基于处方图的变量喷药系统设计与试验(大豆-玉米带状复合种植):实时处方图解析、时空喷嘴匹配与延时补偿(原标题:Design and Experiment of a Prescription-Map-Based Variable-Rate Spraying System for Soybean-Maize Strip Intercropping)

1. Introduction

Maize and soybean are globally important food, feed, and oilseed crops, playing vital roles in ensuring food security, supporting livestock production, and facilitating agricultural processing and utilization. Over time, maize–soybean cropping systems have been continuously optimized [1,2,3,4]. Soybean–maize strip intercropping optimizes the spatial arrangement of the two crops, enabling stable maize production while expanding soybean cultivation, thereby improving land-use efficiency and promoting coordinated grain and oilseed production [5,6,7]. However, soybean and maize differ markedly in inter-row canopy structure, crop-protection requirements, and herbicide tolerance, making conventional large-scale uniform spraying prone to crop injury and missed applications [8,9]. Prescription-map-based variable-rate spraying enables real-time adjustment of application rates according to field weed distribution, crop growth status, and machinery operating parameters, providing an important approach to advancing precision crop protection [10,11,12].

With the development of precision agriculture, variable-rate pesticide application has become an important research direction in agricultural engineering. In recent years, it has attracted extensive attention from researchers worldwide and has gradually emerged as a research hotspot [13,14,15,16].

Existing studies show that variable-rate pesticide application has achieved clear progress in prescription map generation, remote-sensing image recognition, spray actuator control, and improvement of application effectiveness [17]. Hamouz et al. constructed field-zoned application maps, and their experiments showed that site-specific weed control reduced herbicide use by 15.6–100% compared with conventional whole-field uniform herbicide application [18]. With advances in remote sensing and aerial operation technologies, prescription map generation has gradually expanded from traditional manual sampling and grid surveys to airborne platforms, unmanned aerial vehicles, and multispectral image recognition. Yang et al. integrated an airborne multispectral imaging system to generate prescription maps and used the imaging system to evaluate post-application effects [19]. Campos et al. used an unmanned aerial vehicle equipped with a multispectral camera to acquire crop canopy information for prescription-map generation and conducted field evaluations of prescription-map-based variable-rate spraying [20]. Guan et al. reviewed advances in variable-rate spraying from two perspectives: information acquisition based on prescription maps or real-time sensors, and three variable-rate control strategies involving pressure regulation, direct injection, and pulse-width modulation [21]. Yin et al. developed an indoor prescription-map-based variable-rate herbicide spraying system, in which weed images were acquired to generate prescription maps, and precise targeted spraying was achieved by selectively controlling individual nozzles [22]. Abioye et al. developed a deep-learning-guided variable-rate spraying robot that integrates visual perception and deep learning for target plant identification, enabling targeted and variable-rate spraying and providing a new approach to integrating prescription maps, machine vision, and robotic actuation [23].

Despite substantial advances in variable-rate spraying, offline prescription-map-based application still involves time-consuming preparatory steps, including data acquisition, processing, and prescription map generation [10]. During dynamic field operations, the microcontroller must continuously parse positioning data, perform spatial matching, and retrieve prescription values, placing stringent real-time demands on online prescription map interpretation under limited computational and memory resources [24]. Moreover, inherent response delays from control-command output to solenoid-valve actuation and spray formation can translate into spatial application offsets as the sprayer moves, thereby compromising flow-rate control accuracy [25,26].

Accordingly, this study developed a variable-rate spraying system for soybean–maize strip intercropping that innovatively integrates fast embedded prescription-matrix addressing, nozzle-level spatial position prediction, and dynamic delay compensation to achieve precise spatiotemporal matching between prescription information and actual spray locations. The main contributions are as follows: A prescription-map-based strip-specific variable-rate spraying system was proposed for soybean–maize strip intercropping, providing technical support for differentiated and potentially more efficient pesticide application; a rapid prescription-map interpretation method was implemented on the embedded controller to enable real-time matching between machine position and target application rate, thereby improving prescription retrieval efficiency; and a prescription map–nozzle matching delay-compensation method was developed and validated through outdoor droplet-deposition, delay-compensation, and strip-specific variable-rate spraying tests, demonstrating the feasibility of the proposed system.

2. Materials and Methods

2.1. System Architecture

2.1.1. Overall System Design

The overall design of the strip-specific variable-rate spraying system for soybean–maize strip intercropping is shown in Figure 1. Because maize and soybean differ markedly in pesticide tolerance and application requirements, the system adopts two independent pesticide delivery pipelines to enable separate strip-specific application for maize and soybean. To prevent spray liquid from drifting between different crop zones during application, anti-drift nozzles were selected, and cover structures were installed outside the soybean nozzles to reduce the risk of droplet drift.

Figure 1. Overall structure of the variable-rate spraying system. 1. water tank; 2. return valve; 3. check valve; 4. maize pesticide tank; 5. peristaltic pump; 6. Venturi injector; 7. pesticide mixer; 8. pressure sensor; 9. flow sensor; 10. high-speed solenoid valve; 11. anti-drift nozzle; 12. shroud; 13. microcontroller; 14. GNSS receiver; 15. water pump; 16. RTK-BDS; 17. soybean pesticide tank.

Before field operation, strip-specific variable-rate prescription maps are generated on a GIS platform using crop growth information and converted into a format suitable for embedded processing. During operation, the main controller receives real-time RTK-BDS positioning data, calculates the spatial position of each nozzle based on the boom geometry, and retrieves the corresponding target application rate through prescription-map addressing. Flow and pressure sensors in the two independent spray circuits provide real-time feedback, and the controller adjusts the PWM duty cycles of the high-speed solenoid valves according to the deviation between the target and measured flow rates, thereby enabling independent variable-rate application to the maize and soybean strips. In the hydraulic system, the spray solution sequentially passes through the pressure sensor, flow sensor, and high-speed solenoid valve before being delivered to the drift-reducing nozzles, with every three nozzles forming an independent control unit.

2.1.2. Control System Design

STM32 series microcontrollers are characterized by high performance, low power consumption, and low cost, and are therefore commonly used as core controllers in control systems [27,28,29]. Based on the prescription map construction and interpretation method described above, together with the variable-rate pesticide application control algorithm, a soybean–maize variable-rate pesticide application control system was designed with an STM32F103ZET6 microcontroller (STMicroelectronics, Geneva, Switzerland) as the core. The system mainly consists of a power supply module, signal acquisition module, main controller module, actuator module, and communication module.

The overall architecture of the variable-rate spraying control system is shown in Figure 2. During operation, the RTK-BDS positioning terminal continuously outputs the sprayer’s longitude, latitude, heading, and travel speed, which are transmitted to the main controller via a serial interface. Using the preloaded prescription-map data and real-time positioning information, the controller performs field-coordinate transformation, grid identification, and prescription-map addressing to determine the target application rate for the current or predicted spray position. Meanwhile, flow-rate and pressure measurements from the two independent spray circuits are fed back to the main controller. An Improved Beetle Antennae Search (IBAS)-optimized Fuzzy-PID controller calculates the control output from the deviation between the target and measured flow rates. The resulting PWM duty cycles are transmitted to the actuator module via the communication bus, enabling independent variable-rate application to the soybean and maize strips.

Figure 2. Overall structure of the variable-rate pesticide application control system.

The system software employs a hybrid task-scheduling architecture that combines a main-loop routine with interrupt-driven processing. Time-critical signals, such as flow-meter pulses, are processed using timer input-capture interrupts. The main loop sequentially performs RTK-BDS positioning updates, nozzle-position calculation and prescription-map addressing, sensor-data processing for the two spray circuits, flow-control computation, and command output. To reduce the computational burden of multichannel actuation on the MCU, the main controller only calculates and updates the PWM duty cycle for each channel, while the actual PWM signals are generated independently by an external multichannel PWM module to drive the high-speed solenoid valves. This architecture reduces the real-time processing load associated with continuous PWM generation on the main controller.

In terms of hardware, an NZ510G high-precision intelligent positioning terminal (Yanshi Zhikong, Beijing, China) was selected as the GNSS receiver positioning device for the variable-rate spraying system. Spray output was regulated by pulse-width modulation (PWM), whereby the duty cycle of the solenoid valve at each nozzle was adjusted by controlling its opening and closing durations. Consequently, the response speed, corrosion resistance, and operational stability of the solenoid valves directly affect spray-control performance. Considering the field operating environment and the properties of pesticide solutions, normally closed 2WA030-08B solenoid valves (AirTAC (China) Co., Ltd., Ningbo, China) were selected.

2.2. Prescription Map Construction

2.2.1. Overview of ArcMap Software

ArcMap is a geographic information system software developed by Esri and is one of the major desktop applications of the ArcGIS 10.8.1 platform. ArcMap supports various spatial data formats, including vector data, raster data, and spatial database data, and has been widely applied in geographic information-related fields such as environmental management, urban planning, agricultural monitoring, disaster risk assessment, and emergency management [30,31,32].

The proposed system requires a GIS platform with high spatial accuracy and compatibility with the data structure used by the embedded controller. ArcMap provides coordinate registration, projection transformation, regular grid generation, and attribute-data processing, enabling spatial alignment between prescription maps and RTK positioning data. Regular grids generated using the Fishnet tool facilitate spatial matching and prescription-value retrieval, while attribute data can be exported and converted into binary files for embedded processing. Thus, ArcMap satisfies the system requirements for spatial accuracy, grid-based processing, and data export.

2.2.2. Acquisition of Field Position Information

Field position information is a prerequisite for prescription map construction and variable-rate pesticide application control. In this study, a set of simulated prescription map data conforming to the spatial distribution pattern of soybean–maize strips was generated for testing. The dataset included the spatial position information of the operating area and the corresponding prescription values, as shown in Table 1. These coordinates were used for prescription-map construction.

Table 1. Longitude, latitude, and prescription values used for constructing the simulated prescription map.

To facilitate subsequent prescription map interpretation and invocation by the pesticide application control system, the above data were imported into ArcMap for organization and preprocessing, including spatial extent inspection, attribute field verification, and coordinate reference system unification. Through data preprocessing, the prescription map data and BeiDou positioning information were aligned to a unified spatial reference, and the correspondence between the spatial positions of the operating area and the prescription values was established.

After processing, the basic spatial dataset required for subsequent prescription unit division, prescription map construction, and addressing-based matching was obtained, providing data support for real-time prescription retrieval by the variable-rate spraying system.

2.2.3. Prescription Unit Division and Application-Rate Assignment

To enable the spatial representation of prescription map data and subsequent invocation by the variable-rate spraying system, the original longitude, latitude, and prescription value data of the operating area must be converted into prescription units with spatial position attributes, and the correspondence between prescription units and application rates must be established. The original data used in this study were stored in tabular form and mainly included fields such as longitude, latitude, and the corresponding prescription value.

To spatially represent the prescription data for subsequent variable-rate spraying, the raw dataset containing longitude, latitude, and prescription values was imported into ArcMap 10.8.1. As shown in Figure 3, a prescription-point layer was generated using the Add XY Data function, and the coordinate reference system was standardized to ensure spatial consistency with RTK-BDS positioning data. The resulting event layer was then exported as a vector layer, and the correspondence between spatial locations and prescription values was verified, providing the basis for subsequent prescription-cell delineation, prescription-map generation, and spatial addressing.

Figure 3. Generation of the operating-area layer.

As shown in Figure 4, a regular rectangular grid was generated over the operating area using the Fishnet tool in ArcMap. For the 4:2 soybean–maize strip intercropping configuration, the transverse grid resolution, Δx, was set equal to the actual crop-strip width to accurately represent strip-specific application boundaries. Prescription values were then assigned to the corresponding grid cells, establishing a one-to-one correspondence between spatial location and target application rate. Each grid cell therefore contained a unique spatial location and application attribute, representing the target application requirement for the corresponding area.

Figure 4. Generation of the regular rectangular grid.

2.2.4. Construction of the Simulated Prescription Map

After prescription-cell delineation and assignment of application rates, a prescription map of the operating area was generated to visualize the spatial distribution of application requirements and provide input data for subsequent prescription-map interpretation and variable-rate spraying control. In ArcMap, the operating area was discretized into regular grid cells based on the assigned prescription-cell layer, and the cells were classified according to their prescription values and displayed using different colors to distinguish application-rate levels, as shown in Figure 5. The completed prescription map was then visually inspected to verify the prescription-cell delineation and spatial distribution of prescription values, thereby confirming the completeness and consistency of the prescription-map data.

Figure 5. Prescription map for soybean–maize strip intercropping.

On the basis of the visualized prescription map, the structured attribute data of the prescription map must be further exported to meet the requirements of the subsequent prescription map interpretation system for grid identification and rapid addressing. In this study, row number r and column number c fields were established for each grid cell in ArcMap and saved together with the prescription value field, forming an attribute table containing both grid index information and prescription value information. The relevant attribute data were then exported as a .csv file and further converted into a .bin file readable by the microcontroller. Through this process, the visualized prescription map generated in ArcMap was converted in