NeuroBand: A Wearable Multimodal Sensing System with CNN for Sleep Staging and Quality Assessment

Aiming at the problem that traditional polysomnographic sleep monitoring devices are large and expensive, this study designs an intelligent sleep monitoring system based on multimodal sensing and lightweight neural network. The system contains STM32 architecture bracelet to collect heart rate, blood oxygen, and body temperature, and a head ring to collect single-channel electroencephalogram (EEG) signal) to construct a portable real-time monitoring platform. A lightweight model fusing Residual Network (ResNet), Long Short-Term Memory Network (LSTM) and Squeeze-andExcitation Attention (SE attention) mechanism is proposed, and weighted cross-entropy loss is introduced to alleviate the category imbalance problem. On the Sleep-EDF dataset, the model achieves $86.39 \%$ accuracy, $81.95 \%$ macro-mean F1 with 0.81 Kappa coefficient. The accompanying software displays physiological metrics in real time and provides personalized assessment suggestions through sleep structure analysis. This study provides a complete technological closure for wearable sleep health management with significant application value.