基于旋转目标检测的青贮机精准抛送系统设计与试验  

Design and experiment of the precise throwing system for silage machines based on rotating target detection

作　　者：尤泳[1] 李芳旭 纪中良 孙玉玲 马朋勃 卢琦琦 王海翼 王德成[1] YOU Yong;LI Fangxu;JI Zhongliang;SUN Yuling;MA Pengbo;LU Qiqi;WANG Haiyi;WANG Decheng(College of Engineering,China Agricultural University,Beijing 100083,China;Shandong Wuzheng Takakita Stockbreeding Machinery Co.,Ltd.,Rizhao 276800,China)

机构地区：[1]中国农业大学工学院,北京100083 [2]山东五征高北农牧机械有限公司,日照276800

出　　处：《农业工程学报》2024年第21期36-45,共10页Transactions of the Chinese Society of Agricultural Engineering

基　　金：山东省重大科技创新工程项目(2022CXGC020704);国家现代农业产业技术体系(CARS-34)。

摘　　要：现阶段青贮收获机主要采用驾驶员或其他操作人员手动控制抛送筒转动的方式完成物料装车,存在操作要求高,劳动强度大,影响收获效率且易造成田间损失等问题。针对上述问题,该研究根据机器视觉原理,设计青贮收获机精准抛送填装系统,通过构建R-YOLOv5旋转目标检测算法,实现对旋转车厢、车内物料、抛送物料流的识别及落料位置判断;根据反馈的期望落料点与实际落料点信息,并通过Arduino控制器实现青贮收获机抛送筒的运动控制,将青贮物料精准抛送至跟车车厢,实现物料的高效填装。试验结果表明:当抛送筒在-60°~60°范围内旋转且角速度低于15°/s时,所构建的R-YOLOv5目标检测算法对料车车厢的识别平均精度高于90%;实际距离与图像像素比值为0.533 cm/px时,青贮收获机抛送筒的理论运动轨迹与实际运动轨迹平均误差保持在4%之内,满足实际作业要求。研究结果可为研发自走式青贮收获机精准抛送填装系统提供借鉴。When the silage harvester operates in the field,the filling and throwing system has a great influence on the efficiency of the silage harvester.Aiming at the feeding process of silage harvester,this study designs a precise throwing system for silage machine based on rotating target detection in order to realize the adaptive throwing of silage crumbs to the wagon compartment.In this study,a filling rule is proposed through the definition of ideal drop point and actual drop point.At the same time,an improved YOLOv5s(Rotation-YOLOv5,R-YOLOv5)was proposed to further realize the accurate detection of rotating targets using machine vision.The feedback of the target and the actual falling point were then calculated to control the movement of the throwing cylinder.The self-adaptive control was achieved in the throwing cylinder using R-YOLOv5.The network structure of the baseline was also optimized to predict the rotating carriages.A new prediction channel of rotation angle was then added to the Head part of the original YOLOv5.The virtual environment was built using Python3.9.Pytorch1.6 was chosen to train the R-YOLOv5 rotating target detection model on the Pytorch deep learning framework.Once the number of training rounds reached 60,the mAP values were all stable at 0.97,indicating a stable system.The improved R-YOLOv5 was recognized as a rectangle with a rotation angle.The recognition range was synchronized with the angle of the carriages.The training data was highly compatible with the carriages for better recognition.In addition,the throwing control was set under the distance between the actual and target drop position.A rotating carriage identification and a field test were carried out to develop the control flow.The test results show that the improved R-YOLOv5 was better performed on the rotating compartments at different rotating angles with an average of more than 97%before,indicating better detection;When the projectile barrel was rotated in the range of-60°-60°,angular velocity was less than 15%/s.The average

关 键 词：机器视觉 R-YOLOv5 青贮收获机 精准抛送 旋转目标检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S24[自动化与计算机技术—计算机科学与技术]