Deep Learning Dynamic Graphs for Food Microstructures (Delifood)

The global food system significantly contributes to greenhouse gas emissions (26%), land (40%), and water (70%) use, an increase in eutrophication, and biodiversity loss. A key obstacle to sustainable food system transformation is consumers’ rejection of alternative foods. Traditional, macroscopic analyses such as stress measures and chemical composition give a superficial understanding of material properties that are vital to understanding how foods form, age, and react to cooking and eating. Recent developments in super-resolution and non-linear microscopy promise to revolutionize our view of the dynamic behavior of foods. Likewise, recent developments in deep learning promise a revolution in how we can measure and model non-linear, dynamic features of geometric structures and their relations in images.
We hypothesize that network-like microscopic structures are the key to understanding food texture over time. In this project, we lay the foundation for a paradigm shift in how food structures and textures are measured and modeled: we will develop new protocols to image food structures in model systems over time and in different settings, and we will develop data-driven dynamic graph-based deep learning methods to model these systems.