Quantitative analysis of starch adulterants in peanut coating powder based on FT-NIR and chemometric methods

Peanut skin, known for their high medicinal value, can experience reduced efficacy when adulterated with starchy substances. This study presents an efficient detection method based on Fourier Transform Near-Infrared Spectroscopy (FT-NIR) technology for adulterated peanut skin powder found in the market. Potato starch and corn starch were selected as common adulterants, and a range of samples with different adulteration levels was prepared. Then, Chemometric methods were employed to build quantitative analysis models for the adulterants. First, the collected spectral data were preprocessed. Subsequently, feature selection was conducted using Competitive Adaptive Reweighted Sampling (CARS), Bootstrapping Soft Shrinkage (BOSS), and Variable Combination Population Analysis (VCPA) algorithms. Finally, Partial Least Squares Regression (PLSR), Support Vector Machine (SVM), and Random Forest (RF) models were applied to quantitatively analyze the adulteration levels of different substances, and their predictive performances were compared. Experimental results revealed that the BOSS feature selection method combined with the SVM model yielded the highest prediction accuracy for potato starch adulteration, achieving an $R_P^2$ of 0.9953. For corn starch adulteration, the highest accuracy was achieved using the full-variable dataset with the SVM model, reaching an $R_P^2$ of 0.9957. These findings demonstrate that integrating FT-NIR technology with advanced machine learning techniques enables high-precision identification and quantification of different adulterants in peanut skins, providing an efficient and reliable approach for food quality monitoring.