摘要
目前青梅的缺陷识别检测仍然依靠人工挑选方式来完成,但人工挑选方式受工作经验、劳动强度等因素制约,已经难以适应产业的发展。为有效提高青梅表面缺陷检测的自动化程度和检测精度,本试验应用机器视觉技术针对青梅表面的缺陷检测展开研究。通过搭建青梅表面图像静态采集系统,采用图像处理软件HALCON对青梅表面进行了单通道灰度图像提取、图像滤波、灰度二值化及特征提取等预处理操作,实现了对青梅表面图像的去背景化,并利用去边缘法在青梅H通道分量图像中成功提取到青梅表面缺陷。最后采用高斯混合模型构建青梅表面缺陷检测分类器,并创建了一套基于机器视觉的青梅表面缺陷检测系统。具体选取了348张青梅缺陷图像作为训练测试样本,其中78%的图像作为训练集,22%的图像作为测试集,结果表明:该分类器对青梅溃烂、伤疤、雨斑缺陷的检测准确率分别为100%,97.22%,92.31%,对完好青梅检测准确率为94.44%,验证了将高斯混合模型应用在青梅缺陷检测方面的有效性。
Green plum has high nutritional and medicinal values,which is very beneficial to human protein composition and normal metabolism.The defects and main index components of green plum have an important influence on its fine and deep processing.At present,the defect identification and detection of green plum still depends on manual selection,but the manual selection method is restricted by work experience,labor intensity and other factors,which is difficult to be adapted by the industrial production.In order to improve the quality and added value of green plum products,this experiment used machine vision technology to carry out the research on green plum surface defect detection and recognition technology,so as to effectively improve the degree of automation and detection accuracy of green plum surface defect detection.In this study,a static image acquisition system for the surface of green plum was built.The single channel gray image extraction,image filtering,gray binarization and feature extraction were performed on the surface of green plum using the image processing software HALCON.The background of the surface image of green plum was removed,and the H-channel component image of green plum was obtained using the de-edge method.After comparing the 77 defect images of green plum in the training set with the Gaussian mixture model(GMM),multi-layer perceptron(MLP),and support vector machine(SVM)classifiers,it was found that the number of misjudgments of GMM classifier was smaller than those of MLP and SVM.Among them,the accuracy of GMM s defect detection using GMM s and classification of green plum reached 96.10%.Compared with the MLP and SVM algorithms,the accuracy of defect detection using GMM s increased by 14.28%and 7.79%,respectively,indicating that the use of multi-modality to accurately quantify green plum compared with MLP and SVM,Gaussian mixture model was more suitable for the detection of green plum surface defects.Therefore,the Gaussian mixture model was used to classify the surface defect detection classifier of greengage,and a set of machine vision-based surface defect detection system of greengage was established.348 defect images of green plum were selected as training test samples,78%of which were used as training set and 22%as testing set.The results showed that the accuracy of the classifier in detecting canker,scar and rainspot defects of green plum was 100%,97.22%,92.31%,respectively,and 94.44%for perfect green plum.The validity of Gaussian mixture model in defect detection of green plum was verified.
作者
刘阳
丁奉龙
刘英
沈鹭翔
董瑞文
LIU Yang;DING Fenglong;LIU Ying;SHEN Luxiang;DONG Ruiwen(College of Mechanical and Electronic Engineering,Nanjing Forestry University,Nanjing 210037,China)
出处
《林业工程学报》
CSCD
北大核心
2020年第4期139-144,共6页
Journal of Forestry Engineering
基金
江苏省农业科技自主创新资金[CX(18)3071]。
关键词
机器视觉
青梅
缺陷检测
分级
高斯混合模型
machine vision
green plum
defect detection
classification
Gaussian mixture model(GMM)