A review of chewing detection for automated dietary monitoring

A healthy dietary lifestyle prevents diseases and leads to good physical conditions. Poor dietary habits, such as eating disorders, emotional eating and excessive unhealthy food consumption, may cause health complications. People's eating habits are monitored through automated dietary monitoring (ADM), which is considered a part of our daily life. In this study, the Google Scholar database from the last 5 years was considered. Articles that reported chewing activity characteristics and various wearable sensors used to detect chewing activities automatically were reviewed. Key challenges, including chew count, various food types, food classification and a large number of samples, were identified for further chewing data analysis. The chewing signal's highest reported classification accuracy value was 99.85%, which was obtained using a piezoelectric contactless sensor and multistage linear SVM with a decision tree classifier. The decision tree approach was more robust and its classification accuracy (75%-93.3%) was higher than those of the Viterbi algorithm-based finite-state grammar approach, which yielded 26%-97% classification accuracy. This review served as a comparative study and basis for developing efficient ADM systems.