Discriminate of Moldy Chestnut Based on Near Infrared Spectroscopy and Feature Extraction by Fourier Transform

Abstract

As near infrared spectra has the characters of multi-variables and strong correlations, to solve the problem, Fourier transform (FT) was used to extract feature variables of shelled chestnuts spectra. FT coefficients and the status of 178 chestnuts were selected as inputs and outputs of the back-propagation neural network (BPNN) classifier to build a recognition model. For comparison, principal component analysis (PCA) was utilized to compress the variables, which then was introduced as input of the neural network model. The results demonstrate that FT is a powerful feature extraction method and is better than PCA as a feature extraction method when employed together with BPNN. When the preprocessing method of standard normal variate transformation(SNV) was carried out and the first 15-point FT coefficients were used as the input, an optimal network structure of 15-6-1 was obtained, where discriminating rates of qualified chestnut, surface moldy chestnut and internal moldy chestnut in prediction set are 100%, 100% and 92.31%, respectively.

NIR-Based Wood Water Content Prediction with an Integration of ANN and PCA

“…Artificial neural network (ANN) is an optimal approach of the multivariate techniques for obtaining accurate and reliable analysis results. It has obvious superiority in nonlinear data processing [7][8][9]. Back Propagation (BP) neural network is a kind of multilayer feed-forward network with the advantages of plasticity and simple structure.…”