Photons for Food Intake: measuring macronutrients with hyperspectral imaging and machine learning

What if a device could determine the nutritional content of a meal simply by imaging it? That is the vision behind this research. Measuring what people actually eat is surprisingly difficult. Current methods rely on individuals remembering and reporting their own food intake, which is often inaccurate and time-consuming. This thesis explores a more objective alternative: hyperspectral imaging, a camera technology that captures detailed chemical information from food, far beyond what the human eye can see. Combined with machine learning, the technology behind many modern pattern recognition systems, these cameras can predict the fat and protein content of foods such as sandwiches and cheese, without any physical contact or destruction of the sample. This research demonstrates that this combination is effective, and lays the groundwork for a future automated system: point a device at a plate, and instantly obtain its nutritional composition. A meaningful step toward more accurate and objective measurement of dietary intake.