激光探测多信息融合的果园变量喷雾模型与实时控制研究

The research objective is to develop an intelligent controller for precision sprayers that can continuously match canopy characteristics and field environment to deliver agrichemicals accurately to nursery and fruit crops, and thus reduce pesticide use,save fuel, reduce the chance of worker exposure and save labor costs in nursery production. A high speed wide angle laser sensor will be used to detect tree objects on left and right lines of the sprayer simultaneously during operation. The raw data from laser sensor and sprayer velocity sensor after real-time field road condition correction will be converted to three-dimension images. Image processing algorithms will be applied for calculating tree canopy characteristics such as tree height, canopy volume and foliage density. A spray model will be developed to optimize variable flow rate based on tree canopy characteristics, field environment parameters and expert knowledge. Embedded computer programs using VC++ language software will be developed to convert laser signals into canopy location, size, shape and density,and ultimately control spray nozzles with variable rates. An electronic flow rate control system accommodating with microprocessors and Pulse Width Modulation(PWM) controlled solenoid valves will be designed to manipulate the output of spray nozzles independently to match tree structures. The communication interfaces based on bus technology will be used to fulfill communications between the embedded computer and the control system. The universal laser-guided spray controller for precision sprayer applications can be used to protect floriculture, nursery, vegetable, field crops against damages from diseases and pests, while safeguarding environmental quality, food and worker safety. Reducing pesticide consumption through targeted application based on precision instrument, information fusion and online decision-making model methods as proposed in the project provides good ideas in the development of intelligent strategies and methodologies for pesticide spray application technologies.

集对靶性、适量性和均匀性等优点，研究能自适应场地环境持续匹配植株有无和特征变化的智能变量喷雾模型系统，旨在提高农药利用率，降低喷施误靶率，减少环境污染。基于激光传感器实时探测的喷雾机双侧姿态各异苗木冠层点云数据，综合雷达探测喷雾机实时行走速度和数字陀螺仪检测的场地路况数据等多传感器信息，在线重构并校正目标三维树形轮廓和提取多靶标区域冠层的体积密度等特征量；融合植株目标特征、喷雾机行进速度、喷雾压力、路面平整状况、场地风速和病虫害专家知识等多信息构建多靶标独立喷嘴目标区域变量喷雾模型；研究系统输出动态延迟补偿预测机制解决变速喷雾速度、场地风速和系统输出延时等对喷雾目标对靶的影响；实时控制脉宽调制变量单元调节多通道喷嘴实现变流量输出，验证变量喷雾系统的喷雾沉降量、均匀性、对靶性和节约量等。提炼以精密探测、信息融合、在线决策、控制实现为主线的精密施药一般性科学问题，为智能喷雾研究奠定理论基础。

变量喷雾是实现农业精密施药的重要方法，能提高喷施均匀性、降低误靶率、节约农药量和减少环境污染。之前学者研究主要集中于对靶喷雾或仿形喷雾，本课题研究能匹配植株精细特征变化的精密喷雾，采用激光雷达等多传感器目标检测与特征量提取方法，融合激光和IMU姿态角实时矫正模拟复杂地形进行喷雾目标的激光检测与三维重构，基于K-means和近邻回归算法修复植株深度检测图像，提出双信息特征源目标植株图像拼接和植株点云超限补偿信息融合方法，建立了自适应智能变量喷雾动态决策模型，设计了系统动态延时补偿机制来解决对精确对靶喷雾的影响，控制多路喷嘴独立变流量指令的实时更新，由多通道变流量喷雾控制器实现变量喷雾输出，实现变速喷雾行走下的自动精确对靶变量喷雾。设计实施了智能变量喷雾控制系统的目标喷雾沉降量、均匀性和节约量对比等实验。实验结果表明本课题研究的变量喷雾模型系统能自适应场地条件实现变量喷雾，相比于恒量喷雾减少误靶率80%以上，变量模式下的喷雾面积沉降量占比相对于恒量喷雾模式30%~55%，变量模式喷雾量为恒量模式的12.1%到43.3%，减少农药用量50%以上。