Discrimination Analysis of Moldy Chinese Chestnut Using Artificial Neural Network Model based on Near Infrared Spectra

摘要

The nondestructive discrimination of the shelled chestnuts was studied with near infrared spectra, which could provide a new method for quality detection of other shelled agricultural products. 178 chestnut samples were prepared and their diffuse reflectance spectroscopy were collected in the wave number range of 12 000 - 4 000 cm-1.First，six preprocessing techniques including smooth、vector normalization、min-max normalization、standard normal variate transformation、multiplication scattering correction and first derivative were used to process the original spectrum. Then，principal component analysis was applied to compress thousands of spectral data into several variables and collect spectral information. The principal components extracted by PCA were employed as the inputs of the BP neural networks. Effects of 6 preprocessing techniques for the model based on BP neural network were compared. The results indicated that prediction precision varied to different preprocessing techniques. The optimum network structure of 7-4-1 was obtained after vector normalization method done. Discriminating rate of qualified chestnut, surface moldy chestnut and internal moldy chestnut in prediction set 94.74%、94.44% and 92.31%, respectively, were achieved.