Acrylamide prediction on chestnut-based biscuits, a shaped matter

The detection and removal of contaminants through non-contact sensors remain challenging in bakery products. The chemical composition and the brown colour of goods made with alternative flours, such as chestnut-flour-based dough, may compromise the sensitivity of colour and hyperspectral analysis, which are widely acknowledged non-contact techniques for acrylamide detection in heated goods. Furthermore, empirical models proposed in the literature often neglect the effect of product shape and its relationship with acrylamide formation. In the present study, chestnut flour-based biscuits were shaped into four shapes and baked at incremental times. Colour, hyperspectral, and thermal analyses of the biscuit surface were correlated with acrylamide content, measured by an ELISA-based method. In addition, a morphological evaluation was performed using three-dimensional photogrammetry, enabling the extraction of shape descriptors relevant to the biscuit's surface warming dynamics. Acrylamide content was predicted using features derived from these datasets. Colour analysis outperformed the other feature types. Lightness and the blue-yellow component yielded more meaningful results than the green-red component. Among the different biscuit shapes, the measured surface area and surface roughness emerged as relevant factors for adjusting acrylamide content in each biscuit, highlighting the importance of morphology in predictive modelling. Finally, the best prediction model could estimate acrylamide content across the investigated baking conditions, with an average prediction error of 44.59 μg kg−1, by combining two colour-derived predictors with two morphological descriptors.