Metabolic rate estimation method using image deep learning

Thermal comfort is an important factor in evaluating indoor environmental quality. However, accurately evaluating thermal comfort conditions is challenging owing to the lack of suitable methods for measuring individual factors such as the metabolic rate (Mvalue). In this study, aMvalue evaluation method was developed using deep learning. The metabolic equivalent of task was measured for eight typical indoor tasks based on the ASHRAE Standard 55 (lying down, sitting, cooking, walking, eating, house cleaning, folding clothes, and handling 50 kg books) in 31 subjects (males: 16; and females: 15); the measurements were analyzed in terms of gender and body mass index (BMI). The experimental results were assessed using the reliability of the measured data, theMvalue difference in terms of gender and BMI, and the measurement accuracy. We developed aMvalue self-evaluation model using artificial intelligence, which achieved an average coefficient of variation (CV) of 12%. A third-party evaluation model was used to evaluate theMvalue of one subject based on the learning data acquired from the other 30 subjects; this model yielded a low CV of 54%. For high-activity tasks, males generally had higherMvalues than females, and the higher the BMI was, the higher was theMvalue. Contrarily, for low-activity tasks, the lower the BMI was, the higher was theMvalue. The breakthroughMvalue evaluation method presented herein is expected to improve thermal comfort control.