仓储小麦隐蔽性害虫的检测模型及算法  

Model and Algorithm for the Detection of Hidden Insects in Wheat

作　　者：史卫亚[1,2] 乔娜娜[1,2] 梁义涛[1,2] 王锋[1,2] SHI Weiya QIAO Nana LIANG Yitao WANG Feng(School of Information Science and Engineering. Henan University of Technology, Zhcngzhou 450001 , China Food Information Processing and Control Key Laboratory of Education Ministry,Henan University of Technology, Zhengzhou 450001, China)

机构地区：[1]河南工业大学信息科学与工程学院,河南郑州450001 [2]河南工业大学粮食信息处理与控制教育部重点实验室,河南郑州450001

出　　处：《食品与生物技术学报》2016年第6期577-583,共7页Journal of Food Science and Biotechnology

基　　金：国家自然科学基金项目(31171775);国家863计划项目(2012AA101608);河南教育厅自然科学基金项目(14B550002)

摘　　要：粮食在存储过程中极易发生虫蚀现象,因此需要一种快速高效的检测手段来检测粮食是否染虫。结合机器学习和生物光子学的相关理论,分别测量正常和含虫小麦的自发光子数,然后提取8个统计特征和13个直方图特征,分别采用线性判别分析LDA和二次辨别分析QDA算法对正常小麦和含虫小麦进行识别,同时针对小样本情况下协方差矩阵奇异性问题,引入正则化判别分析RDA方法,对QDA算法进行优化,提高分类正确率。实验结果证实了所提方法的有效性。In order to prevent the loss of grain mass and quality, a fast and efficient method for the early detection of insects in grains is urgently needed during trade and storage. Based on the biophoton analytical technology （BPAT）,the experiments were made to measure spontaneous photon counts of wheat kernels and infested ones. Then statistical characteristics and histogram distribution were extracted and linear discriminant analysis （LDA） and quadratic discriminant analysis（QDA） were used to discriminate between normal and infested grains. In addition, due to the singularity and instability of the per class covariance matrices in the small sample,regularized discriminant analysis （RDA） was used to optimize QDA and increase the classification accuracy. Therefore, the proposed method is workable.

关 键 词：生物光子 统计特征 直方图特征 识别 

分 类 号：TS210.7[轻工技术与工程—粮食、油脂及植物蛋白工程]