基于计算机视觉对苹果脆片外观品质分级  

External quality classification of apple chips based on computer vision

作　　者：魏康丽 王振杰[1] 孙柯[1] 殷旭[1] 赵保民 屠康[1] 陈飞 朱金星 潘磊庆[1] 

机构地区：[1]南京农业大学食品科技学院,江苏南京210095 [2]江苏派乐滋食品有限公司,江苏徐州221008

出　　处：《南京农业大学学报》2017年第3期547-555,共9页Journal of Nanjing Agricultural University

基　　金：国家自然科学基金项目(31671925;31671926);苏北科技发展计划--科技富民强县项目(BN2015025);中央高校基本科研业务费专项资金(KYLH201504)

摘　　要：[目的]为了开发苹果脆片自动化分级技术,利用计算机视觉对苹果脆片外部品质进行无损检测研究。[方法]通过对苹果脆片的大小、形状、颜色、纹理特征的感官评分建立苹果脆片外部品质等级评价标准。首先,利用计算机视觉系统获取脆片图像,经图像处理后,进行特征提取;然后,通过比较数据离差标准化和标准差标准化处理对费歇尔线性判别(FLD)、偏最小二乘法判别分析(PLS-DA)、支持向量机(SVM)3种模型建模效果的影响,确定合适的数据预处理方式;并分别基于这3种模型对预测样本进行分级;最后,采用连续投影算法(SPA)筛选最佳特征参数,并比较3种模型的建模和预测结果,确定最佳分级模型。[结果]数据经离差标准化处理后建模效果更好;FLD、PLS-DA、SVM 3种模型中SVM模型分级效果最好,其建模集和预测集分级准确率分别为99.1%和98.0%;基于SPA算法筛选得到了面积、圆形度、对比度的均值、能量的均值、对比度的均方差、熵的均方差这6个特征参数,基于特征参数的SVM模型建模集和预测集分级准确率分别为98.7%和97.3%。[结论]该研究实现了苹果脆片的外观品质的自动检测和分级,为苹果脆片的计算机视觉自动分级提供了技术支持。[ Objectives] A non-destructive classification technology for apple chips was built based on computer vision technology. [ Methods] In this paper, grading was implemented according to sensory evaluation score of apple chips based on size, shape, color and texture. Following image acquisition and processing, the features were extracted. Then the data was pretreated using deviation normalization and standard deviation normalization to determine the appropriate handling method by comparing modeling results of three models, including fisher linear discrimination （ FLD）, partial least squares discriminant analysis （PLS-DA） and support vector machine （SVM）. Then these three models were utilized to classify samples. Furthermore, the best classification model was determined by comparing the modeling and classification results based on the feature parameters screened by successive projections algorithm （SPA）. [ Results] The classification performance based on data pretreated by deviation normalization were better than those by standard deviation normalization, and SVM performed best among three models. For the calibration and prediction samples, classifica- tion accuracy was 99.1% and 98.0%, respectively. Six characteristic parameters were selected by SPA, including area, circularity, the mean value of contrast and energy, and the mean square error of contrast and entropy. On the basis of six parameters, SVM had the classification accuracy of 98.7% and 97.3% for calibration samples and prediction samples, respectively. [ Conclusions] This study realized the automatic detection and classification of external quality of apple chips, and provided a technical support for automatic classification of apple chips based on computer vision.

关 键 词：苹果脆片 计算机视觉 连续投影算法 模型 分级 

分 类 号：TS255[轻工技术与工程—农产品加工及贮藏工程]