Using microwave metrology to count calories

• We propose a nearest neighbor estimator to estimate the calories of food stuffs.
• Homogeneous mixtures of oil, water, sugar are investigated.
• The estimator is based on a novel distance metric for magnitude spectra.
• Performance is compared to an oracle estimator.
• The estimator achieves an average absolute error of 10%.

One non-destructive way to probe the caloric content of food is to examine the transmission spectra of low energy microwaves over a broad band (approximately 1–8 GHz). This paper proposes a simple, but generalizable, nearest neighbor scheme to estimate the calories of homogeneous mixtures of oil, sugar, and water. The approach’s performance is empirically quantified and is also compared to the performance of an oracle estimator. We report that for this study the estimator achieves an average absolute error of approximately 10%. A heuristic extension of the nearest neighbor estimator is also discussed that can in some cases improve the average error performance. The work represents the first steps toward accurate and reliable calorie estimation techniques for complex, non-homogeneous foods of varying shapes and amounts.