This thesis (open access) set out to prove the concept that multi-spectral imaging (MSI) could be used to build a classification model to differentiate chicken breast with undeclared added water from that with no, or legally-permitted low, added water content.
The researcher built a model based upon an in-house reference set of chicken breast samples; 12 with no added water, 12 with water added at a level that need not be legally declared (3 – 5%) and 12 with water added at a level that should be legally declared on-pack (9-11%). The protein/water content of the samples was then calculated using classical analysis, in order to label the MSI scans. MSI used two cameras , FX10 and FX17. After annotation, the samples were saved and analysed in MATLAB for model development
The researcher concluded that the method holds promise but would need a much more robust database. With this limited database, the model could distinguish added-water from non-added-water samples but could not robustly distinguish between amounts of added water which would be legal if undeclared and those which would not be legal.
Photo by Philippe Zuber on Unsplash
