Rapid and non-destructive discrimination of stale seeds of Torreya grandis based on near-infrared spectroscopy and variable wavelength selection

Abstract: Due to the oxidation of unsaturated fatty acids during storage, the taste and quality of Torreya grandis seeds declines. Unscrupulous merchants, seeking huge profits, blend Torreya stale seeds with fresh ones for sale, infringing upon consumers’ interests. A fast and non-destructive identification method is needed . Methods: In this research, near-infrared spectroscopy was used to conduct rapid and non-destructive discrimination on stale Torreya grandis seeds. Spectra of shelled Torreya grandis seeds samples were collected in the wavelength ranges of 200-1160 nm and 900-1700 nm using two near-infrared spectrometers. Nine methods were employed to preprocess the spectral data. Then, four wavelength selection methods, namely interval optimization selection algorithm (ICO), competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), and variable combination population analysis (VCPA), were utilized to screen the spectral characteristic variables of stale Torreya grandis seeds. Linear discriminant analysis (LDA), support vector machine (SVM), and backpropagation neural network (BP) methods were applied to establish discrimination models for stale Torreya grandis seeds. Results: The results indicate that for spectrometer 1, the CARS method is the optimal wavelength selection method, and the CARS-SVM model exhibits the best performance, with sensitivity, specificity, and accuracy all reaching 100% in the prediction set. For spectrometer 2, standardization and SNV are superior preprocessing methods. The VCPA variable selection method outperforms the other three methods, and the established optimal model is VCPA-BP, with the sensitivity, specificity, and accuracy of the model’s prediction set being 98.18%, 93.02%, and 95.04%, respectively. Conclusions: Thus, it can be concluded that the discrimination models established based on the data from both spectrometers can effectively discriminate stale Torreya grandis seeds, and the overall performance of spectrometer 1 is superior to that of spectrometer 2. This study can provide a detection method for the rapid and non-destructive discrimination of stale Torreya grandis seeds, effectively guaranteeing the quality of Torreya grandis seeds.