Abstrakt

Detection of hidden insect of wheat by biological photon technique

Shi Weiya, Qiao Nana, Liang Yitao, Wang Feng

In order to prevent grain mass and quality loss, a fast and efficient method for early detection of insect infestation of grain is urgently needed during trade and storage. Based on the biophoton analytical technology (BPAT), this work adopted a newmethod of extracting feature by combining statistical characteristics and histogram distribution. Considering the sample covariance matrices of any single class could be singular, the feature vector was compressed by principal component analysis (PCA) and given as inputs to classifiers for the identification of uninfested wheat and infested wheat, such as linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), mahalanobis and linear support vector machines (linear SVM). For further improving the classification accuracy, regularized discriminant analysis (RDA)was presented to optimizeQDAandmahalanobis algorithms. The results proved that the proposed method is workable