CAJ | 학술논문

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％。
이신강무핵백선포도위연구대상，채용궤기시각기술예측포도수적질량。수선，제취 RGB 도상，주 G ， B쌍통도분량가운산R＋B ，채용고사저통려파법려제도상중적조음，채용 Gamma 변환법조정도상회도，종이증강전경여배경적대비도。기차，채용자동역치분할법분할도상，경수학형태학적부식화개운산획득최가이치도상，제취이치도상중목표구역적궤하특정。최후，채용일원선성회귀、다원선성회귀화편최소이승회귀예측포도수적질량。결과표명，제취분할후적포도수면적、주장、장축급단축장도등특정건립적편최소이승회귀모형，기예측포도수질량효과최가，상관계수r2위96．91％。
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%.