油茶果壳籽实时分选系统设计与试验  

Design and Experiment of Real-time Sorting System for Camellia Shells and Seeds

作　　者：段宇飞 孙记委 董庚 卢周成 霍俊 王焱清[1,2] Duan Yufei;Sun Jiwei;Dong Geng;Lu Zhoucheng;Huo Jun;Wang Yanqing(Institute of Agricultural Machinery,Hubei University of Technology,Wuhan 430068,China;Hubei Province Engineering Technology Research Center for Intelligent Agricultural Machinery,Wuhan 430068,China)

机构地区：[1]湖北工业大学农机工程研究设计院,武汉430068 [2]湖北省农机装备智能化工程技术研究中心,武汉430068

出　　处：《农机化研究》2024年第10期79-86,共8页Journal of Agricultural Mechanization Research

基　　金：湖北省重点研发计划项目(2020BBA042);湖北工业大学科研启动基金项目(BSQD2017076)。

摘　　要：为了实现油茶果脱壳后果壳与茶籽的高效分选,进行了油茶果壳籽分选装置上料模块、视觉采集模块、执行模块和控制系统的设计。其中,整体装置采用一种阵列式凹槽传送机构,避免执行喷吹时产生对茶籽的带出干扰。利用有效的图像预处理方法获取壳籽目标,并提取目标图像的颜色及纹理信息共18个;考虑到分选的高效实时性,进一步对特征参数实施Spearman相关性分析,将特征减少至6个,建立随机森林分类模型,并与K最邻近分类算法(KNN)、支持向量机(SVM)及偏最小二乘法判别分析(PLS-DA)模型进行对比,发现随机森林模型的分类效果最佳。试验中,分选系统识别准确率达到95.5%,平均检测时间为41.9ms,分选速率达20个/s,表明系统实现了油茶果的壳籽高效率实时分选。In view of achieve efficient sorting of shells and seeds after shelling the Camellia,this study was conducted by designing the loading module,visual acquisition module,execution module and control system of the Camellia shells and seeds sorting device.The overall device adopts an array type groove transfer mechanism,where the grooves were designed as conical tables and square holes existed at the bottom.When the Camellia shells and seeds reach the blowing position,the bottom of the groove is parallel to the surface of the nozzle,and the airflow faces the square holes at the bottom of the groove and corresponds to each other,so as to avoid interference with brought out seeds when the shell is sprayed.The PLC was used to communicate by lower computer with the upper computer in real time and control the sorting actuator.The Matlab 2016a software was used by upper computer to process the Camellia image in real time and output the recognition results.A series of image pre-processing,such as Gaussian noise reduction,color difference operation,threshold segmentation,morphological operation,connected domain processing and image masking,were performed to avoid the interference of background images,and a complete target image of Camellia shells and seeds was obtained,with 18 color and texture information extracted.The analysis of the color information of the Camellia shells and seeds shows that the G-component average can distinguish the inner shell(inner facing upward shell),while in the sorting of the shell(outer facing upward shell)and seeds,considering the efficient real-time sorting,Spearman correlation analysis is further implemented for the feature parameters to reduce the features to 6,and a random forest classification model was established,in which the number of decision trees was based on the out-of-bag mean square error and training time,50 trees were selected,and the random forest model was compared with the K-nearest neighbor classification algorithm(KNN),support vector machine(SVM)and partial least squares d

关 键 词：油茶果 自动分选 机器视觉 特征选择 随机森林 

分 类 号：S226.5[农业科学—农业机械化工程] S220.3[农业科学—农业工程]