Apple Grading Using Principal Component Analysis and Kernel Fisher Discriminant Analysis Combined with NIR Spectroscopy

Abstract

Principal component analysis (PCA) and kernel Fisher discriminant analysis (KFDA) were applied to grade Fuji apples combined with near infrared reflectance (NIR) spectroscopy. Firstly, PCA was used to reduce the dimensionality of NIR spectra acquired by the Antaris II FT-NIR spectrophotometer on apples. Secondly, nonlinear discriminant information was extracted by kernel Fisher discriminant analysis (KFDA). Finally, the k-nearest neighbors algorithm with leave one out strategy was utilized to classify apple samples into two grades. LDA can only solve linearly separable problems, and it is not suitable in solving some nonlinear problems. But unlike LDA, KFDA can solve nonlinearly separable problems, and it projects data onto a high-dimensional feature space by the nonlinear mapping. Experimental results showed that KFDA achieved higher classification rate compared with LDA.