Novel nondestructive NMR method aided by artificial neural network for monitoring the flavor changes of garlic by drying

Abstract

Flavor changes of garlic during drying process were monitored using LF-NMR combined with partial least squares (PLS) and back-propagation artificial neural network (BP-ANN). Results show that with elapsed drying time, the free water (A23) and total moisture content (A) of garlic decrease with different drying conditions. Correspondingly, the sulfide of main flavor components in garlic was significantly reduced, but alcohol and acid components increased slightly and the overall aromatic flavor showed a downward trend, which was consistent with the GC-MS volatile component detection results. Electronic nose sensors S2, S5, S8, S10 were determined as feature sensors by principal component analysis (PCA) and linear discriminant analysis (LDA). The univariate linear model of NMR parameters and electronic nose characteristic sensors show high correlation. Furthermore, ANN and PLS garlic flavor prediction model were established, although the PLS model was not as good as the BP-ANN model (RP 2 of 0.9713 and 0.9975) to monitor flavor changes, it also yields relatively accurate prediction performance with RP 2 of 0.9418 and 0.9633 for mid-shortwave infrared drying and microwave vacuum drying, respectively.