A Deep-Learning-Assisted On-Mask Sensor Network for Adaptive Respiratory Monitoring

Wearable respiratory monitoring is a fast, non-invasive, and convenient approach to provide early recognition of human health abnormalities like restrictive and obstructive lung diseases. Here, a computational fluid dynamics assisted on-mask sensor network is reported, which can overcome different user facial contours and environmental interferences to collect highly accurate respiratory signals. Inspired by cribellate silk, Rayleigh-instability-induced spindle-knot fibers are knitted for the fabrication of permeable and moisture-proof textile triboelectric sensors that hold a decent signal-to-noise ratio of 51.2 dB, a response time of 0.28 s, and a sensitivity of 0.46 V kPa(-1). With the assistance of deep learning, the on-mask sensor network can realize the respiration pattern recognition with a classification accuracy up to 100%, showing great improvement over a single respiratory sensor. Additionally, a customized user-friendly cellphone application is developed to connect the processed respiratory signals for real-time data-driven diagnosis and one-click health data sharing with the clinicians. The deep-learning-assisted on-mask sensor network opens a new avenue for personalized respiration management in the era of the Internet of Things.