计算机视觉技术在核桃壳、仁及分心木识别中的应用研究  

Application of computer vision technology on identification of walnut shell, kernel and distractor wood

作　　者：李成吉 张淑娟[1] 邢书海 陈彩虹 孙海霞[1] Li Chengji;Zhang Shujuan;Xing Shuhai;Chen Caihong;Sun Haixia(College of Engineering,Shanxi Agricultural University,Taigu 030801,China)

机构地区：[1]山西农业大学工学院,山西太谷030801

出　　处：《山西农业大学学报（自然科学版）》2018年第11期20-26,共7页Journal of Shanxi Agricultural University(Natural Science Edition)

基　　金：国家自然科学基金(31271973);晋中市科技重点研发计划(农业)项目(Y172007-4)

摘　　要：[目的]为了研究快速准确识别核桃壳、仁及分心木的方法,[方法]本研究以太谷"清香"核桃为研究对象,搭建计算机视觉系统获取核桃壳、仁及分心木的图像信息,提取各样本的12个颜色特征值(RGB和HSI各分量的均值与方差),以及利用灰度共生矩阵(gray level co-occurrence matrix,GLCM)分别提取各样本能量、熵、惯性矩、相关性、逆差矩4个方向(0°,45°,90°,135°)共20个纹理特征参数。以颜色特征值、纹理特征值、颜色-纹理特征值融合作为输入建立3种最小二乘支持向量机(LS-SVM)模型(Y-LS-SVM、W-LS-SVM、Y-W-LS-SVM),并对各预测集样本进行判别。[结果]结果表明,Y-LS-SVM模型对核桃仁、分心木及核桃壳的判别准确率分别为91.1%、82.4%、96.6%;W-LS-SVM模型对核桃仁、分心木及核桃壳的判别准确率分别为100%、85.3%、84.7%;Y-W-LS-SVM模型对核桃仁、分心木及核桃壳的判别准确率分别为93.3%、97.1%、100%。Y-W-LS-SVM模型的判别效果最好。[结论]本研究表明,基于计算机视觉技术能够很好地对核桃壳、仁及分心木进行识别,为核桃壳、仁及分心木在线分选技术提供理论基础。[Objective]Present study was aimed at developing a computer vision technology based method for rapid and accurate identification of walnut shell,kernel,and distraction wood.[Methods]Local walnut variety"Qingxiang"was selected as research material.The computer vision system was built up to obtain the image information of walnut shell,kernel and distraction wood.A total of 20 texture feature parameters including the energy,entropy,moment of inertia moment,relevance,and deficit four direction(0°,45°,90°,135°)were extracted using 12 color eigenvalues(the mean and variance of each component of RGB and HSI of each sample)coupled with gray level co-occurrence matrix(GLCM),.Three least squares support vector machine(LS-SVM)models,named as Y-LS-SVM,W-LS-SVM,and Y-W-LS-SVM,were established based on color feature values,texture feature values,and color-texture feature values as inputs,and each prediction of a set of samples was discriminated.[Results]The results showed that the accuracy of Y-LS-SVM model for walnut kernel,distraction wood,and walnut shell was 91.1%,82.4%,96.6%,respectively;the accuracy of W-LS-SVM model for walnut kernel,distraction wood,and walnut shell was 100%,85.3%,84.7%,respectively;and the accuracy of Y-W-LS-SVM model for walnut kernel,distraction wood and walnut shell was 93.3%,97.1%,100%,respectively.Therefore,the Y-W-LS-SVM model possessed the best discrimination effect.[Conclusion]This study indicated that the computer vision technology is able to successfully identify walnut shell,kernel and distraction wood,and the results provide theoretical basis for online sorting technology of walnut shell,kernel and distraction wood.

关 键 词：计算机视觉 核桃 识别 最小二乘支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S664.1[自动化与计算机技术—计算机科学与技术]