DETERMINATION OF RICE SEED VIGOR BY LOW-FIELD NUCLEAR MAGNETIC RESONANCE COUPLED WITH MACHINE LEARNING

Abstract

Physiological index data and low-field nuclear magnetic resonance (LF-NMR) spectral data of rice seed samples from three varieties harvested in different years were collected through a combination of the standard germination test and an LF-NMR test. Three parameters of seed vigor: germination energy, germination percentage, and germination index, were calculated based on the physiological index data of the rice seed samples to determine their vigor over the years after harvest. LF-NMR Carr-Purcell-Meiboom-Gill (CPMG) sequence echo-peak data were used as the input, and rice seed vigor was used as the output to establish discriminative models using principal component analysis, support vector machine, logistic regression, K-nearest neighbor, artificial neural network, and Fisher’s linear discriminant. The results showed that models constructed using any algorithm, except for principal components analysis-algorithm distinguished between seeds with high and low vigor, while models constructed using Fisher’s linear discriminant algorithm gave the best results. This study provided a rapid, accurate, and non-destructive method to test rice seed vigor, offering theoretical support and a reference for rice seed-sorting and storage research.