Quantitative detection of aflatoxin B1 in peanuts using Raman spectra and multivariate analysis methods

Aflatoxin B1 (AFB1), among the identified aflatoxins, exhibits the highest content, possesses the most potent toxicity, and poses the gravest threat. It is commonly found in peanuts and their derivatives. This study employs Raman spectroscopy to monitor the AFB1 levels in moldy peanuts, providing a reliable theoretical basis for peanut storage management. Firstly, different degrees of moldy peanuts are spectrally characterized using a portable Raman spectrometer. Subsequently, a two-step hybrid strategy for feature selection is proposed, combining backward interval partial least squares (BiPLS) and variable combination population analysis (VCPA), aiming to simplify model complexity and enhance predictive accuracy. Finally, partial least squares (PLS) regression models are constructed based on different feature intervals and wavelength points. The research results reveal that the PLS regression model using the optimized feature intervals and wavelength points exhibits improved predictive capability and generalization performance. Notably, the BiPLS-VCPA-PLS model, established through the two-step optimization, selects nine wavelength variables, achieving a root mean square error of prediction (RMSEP) of 33.3147 mu g center dot kg-1, a correlation coefficient of the prediction set (RP) of 0.9558, and a relative percent deviation (RPD) of 3.4896. These findings demonstrate that the two-step feature optimization method, combining feature interval selection and feature wavelength selection, can more accurately identify optimal variables, thus enhancing detection efficiency and predictive precision.