基于机器视觉的大枣表面缺陷检测

以中国灵武长枣为试验对象,采用Halcon12.0视觉处理软件,通过支持向量机的方法,在HSI颜色空间中提取H分量的均值、方差作为颜色特征值。通过试验选择高斯核函数,当核函数为0.2,正则常数为0.005时达到较好的分级效果,准确率达到94.6%,大大提高了大枣无损在线检测效率,降低了劳动强度和成本。

Detection of the Chinese Jujube surface defects by machine vision

Based on LingWu jujube as experimental object, used the Halcon12.0 visual processing software by the method of support vector machine (SVM) in IHS color space to extract the mean value and mean variance of H component as the color eigenvalues. Selected the gaussian kernel function by the experiments. When the kernel parameter was 0.2, and the regular constant was 0.005, the accuracy rate was 94.6%, which greatly improved the efficiency of nondestructive on-line detection, decreased the labor cost and labor intensity, and eliminated the scruple on the accuracy of on-line detection for jujube to processors. It has large research significance in fruit grading.