SE-Res-U-Net: an improved U-Net architecture for efficient sleep state detection and classification
Ghulam Irtaza1,2, Naila Sammar Naz1, Muhammad Usman Saeed3
1School of Computer Science, National College of Business Administration and Economics, Lahore, 54000, Pakistan.
Scientific Reports
|November 10, 2025
Summary
This study introduces an improved 1D U-Net model for accurate sleep state classification using physiological signals. The model achieves high accuracy, offering a scalable solution for clinical and home-based sleep analysis.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence in Healthcare
- Sleep Medicine
Background:
- Accurate sleep state detection is crucial for diagnosing sleep disorders.
- Traditional methods like polysomnography are resource-intensive and not scalable.
- There is a need for efficient and automated sleep analysis tools.
Purpose of the Study:
- To develop an improved 1D U-Net model for efficient sleep state detection and classification.
- To enhance feature extraction and representation for physiological signals.
- To provide a scalable solution for sleep analysis in clinical and home settings.
Main Methods:
- An improved 1D U-Net architecture incorporating residual blocks in the encoder for deep feature extraction.
- Squeeze and excitation blocks with attention mechanisms in the decoder to enhance channel-wise features.
- Processing of 1D physiological data, including electroencephalogram (EEG) signals.
Main Results:
- Achieved up to 94% accuracy on benchmark datasets (Sleep-EDF-20, Sleep-EDF-78, SHHS).
- Obtained a mean F1 score of 87.1% and Cohen's kappa of 0.84.
- Demonstrated reduced computational overhead compared to baseline models.
Conclusions:
- The proposed model offers an effective and efficient approach for sleep state classification.
- The model's performance and reduced computational cost make it suitable for scalable applications.
- This technology has practical significance for both clinical diagnostics and remote/home-based sleep monitoring.
