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

Abstract 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.