基于灰度截留分割与十色模型的马铃薯表面缺陷检测方法  

Method of potato external defects detection based on fast gray intercept threshold segmentation algorithm and ten-color model

作　　者：李锦卫[1] 廖桂平[1] 金晶[1] 虞晓娟[1] 

机构地区：[1]湖南农业大学农业信息研究所,长沙410128

出　　处：《农业工程学报》2010年第10期236-242,共7页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家自然科学基金项目(31071328)

摘　　要：为探索基于计算机视觉的马铃薯表面缺陷检测新方法,该研究提出能将马铃薯表面疑似缺陷一次性分离出来的快速灰度截留分割方法和用于缺陷识别的十色模型。选择面积比率和十色比率作为缺陷判别特征,对分割出来的深色部位采用阈值法进行缺陷识别。采用基于快速G与亮度截留分割的2种方法对发芽进行识别。通过对326个马铃薯样本的652幅正反面图像进行试验,基于十色模型的缺陷识别方法对分割出来的深色区域的正确识别率为93.6%,基于快速G与亮度截留分割2种方法结合对有芽体图像的正确识别率为97.5%,马铃薯表面缺陷正确检测率为95.7%。结果表明,该套方法能快速、有效、方便地检测出黄色薯皮马铃薯的表面缺陷。Correct detection of external defects on potatoes is the key technology in the realization of automatic potato grading and sorting station.This paper reports a novel inspection approach to external defects of potato in three potato cultivars.Fast gray interception segmentation threshold method was proposed to extract the dark part of potato surface.Ten color model was used as color feature for defects detection.Area ratio and ten color ratio threshold were used to identify defects in the segmented dark part of potato surface.Fast interception segmentation based on G and Intensity was used to detect the sprouting potatoes.The tests were conducted with 652 double-faced images for 326 potato tubers.The recognition accuracy rate for the dark part of potato surface segemented,the sprouting potatoes and potatoes external defects detection was 93.6%,97.5% and 95.7%,respectively.The results showed that the method was fast,valid and convenient for defect detection on yellow-skin potatoes.

关 键 词：计算机视觉 模式识别 马铃薯 缺陷检测 

分 类 号：S532[农业科学—作物学] TP391.41[自动化与计算机技术—计算机应用技术]