MsuSleepNet: Automatic Sleep Staging Scoring Based on U-net and Multi-Scale Features

Automated sleep staging is critical for the assessment and diagnosis of sleep disorders. Although previous attempts to classify sleep stages have achieved high classification performance, they still face several problems: 1) How to effectively extract deep-level salient wave features in single-channel sleep data; 2) How to portray the characteristic representation of different time and frequency scales between sleep stages and the multi-scale transformation rules. To address these challenges, we propose an automatic sleep staging model named MsuSleepNet. This model is a time-series fully convolutional network based on the U-net architecture. It combines multi-scale convolution operations and channel attention mechanisms to expand the receptive field, and can consider multiple feature sequences on different time and frequency scales. We evaluated on publicly available standard datasets and achieved good sleep staging results.