Personal(ised) sensors or ”wearables” could, in the future,be applied to provide personalized nutritional advice. A challenge here is to assess dietary intake. To date, this has mainly been done with questionnaires or interviews, for example, food frequency questionnaires or 24-hour recalls. However, these methods are prone to bias due to conscious or unconscious misreports,and more objective measurement methods are desirable.By applying and combining spectral imaging techniques like hyperspectral imaging (HSI) and RGB-depth (RGBD) imaging, information on the macro-composition, identity and quantity of food consumed can be obtained. In this work, we demonstrate that HSI was effective for estimation of the fat content and layer thickness of butter on slices of bread with root mean squared errors of predictions of 4.6 (fat w/w %) and 0.056 mm respectively.Identification and volume estimation of vegetables and preparation methods were successful with RGBD imaging. Using Convolutional Neural Networks, all samples were correctly identified. For volume estimations of vegetables, R-square scores between 0.80 – 0.96 were achieved.
|Title of host publication||OCM 2021 - Optical Characterization of Materials: Conference Proceedings|
|Editors||J. Beyerer, T. Längle|
|Publication status||Published - 26 Feb 2021|