A Holistic Eating Monitoring System Enabled by Textile Strain Sensors and Hierarchical Network

Wearable sensors have prompted human dynamic monitoring by quantifying physiological and biomechanical signals in daily life. However, existing approaches lack an integrated solution that supports continuous and fine-grained measurement during eating behavior. In this work, we propose a holistic eating behavior monitoring system that combines graphene-based textile strain sensors with a hierarchical deep learning framework. The proposed wearable sensor enables simultaneous and comfort able acquisition of throat and jaw deformation signals, while the hierarchical network jointly models eating-related activities and biomechanical attributes from raw strain data. Furthermore, model visualization analyses, including activation maps and event mask dynamics, reveal clear correspondence with underlying physiological processes, supporting the explainability of the proposed framework. To facilitate user interpretation and behavioral insight, a large language model coaching module converts the extracted biomechanical indicators into personalized natural language feedback. This holistic sensing and learning system establishes a foundation for personalized healthy dietary guidance, enabling dynamic monitoring for digestive health, metabolic regulation, and nutrition management.