无核白鲜葡萄机器视觉质量预测研究!  

Prediction of the Weight of Xinjiang Grape by Machine Vision Techniques

作　　者：李俊伟[1] 郭俊先[1] 张学军[1] 胡光辉[1] 程国首[1] 

机构地区：[1]新疆农业大学机械交通学院,乌鲁木齐830052

出　　处：《农机化研究》2014年第7期57-61,66,共6页Journal of Agricultural Mechanization Research

基　　金：新疆维吾尔自治区科技厅基金项目(2009211B07);国家自然科学基金项目(61005022)

摘　　要：以新疆无核白鲜葡萄为研究对象,采用机器视觉技术预测葡萄穗的质量。首先,提取RGB图像,做G,B双通道分量加运算R+B,采用高斯低通滤波法滤除图像中的噪音,采用Gamma变换法调整图像灰度,从而增强前景与背景的对比度。其次,采用自动阈值分割法分割图像,经数学形态学的腐蚀和开运算获得最佳二值图像,提取二值图像中目标区域的几何特征。最后,采用一元线性回归、多元线性回归和偏最小二乘回归预测葡萄穗的质量。结果表明,提取分割后的葡萄穗面积、周长、长轴及短轴长度等特征建立的偏最小二乘回归模型,其预测葡萄穗质量效果最佳,相关系数r2为96.91%。The object of this study is to forecast weight of Xinjiang Wu Hebai grape spike by using machine vision tech -nology .Extracting the RGB image , add operation R＋B of G and B dual channel component , gaussian low pass filtering method for filtering noise in images and gamma transform method for adjusting the image gray level are used to enhance the visibility of the foreground and background .Furthermore ,the automatic threshold segmentation method is used to split image;The corrosion and opening function of mathematical morphology is used to get best binary image and extract the ge -ometrical characteristics of the target area in binary image .For the last , monadic linear regression , multiple linear regres-sion and partial least-squares regression are used to predict grape spike weight .Results show that partial least-squares regression model which is established by the area ,perimeter ,length of long axis and short axis of the grape spike in the segmentation images after extraction ,predicts the best weight effect of this method with correlation coefficient r2 96 .91%.

关 键 词：机器视觉 图像处理 无核白鲜葡萄 偏最小二乘回归 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]