Sleepal AI Lamp Achieves 92.8% Accuracy in Contactless Sleep Stage Detection

A contactless sleep monitoring device called the Sleepal AI Lamp has demonstrated remarkable accuracy in sleep stage classification. Using radar-based technology, the system achieved 92.8% accuracy for binary sleep-wake classification and 78.5% accuracy for four-stage sleep classification. This performance was validated against gold-standard polysomnography (PSG) using a dataset of 1022 overnight recordings. The device extracts multi-scale respiratory and motion features from radar signals, which are processed through a frequency-augmented deep learning model. For the binary classification task, the model also achieved a macro-averaged F1 score of 0.895, while the four-stage classification (wake, light NREM, deep NREM, REM) achieved a Cohen's kappa coefficient of 0.785. This research addresses limitations of conventional PSG, which is intrusive and unsuitable for long-term monitoring despite being considered the gold standard for sleep assessment. Sleep staging remains essential for evaluating sleep quality and diagnosing sleep-related disorders. The study was published on arXiv under identifier 2604.16442v1.