Towards the establishment of a green and sustainable analytical methodology for hyperspectral imaging-based authentication of wholemeal bread

Abstract

The increased interest in whole-grain products, along non unified European regulations on the composition of wholemeal bread could lead to its misleading labelling. Therefore, to ensure the safety of consumers, both health-wise and from possible fraud, a novel hyperspectral imaging-based authentication method is being proposed, given that no previous study has combined imaging techniques with authentication of wholemeal flour content. Quantification based on pixel counting by classification (QPC) utilizes multivariate analysis methods such as PLS-DA and SVM to classify pixels within a bread sample (as Wholemeal and White flour), based on their visible-near-infrared spectra, to later estimate its proportion of wholemeal flour. This is in accordance with the heterogeneous nature of bread samples, in which individual pixels belonging to both wholemeal and white flour can be accounted for, which was proved by implementing unsupervised training techniques such as hierarchical cluster analysis (HCA). Results show that the quantification model was able to successfully predict wholemeal flour content with a maximum deviation of 8 g wholemeal flour/100 g flour from the estimated value.