Monitoring of solid-state fermentation of wheat straw in a pilot scale using FT-NIR spectroscopy and support vector data description

Abstract

Fourier transform near-infrared (FT-NIR) spectroscopy coupled with support vector data description (SVDD) as an ideal tool was attempted to rapidly and accurately monitor physical and chemical changes in solid-state fermentation (SSF) of crop straws without the need for chemical analysis. Raw spectra of fermented samples were acquired with wavelength range of 10,000–4000cm−1. SVDD algorithm was employed to build a one-class classification model, and some parameters of SVDD algorithm were optimized by cross-validation in calibrating model. Simultaneously, four traditional two-class classification approaches (i.e., linear discriminant analysis, LDA; K-nearest neighbor, KNN; back propagation neural networks, BPNN; support vector machine, SVM) were comparatively utilized for monitoring time-related changes that occur during SSF. Compared to the four models, SVDD model revealed its incomparable superiority in handling the problem of imbalance training sets. The discrimination rate of SVDD model was 90% in the validation set when the ratio of samples from stationary stage to those from other stages was one to eight. This study demonstrates that FT-NIR spectroscopy combined with SVDD is an efficient method to develop one-class classification model for the rapid monitoring of SSF.