crop growth changes, and generate a plant protection operation evaluation report.

[0023] The beneficial effects of the present invention after adopting the above structure are as follows:

[0024] (1) A multi-source perception scheme combining hyperspectral imaging and binocular vision is adopted, which integrates crop spectral features and canopy structure features. It can achieve accurate identification of early pests and diseases and quantitative assessment of crop growth. Compared with the traditional monocular vision identification scheme, the accuracy of pest and disease identification is increased by more than 15%, and the error of growth assessment is reduced by more than 10%, providing a precise data basis for variable spraying decision-making.

[0025] (2) The airborne edge computing unit is used to realize the local real-time calculation of spraying decision, without the need to transmit image data to the ground for processing. The variable control response delay is reduced from the second level to the hundreds of milliseconds level, which can match the operating flight speed of the UAV from 3m / s to 6m / s, realize the real-time dynamic variable adjustment on the flight path, avoid the problem of misaligned application caused by transmission delay, and improve the accuracy of variable spraying.

[0026] (3) The variable spraying execution unit with independent control of multiple nozzles can realize independent flow rate adjustment and droplet size adjustment in the horizontal zone. Compared with the traditional overall flow rate adjustment scheme, the variable adjustment particle size is finer, which can achieve precise application of pesticides for strip-shaped distribution of pests and diseases. At the same time, the droplet size can be adjusted according to the height of the crop canopy to improve the deposition rate of pesticides in the crop canopy.

[0027] (4) Combining real-time meteorological data to optimize spraying parameters and dynamically adjusting spraying strategies through droplet drift prediction models can reduce pesticide drift loss under wind conditions and improve pesticide utilization. Compared with constant spraying schemes, the average amount of pesticide used per mu can be reduced by 20% to 30%, effectively reducing pesticide waste and farmland ecological environment pollution.

[0028] (5) The ground control unit has the functions of operation planning, real-time monitoring and effect evaluation. It can realize the full-link digital management of plant protection operations, and quantitatively evaluate the plant protection effect by comparing spectral data before and after the operation, providing data support for operation parameter optimization and field management, and improving the management efficiency and standardization level of plant protection operations. Attached Figure Description

[0029] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0030] Figure 1 This is a diagram illustrating the overall system architecture and data flow of the present invention.

[0031] Figure 2 This is a timeline diagram of the entire process of a single plant protection operation according to the present invention;

[0032] Figure 3 This is the internal logic diagram of the spray decision model of the present invention;

[0033] Figure 4 This is a comparison diagram of system configurations for three embodiments of the present invention. Detailed Implementation

[0034] The technical solution of the present invention will be further described in detail below with reference to specific embodiments, but the scope of protection of the present invention is not limited to the following embodiments.

[0035] Example 1

[0036] This embodiment is a precision plant protection system based on variable spraying technology for wheat aphid control in field applications. The overall system architecture includes a hexacopter drone flight platform, a variable spraying execution unit, a multi-source sensing unit, an airborne edge computing unit, and a ground control unit.

[0037] In this embodiment, the drone flight platform adopts a six-rotor agricultural drone with a payload of 20kg, a maximum takeoff weight of 45kg, an effective payload of 20kg, a normal operating flight speed of 4m / s, an operating flight altitude of 2m to 3m above the crop canopy, and a single full-charge operation time of 15min to 20min. The flight control module supports autonomous flight of the flight path and real-time attitude adjustment, and has RTK centimeter-level positioning capability, which can ensure the accuracy and consistency of the flight path.

[0038] The multi-source sensing unit is mounted on the central shock-absorbing gimbal at the bottom of the UAV fuselage. It includes a hyperspectral imaging module, a binocular vision module, and a weather sensing module. The hyperspectral imaging module covers a wavelength range of 400nm to 1000nm, contains 128 spectral channels, has a spectral resolution of 2.5nm, a spatial resolution of 1024×768, and an acquisition frame rate of 15fps. With the lens shooting vertically downwards, the ground resolution at an operating height of 2.5m is 2cm per pixel, clearly capturing the spectral reflectance characteristics of wheat leaves and identifying spectral changes in leaves caused by early aphid infestations. The binocular vision module... The sensing module has a baseline length of 120mm, a monocular sensor with a 1080P color CMOS sensor, a frame rate of 30fps, and a horizontal field of view of 90°. It is used for wheat canopy height measurement, field obstacle detection, and obstacle avoidance during flight. The depth measurement error is less than 5cm within a 3m range. The meteorological sensor module is integrated into the windward position at the end of the UAV arm. It includes four types of sensors: wind speed, wind direction, temperature, and humidity. The sampling frequency is 1Hz, and the measurement accuracy is ±0.1m / s for wind speed, ±3° for wind direction, ±0.5℃ for temperature, and ±3%RH for humidity. It can collect micro-meteorological data of the operating area in real time.

[0039] The airborne edge computing unit is fixed to a shock-absorbing bracket inside the drone's fuselage. It uses an RK3588 edge computing chip, paired with a 6-TOPS NPU, 8GB of memory, and 128GB of local storage. Running an embedded Linux system, it integrates a crop growth recognition module, a pest and disease classification module, and a spraying decision model module. The crop growth recognition module extracts multi-dimensional vegetation indices such as NDVI, PRI, and CCI from hyperspectral data. Combined with canopy height data acquired through binocular vision, it classifies wheat growth into weak, medium, and strong levels using a pre-set growth assessment model, achieving an accuracy rate of over 93%. The pest and disease classification module... The pest classification module uses a lightweight YOLOv8n network optimized by pruning, which can identify three common pests and diseases of wheat: aphids, powdery mildew, and rust, and classify the severity of damage into three levels: mild, moderate, and severe. The model training dataset contains tens of thousands of field samples, and the average accuracy (mAP) can reach 89%. The spraying decision model module uses a random forest regression algorithm. The input parameters include crop growth level, pest and disease level, wind speed, temperature, and humidity. The output parameters are the spray flow rate, droplet size, and spray pressure for the corresponding area. The model was pre-trained using three years of field trial data, and the single-frame inference time is 35ms, which can meet the needs of real-time decision-making.

[0040] The variable spray actuator unit consists of a 16L food-grade corrosion-resistant tank, six independent electrically controlled centrifugal nozzles, six sets of micro flow control valves, a high-pressure diaphragm pump, and a capacitive level sensor. The six nozzles are evenly spaced laterally along the fuselage, with a total spray width of 4m. Each nozzle corresponds to an independent flow control valve, which can be individually controlled for its on / off state and flow rate. The flow rate adjustment range for a single nozzle is 0.5L / min to 2.5L / min. The droplet size can be continuously adjusted from 80μm to 250μm by adjusting the spray pressure. The parameter adjustment response delay is 150ms, enabling rapid dynamic adjustment. The level sensor uses a capacitive level gauge with a measurement accuracy of ±0.1L, which can monitor the remaining amount of medicine in real time and transmit the data to the ground control unit. When the amount of medicine is insufficient, it automatically triggers a return-to-base reminder.

[0041] The ground control unit is an industrial-grade tablet computer equipped with dedicated control software. It communicates bidirectionally with the drone via a 4G wireless link and includes a job planning module, a real-time monitoring module, and an effect evaluation module. Before operation, high-definition remote sensing images of the field can be imported to automatically identify field boundaries and obstacles such as power poles and trees, generate the optimal reciprocating operation path, and plan breakpoint resuming and refill locations. During operation, it can display real-time information on the drone's position, flight altitude, flight speed, remaining battery power, remaining pesticide dosage, spraying status, and pest and disease identification results, and supports manual emergency intervention and real-time parameter adjustment. After the operation is completed, hyperspectral data before and after the operation can be imported to automatically calculate the pest and disease control effect and crop growth changes, generate standardized plant protection operation reports, and support data export and archiving.

[0042] The specific operation process of this embodiment is as follows: Before the operation, the ground control unit completes field mapping and path planning, and sends the operation route and initial spraying parameters to the UAV flight control and airborne edge computing unit. After the tank is filled and the equipment self-check is completed, the operation is started. After the operation starts, the UAV flies autonomously according to the preset route. The multi-source sensing unit simultaneously collects hyperspectral images, binocular vision images and meteorological data of the field, and transmits them to the airborne edge computing unit in real time. The airborne edge computing unit first performs radiometric correction, geometric correction and continuous frame stitching on the hyperspectral data, extracts the vegetation index value of each pixel, and combines the depth data of the binocular vision to complete the calculation of crop canopy parameters and growth classification. At the same time, the target detection network identifies the location and type of pests and diseases in the image to complete the damage level classification. The spraying decision model combines the growth level, pest and disease level and real-time meteorological data to calculate the damage level. The optimal spraying parameters for different lateral areas within the current field of view are converted into flow and pressure control commands for the corresponding nozzles. These commands are then output to the variable spray execution unit via the CAN bus. The variable spray execution unit adjusts the valve opening and water pump pressure of the corresponding nozzles in real time according to the commands, achieving differentiated variable spraying for different lateral positions. For areas free of pests and diseases, the corresponding nozzles can be completely shut off to minimize pesticide waste. During the operation, all sensing data, control data, and flight status data are cached locally to avoid data loss due to signal interruption. After the operation is completed, the data is batch synchronized to the ground control unit. After the operation, the ground control unit compares the changes in NDVI index and the percentage of pest and disease areas before and after the operation, and combines this with field sampling survey data to quantitatively evaluate the control effect of this plant protection operation, generating a complete evaluation report that includes operation parameters, control effect, and pesticide dosage.

[0043] Example 2

[0044] This embodiment is a drone-based precision plant protection system for controlling red spider mites in citrus orchards. The overall architecture is the same as that of Embodiment 1, but the difference lies in the targeted optimization for the tall canopy and complex obstacle distribution characteristics of orchard scenarios.

[0045] This embodiment uses an octocopter agricultural drone with a payload of 30kg, a maximum takeoff weight of 60kg, an effective payload of 30kg, a normal operating flight speed of 2m / s, and an operating flight altitude of 1.5m to 2m above the citrus tree canopy. It supports terrain-following flight, tree-circling flight, and fixed-point hovering modes, and is equipped with a dual obstacle avoidance system of binocular obstacle avoidance and millimeter-wave radar, making it suitable for the complex flight environment of orchards.

[0046] In the multi-source sensing unit, the hyperspectral imaging module adds a lateral swing gimbal mechanism, which can swing left and right at a preset angle during flight to scan the sides and interior areas of the canopy, improving the probability of identifying pests and diseases inside the canopy and avoiding blind spots in the inner canopy. The binocular vision module adopts a long-range binocular module with a baseline length of 200mm, and the maximum detection distance can reach 50m. It can identify obstacles such as tree trunks, supports and ropes in advance, ensuring the safety of flight between rows in the orchard. The meteorological sensing module adds canopy temperature and humidity measuring points. By extending the sensing probe into the interior of the canopy, it can obtain more accurate micro-meteorological data of the working area and improve the accuracy of fog droplet drift prediction.

[0047] The pest and disease classification model of the airborne edge computing unit has been specifically trained for three common orchard pests and diseases: citrus red spider mites, canker, and anthracnose. It has expanded the sample data for different light angles and canopy positions, and the recognition accuracy can reach more than 92%. At the same time, it has added a canopy penetration assessment function, which can dynamically adjust the droplet size and spray pressure according to the canopy density and leaf layer, thereby improving the pesticide deposition rate inside the canopy. For leaf-back pests such as red spider mites, it can automatically reduce the droplet size and increase the spray pressure to improve the adhesion effect on the leaf back.

[0048] The variable spray execution unit uses eight centrifugal electronically controlled nozzles, arranged in two staggered rows. The upper row of nozzles faces the upper part of the tree canopy, and the lower row of nozzles faces the lower part and inner side of the tree canopy. The flow rate of a single nozzle can be adjusted from 1L / min to 4L / min, and the droplet size can be adjusted from 100μm to 300μm. It can achieve differentiated spraying for tree canopies of different heights and locations, improving the uniformity of overall pest control.

[0049] Example 3

[0050] This embodiment is a drone-based precision plant protection system for the prevention and control of downy mildew in greenhouse vegetables. The overall architecture is the same as that of Embodiment 1, but the difference lies in the miniaturization and high-precision optimization for the closed space and precision operation requirements of greenhouse agriculture.

[0051] This embodiment uses a small quadcopter plant protection drone with a payload of 5kg, a maximum takeoff weight of 12kg, an effective payload of 5kg, an operating flight speed of 1.5m / s, and an operating flight altitude of 0.8m to 1.2m above the vegetable canopy. It supports centimeter-level UWB precise positioning and fixed-point hovering operations. Its compact size allows it to move flexibly between rows in greenhouses.

[0052] The multi-source sensing unit uses a miniature hyperspectral imaging module, covering the band from 400nm to 900nm, with a spatial resolution of 640×480, a frame rate of 20fps, and a weight of only 150g, making it suitable for the payload limitations of small drones. The binocular vision module uses a close-range high-precision binocular module with a baseline length of 60mm and a depth measurement accuracy of ±1cm, which can accurately measure the height of the vegetable canopy and the distance between leaves, making it suitable for the low canopy characteristics of greenhouse vegetables. The meteorological sensing module is equipped with CO2 concentration and light intensity sensors, which can simultaneously monitor the environmental parameters inside the greenhouse and provide auxiliary data for crop growth management.

[0053] The airborne edge computing unit uses a lightweight edge chip with a computing power of 1 TOPS and lower overall power consumption. It optimizes the model for common diseases of greenhouse vegetables such as cucumber downy mildew and tomato gray mold, achieving an accuracy rate of over 90%. The model's single-frame inference time is less than 30ms, making it suitable for the real-time requirements of low-speed, refined operations.

[0054] The variable spraying unit uses four piezoelectric micro-nozzles, with a single nozzle flow rate adjustment range of 0.2L / min to 1L / min and a droplet size adjustment range of 30μm to 150μm. The spray droplets are more uniform and delicate, which is suitable for the precision application needs of greenhouse vegetables and can reduce the diffuse residue of pesticides in the greenhouse.

[0055] Controlled trials and efficacy verification

[0056] To verify the technical effect of the present invention, two comparative examples and the above three sets of examples were set up for field control experiments. The experimental site was the Huang-Huai-Hai Agricultural Experiment Station, the experimental time was the wheat grain-filling stage, the target of control was wheat ear aphid, the operating area of ​​each group was 1 mu, the wind speed of the operating environment was 2 m / s to 3 m / s, the ambient temperature was 25℃ to 28℃, and the relative humidity was 60% to 70%. All groups used the same type and concentration of pesticide formulation.

[0057] Comparative Example 1 uses a conventional constant-flow spraying agricultural drone with a fixed spray flow rate of 1.2 L / min and a fixed droplet size of 150 μm. The flight speed and flight altitude are the same as in Example 1.

[0058] Comparative Example 2 uses a commercially available simple variable spray agricultural drone equipped with a monocular visible light camera and basic image recognition algorithm. It can only adjust the overall spray flow rate and cannot achieve independent adjustment of individual nozzles. The recognition and calculation are completed on the ground. The flight parameters and operating environment are consistent with those of Example 1.

[0059] During the experiment, 20 sampling points were evenly set up in each experimental field. The water-sensitive paper method was used to determine the droplet deposition density and deposition uniformity. The high performance liquid chromatography method was used to determine the actual amount of pesticide deposited on the leaves. Seven days after the pesticide was applied, the aphid population reduction rate at each sampling point was investigated using the five-point sampling method. At the same time, the actual amount of pesticide used per mu was calculated. Finally, the pesticide utilization rate was calculated by the ratio of deposition amount to pesticide application amount. The experimental results are as follows.

[0060] In Example 1, the average pesticide usage per acre was 120 mL, the coefficient of variation for droplet deposition uniformity was 12%, the average pesticide deposition on leaves was 1.8 μg / cm², the aphid population reduction rate was 92%, and the pesticide utilization rate was 48%.

[0061] In the equivalent field verification test corresponding to the orchard scenario in Example 2, the average pesticide usage per mu was 135 mL, the coefficient of variation of droplet deposition uniformity was 14%, the average pesticide deposition on leaves was 2.0 μg / cm², the red spider mite population reduction rate was 90%, and the pesticide utilization rate was 45%.

[0062] In the greenhouse experiment corresponding to the facility vegetable scenario in Example 3, the average pesticide usage per mu was 95 mL, the coefficient of variation of droplet deposition uniformity was 10%, the average pesticide deposition on leaves was 2.1 μg / cm², the downy mildew disease index decreased by 88%, and the pesticide utilization rate was 52%.

[0063] The average pesticide application rate per mu in Comparative Example 1 was 180 mL, the coefficient of variation for droplet deposition uniformity was 25%, the average pesticide deposition on leaves was 1.2 μg / cm², the aphid population reduction rate was 78%, and the pesticide utilization rate was 28%.

[0064] In Comparative Example 2, the average pesticide application rate per mu was 150 mL, the coefficient of variation for droplet deposition uniformity was 20%, the average pesticide deposition on leaves was 1.4 μg / cm², the aphid population reduction rate was 83%, and the pesticide utilization rate was 35%.

[0065] The above experimental data shows that, compared with the traditional constant-volume spraying scheme, the precision plant protection operation system of the present invention reduces the amount of pesticide used per acre by 33%, increases the pesticide utilization rate by 71%, improves the pest and disease control effect by 18%, significantly improves the uniformity of droplet deposition, and has a faster variable control response speed and higher operation accuracy. It can effectively reduce pesticide waste and environmental pollution, and has significant technical advantages and practical application value.

[0066] Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents. In conclusion, if those skilled in the art, inspired by this description, design similar structural methods and embodiments without departing from the inventive spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A drone-based precision plant protection system based on variable spraying technology, characterized in that: It includes an unmanned aerial vehicle (UAV) flight platform, a variable spray execution unit, a multi-source sensing unit, an airborne edge computing unit, and a ground control unit; The multi-source sensing unit is mounted on the bottom of the UAV flight platform and includes a hyperspectral imaging module, a binocular vision module, and a weather sensing module. The airborne edge computing unit is fixed inside the fuselage of the UAV flight platform and is electrically connected to the multi-source sensing unit, the variable spray execution unit, and the flight control module of the UAV flight platform. The airborne edge computing unit includes a crop growth recognition module, a pest and disease classification module, and a spray decision model module. The variable spray execution unit is mounted under the fuselage of the UAV flight platform and includes a medicine tank, an electronically controlled nozzle array, a flow regulating valve group and a liquid level sensor. The ground control unit communicates bidirectionally with the airborne edge computing unit and the UAV flight control module through a wireless communication link for operation path planning, real-time monitoring of operation status and operation data backtracking. During the operation, the multi-source sensing unit collects crop spectral images, visual images, and field meteorological data in the operation area in real time and transmits them to the airborne edge computing unit. The crop growth recognition module extracts vegetation indices based on hyperspectral data and combines them with crop canopy structure data from binocular vision to complete crop growth assessment. The pest and disease classification module completes pest and disease type identification and damage level classification through a lightweight convolutional neural network. The spray decision model module combines crop growth, pest and disease level, and meteorological data to generate spray flow rate, droplet size, and spray pressure parameters for the corresponding area in real time and outputs control commands to the variable spray execution unit. The flow regulating valve group and the electronically controlled nozzle array dynamically adjust the output parameters of each nozzle according to the commands to realize zoned, on-demand variable spraying.

2. The UAV precision plant protection operation system based on variable spray technology according to claim 1, characterized in that: The hyperspectral imaging module has a band coverage range of 400nm to 1000nm, a spectral resolution of 2.5nm, a spatial resolution of 1024×768, an acquisition frame rate of 15fps, an acquisition angle perpendicular to the direction of the UAV's flight and downward, and a ground resolution of 1cm to 3cm per pixel corresponding to the acquisition height.

3. The UAV precision plant protection operation system based on variable spray technology according to claim 1, characterized in that: The binocular vision module has a baseline length of 120mm, a monocular resolution of 1920×1080, and a frame rate of 30fps. It is used for crop canopy height measurement, obstacle detection, and flight obstacle avoidance. The output depth data assists in crop growth assessment and adaptive adjustment of spray height.

4. The UAV precision plant protection operation system based on variable spray technology according to claim 1, characterized in that: The meteorological sensing module includes a wind speed sensor, a wind direction sensor, a temperature sensor, and a humidity sensor. The data sampling frequency is 1Hz. The meteorological data is used to correct the droplet drift prediction model, adjust the spray pressure and droplet size parameters, and reduce pesticide drift loss.

5. The UAV precision plant protection operation system based on variable spray technology according to claim 1, characterized in that: The electronically controlled nozzle array contains 4 to 8 independent electronically controlled nozzles, which are arranged at equal intervals along the transverse side of the UAV fuselage. Each nozzle corresponds to an independent flow regulating valve. The flow rate regulation range of a single nozzle is 0.5L / min to 3L / min, the droplet size regulation range is 50μm to 300μm, and the response delay is less than 200ms.

6. The UAV precision plant protection operation system based on variable spray technology according to claim 1, characterized in that: The spraying decision model module uses an improved lightweight YOLOv8 network combined with a random forest regression algorithm. First, it identifies the pest and disease areas and types through a target detection network, and then outputs the optimal spraying parameters for the corresponding areas through a regression model. After pruning and quantization, the model is deployed on an airborne edge computing unit, with a single-frame inference time of less than 50ms.

7. The UAV precision plant protection operation system based on variable spray technology according to claim 1, characterized in that: The airborne edge computing unit adopts an ARM-based edge computing chip, paired with an NPU unit with 2 TOPS computing power, and has a local storage capacity of no less than 64GB. It supports local caching of all sensing and control data during the operation and synchronizes it to the ground control unit after the operation is completed.

8. The UAV precision plant protection operation system based on variable spray technology according to claim 1, characterized in that: The ground control unit includes an operation planning module, a real-time monitoring module, and an effect evaluation module. The operation planning module supports autonomous path planning and breakpoint resume planning based on plot remote sensing images. The real-time monitoring module synchronously displays flight status, spraying status, identification results, and remaining pesticide dosage. The effect evaluation module completes a quantitative evaluation of plant protection effect based on spectral data before and after the operation.

Images