Predicting food nutrition facts using pocket-size near-infrared sensor

Diet monitoring is one of the most important aspects in preventative health care that aims to reduce various health risks. Manual recording has been a prevalence among all approaches yet it is tedious and often end up with a low adherence rate. Several existing techniques that have been developed to monitor food intake suffer too with accuracy, efficiency, and user acceptance rate. In this paper we propose a novel approach on measuring food nutrition facts, through a pocket-size non-intrusive near-infrared (NIR) scanner. We build efficient regression models that can make quantitative prediction on food nutrition contents, such as energy and carbohydrate. Our extensive experiments on off-the-shelf liquid foods demonstrates the accuracy of these regression models and proves the applicability of using NIR spectra that are collected by small hand-held scanner, on food nutrition prediction.