Hyperspectral imaging for predicting and visualizing the acrylamide levels in roasted coffee

Since it is carcinogenic to humans, the acrylamide content in heat-processed foods should be detected and controlled. This study investigated the spatial distribution and visualization of the acrylamide content in roasted coffee. First, hyperspectral images were collected, from which the spectral-pixel features were extracted. Besides partial least square (PLS), five wavelengths (888, 1123, 1456, 1636, and 1734 nm) based on the successive projections algorithm (SPA) were used to establish prediction models (multiple linear regression (MLR), CatBoost, and XgBoost). SPA-MLR presented the optimal outcome, with a prediction coefficient of determination (Rp2) and a root mean square error of prediction (RMSEP) of 0.81 and 35.90 mu g/kg, respectively. The acrylamide content values for all pixels in the hyperspectral images were subsequently calculated using the prediction equation, thereby producing the spatial distribution and visualization images. The results demonstrated that hyperspectral imaging can effectively predict and visualize the acrylamide levels in roasted coffee.