Applying machine learning methods in studying relationships between mouthfeel and microstructure of oat bread

The aim was to study whether machine learning can be applied in analysing, microstructure of food from microscopy images and whether the state of macromolecules, protein and starch, can be related to sensory mouthfeel. Addition of gluten and transglutaminase modified the microstructure and mouthfeel of five oat bread samples. Light microscopy, instrumental texture profile analysis and descriptive sensory analysis were used to analyse the test breads. Digital image analysis was used to obtain numerical data on starch and protein phases in the breads. Degree of starch gelatinization and protein network properties were extracted from microscopy images by expert ratings. Many of the bread sensory properties could be predicted using instrumental texture and image analysis features. Machine learning of degree of starch gelatinization and protein network structure properties from expert classified microscope images was studied. Color histograms and co-occurrence matrix statistics were suitable preprocessing techniques for the images. In both starch gelatinization and protein network structure classification, the prediction error of an induced model tree was lower than the standard deviation between independent predictions given by experts.