Quantitative detection of metanil yellow adulteration in chickpea flour using line-scan near-infrared hyperspectral imaging with partial least square regression and one-dimensional convolutional neural network

Food adulteration is a serious food safety issue, and it is visually difficult to detect metanil yellow adulteration in chickpea flour. The objective of this study was to develop a non-destructive near infrared (NIR) hyperspectral imaging system (HSI) for quantifying metanil yellow adulteration in chickpea flour. Pure chickpea flour was adulterated with metanil yellow over a range of 0–2% (w/w). The images of 150 adulterated samples and 10 control samples were used for generating calibration and predictions sets. Partial least squares regression (PLSR) models were developed based on different spectral preprocessing techniques, full spectrum and wavelengths selected through competitive adaptive reweighted sampling (CARS), iteratively retaining informative variables (IRIV) algorithms for predicting the adulterant. Further, one-dimensional convolutional neural network (1D-CNN) was used for calibration model development to predict adulterant concentration. When using full spectra, PLSR yielded a model with correlation coefficient of prediction (R2p) of 0.978 with 2nd derivative preprocessing, whereas 1D-CNN produced a model with R2 of 0.992 with no spectral preprocessing for adulterant prediction. Furthermore, IRIV selected wavelengths with no preprocessing and CARS selected wavelengths with 2nd derivative preprocessing along with PLSR yielded R2 of 0.989.