Oleogels represent a promising alternative to trans and saturated fats, as they enable the structuring of liquid oils into semi-solid systems while improving nutritional quality, texture, and stability. Their development relies on oleogelators, which are considered food additives and must be carefully selected and regulated due to differences in functionality, cost, and acceptance.We investigated structural variations in oleogels produced with different oleogelator types and concentrations, using polarized light microscopy images coupled with deep learning and explainable artificial intelligence (XAI). A convolutional neural network (CNN) was employed to classify the samples and explore the relationship between crystalline organization and oil-holding capacity. CNN achieved 95 % accuracy in distinguishing between oleogels formulated with beeswax (BW) and glycerol monostearate (GM), and 77 % accuracy in classifying their concentrations. Morphological analyses revealed that BW-based oleogels form platelet-shaped crystals, while GM-based oleogels exhibit irregular clusters of needle-shaped crystals. XAI, using Grad-CAM++, highlighted that CNN predictions were primarily based on these structural features, particularly at lower concentrations. At higher concentrations (≥6 % w/w), similarities between BW and GM reduced classification accuracy. Overall, integrating microscopy with deep learning and XAI proved to be a robust and innovative methodology for characterizing oleogel structures, offering valuable insights for formulation and development.
扫码关注我们
求助内容:
应助结果提醒方式:
