Abstract: A new method was developed to discriminate the quality of bean oil based on near-infrared transmission spectra and BP neural network. The near-infrared transmission spectra of qualified oil and unqualified oil (refined waste oil, fried oil and degenerative oil) were obtained from 10 000 to 3 500 cm-1. Spectral data were preprocessed by Savitky-Golay and baseline correction. Nine principal components (Cumulative contribution rate is 99.502%) extracted by SPSS 11.0 acted as input nerve cell of neural network, and the BP neural network model was build. The model could discriminate qualified oil from unqualified oil, even unqualified kind. Calculation results showed that the distinguishing rate was 100%．

Key words: near-infrared transmission spectra, principal component analysis, BP neural network, bean oil discrimination