基于机器视觉的拖拉机地头转向控制研究  

Research on Automatic Steering of Tractors in Headlands Based on Machine Vision

作　　者：李霄伟 赵军[1] 车刚[1] 王熙[1] Li Xiaowei;Zhao Jun;Che Gang;Wang Xi(College of Engineering,Heilongjiang Bayi Agricultural University,Daqing 163711,China)

机构地区：[1]黑龙江八一农垦大学工程学院,黑龙江大庆163711

出　　处：《农机化研究》2024年第7期227-233,共7页Journal of Agricultural Mechanization Research

基　　金：“十四五”国家重点研发课题(2021YFD2100901);黑龙江八一农垦大学“三纵”计划支持项目(ZRCPY201805)。

摘　　要：自动驾驶拖拉机可按照规划好的路线进行自动化作业,但离不开预先的路径信息采集,而基于深度学习的图像识别技术在自动驾驶汽车上的应用越来越普遍,可以用图像识别代替路径规划,实现拖拉机对地头的识别。为此,以基于深度学习的图像识别技术为核心,采集秋季收获后田间图像进行训练,设计了地头识别软件,可通过OPC协议与PLC通讯,控制电磁换向阀。利用Carsim仿真拖拉机转向获得拖拉机前轮转角参数,用于系统控制可行性检验,结果表明:训练的识别模块对田间图像的识别准确度为99.11%,且软件对电磁换向阀有实时精准的控制。Carsim中,拖拉机转向仿真可输出各种自动驾驶参数,可为今后研究提供参考。Automated driving the tractor can be carried out in accordance with the planned route automation homework,but this way of automatic operation is dependent on the advance path information collection,and image recognition tech-nology based on deep learning application in automatic driving a car is becoming more common,image recognition can be used instead of path planning,the implementation of the tractor out of recognition.In this paper,the image recognition technology based on deep learning is taken as the core,the field image after autumn harvest is collected for training,and the field recognition software is designed.The software communicates with PLC through OPC protocol and controls the e-lectromagnetic directional valve.Carsim is used to simulate the tractor steering to obtain the front wheel angle parameters for system testing.The results show that the recognition accuracy of the trained recognition module to the field image can reach 99.11%,and the software has real-time and accurate control of the electromagnetic directional valve,and the tractor steering simulation in Carsim can output various automatic driving parameters,which lays a foundation for future research.

关 键 词：机器视觉 深度学习 地头识别 神经网络 液压转向 

分 类 号：S219.032.3[农业科学—农业机械化工程]