基于机器视觉的菠萝自动去眼方法与试验  

Automatic removal of pineapple eye using machine vision

作　　者：刘安稳 谢方平[1,2] 向阳 李亚军[1,3] 雷翔茗[1] LIU Anwen;XIE Fangping;XIANG Yang;LI Yajun;LEI Xiangming(Department of Mechanical and Electrical engineering,Hunan Agricultural University,Changsha 410128,China;Hunan Key Laboratory of Intelligent Agricultural Machinery Equipment,Changsha 410128,China;Intelligent Equipment Research Center,Beijing Academy of Agriculture and Forestry Sciences,Beijing 100097,China)

机构地区：[1]湖南农业大学机电工程学院,长沙410128 [2]智能农机装备湖南省重点试验室,长沙410128 [3]北京农林科学院智能装备研究中心,北京100097

出　　处：《农业工程学报》2024年第1期80-89,共10页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家重点研发计划项目(2022YFD2002003-3)。

摘　　要：菠萝加工过程中的去眼作业目前主要依赖人工手动操作,劳动成本高且作业效率低。为实现自动化菠萝去眼作业,该研究运用YOLOv5目标检测算法对菠萝眼进行快速识别,将所有角度相差90o的两张图片作为一组进行立体匹配分析以获取菠萝眼的三维位置信息。通过旋转菠萝和轴向移动去眼刀具将专用去眼刀具依次对准每一个菠萝眼并快速去除。试验结果表明:YOLOv5目标检测算法对菠萝眼识别效果良好,验证集的准确率、召回率和平均精度均值均高于96%;菠萝眼实际中心与探针刺入位置的平均误差为1.01 mm,最大误差为2.17 mm,均方根误差为1.09 mm;菠萝眼的完全去除率为89.5%、不完全去除率为6.2%,漏检率为4.3%,单个菠萝去眼时间为110.9 s,基本满足自动化去眼作业需要。研究结果可为菠萝自动去眼机研发提供技术参考。Eye removal operation has been one of the most important steps in the process of pineapple production.However,manual operation cannot fully meet the large-scale production at present,due the high labor cost and low efficiency.In this study,an automatic eye removal machine was developed for the pineapple using machine vision.The automatic control system included the feeding clamping,image acquisition and processing,eye removal execution and motion control.A total of 480 images of 40 pineapples were firstly collected at 30°angle intervals throughout entire circumference.Then,data enhancement processing(such as rotation,horizontal and vertical image)was used to improve the robustness of recognition mode.1000 pineapple images were obtained as the datasets.According to 8:1:1,the datasets were divided into the training set,validation set and test set.Four YOLOv5(l,s,m,and x)models were used to train the data set of pineapple eyes.The 100 images in the test set were input into four models for manual detection.The optimal network model was selected after that.The YOLOv5l target detection was used to rapidly identify the pineapple eye.Two pictures with a difference of 90 in all angles were analyzed as a group to obtain the three-dimensional position of pineapple eye.Geometric vector was utilized to obtain the camera threedimensional coordinate system(XC,YC,and ZC)with the origin of the camera,and the three-dimensional space coordinate of pineapple eye(X,Y,and Z).The probe positioning test was carried out to evaluate the positioning accuracy of the system.Among them,the three-dimensional spatial coordinates(X,Y,Z)of pineapple eyes were transformed into(L,θ)spatial coordinates with the horizontal position L of the probe and the pineapple rotation angleθas the references.A numericalcontrolled precision motion was applied to pierce the probes into the pineapple eyes,in order to validate the effect of the recognition and localization.In subsequent experiments for practical engineering applications,the three-dimensional spac

关 键 词：农业机械 机器视觉 设计 菠萝眼 YOLOv5 三维定位 

分 类 号：S24[农业科学—农业电气化与自动化]