D
Data
Text
Single-Epoch Sleep Stage Classification with Transformer Representation Learning
NeuroSleepNet classifies the sleep stage of a single 30-second EEG epoch using only the current segment's microevents, with no dependence on neighboring epochs. It combines convolutional spatial filtering across channels, multi-scale temporal context learning, and a transformer self-attention encoder, and counters the strong class imbalance across the five sleep stages (W, N1, N2, N3, REM) with a logarithmic-scaled weighted cross-entropy loss.