Hyperspectral imaging combined with chemometrics for rapid detection of talcum powder adulterated in wheat flour

Abstract

Detection of illegal additives in wheat flour is of crucial practical significance to ensure the market order and people's health. In this study, the potential of hyperspectral imaging in 900–1700 nm for the rapid detection of talcum powder adulterated in wheat flour was investigated. The raw spectral information within the hyperspectral images of samples were extracted and preprocessed with standard normal variate (SNV), baseline correction (BC), multiplicative scatter correction (MSC) and Gaussian filter smoothing (GFS), respectively. Partial least square (PLS) quantitative models with the preprocessed spectra were constructed and assessed to predict the talcum powder concentration in wheat flour. The results showed that the SNV-PLS model had best performance, with correlation coefficient (r) of 0.98 for both calibration and prediction, residual predictive deviation (RPD) of 4.69, and root mean square error of calibration (RMSEC), cross-validation (RMSECV) and prediction (RMSEP) of 2.86%, 2.99%, 3.13%, respectively. The characteristic wavelengths were then respectively selected by regression coefficient (RC), successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS) algorithm to simplify the SNV-PLS model. Based on the four characteristic wavelengths (907.135, 1339.866, 1392.573 and 1394.22 nm) selected by CARS method, the SNV-CARS-PLS model was constructed and had best performance in predicting talcum powder content, leading to rP of 0.98, RMSEP of 2.88% and RPD of 5.09. The whole study indicated that the hyperspectral imaging in 900–1700 nm range combined with CARS method could be used for further developing a portable detection equipment to realize the rapid and accurate talcum detection in wheat flour.