Deep Residual Networks for Sleep Posture Recognition With Unobtrusive Miniature Scale Smart Mat System

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Sleep Science

Background:

• Sleep posture is a key indicator of sleep quality.
• Accurate sleep posture monitoring is challenging with existing methods.
• Unobtrusive sensing solutions are needed for long-term sleep analysis.

Purpose of the Study:

• To develop and validate an unobtrusive smart mat system for recognizing sleep posture.
• To assess the system's accuracy and reliability in classifying different sleep positions.
• To explore the system's potential applications in sleep studies and healthcare.

Main Methods:

• A smart mat system utilizing a dense flexible sensor array and printed electrodes.
• An algorithmic framework incorporating preprocessing and Deep Residual Networks (ResNet) for posture classification.
• System evaluation on seventeen subjects for short-term and overnight sleep studies.

Main Results:

• The system achieved up to 95.08% accuracy in short-term tests and 86.35% accuracy in overnight studies.
• Classification included four categories: supine, prone, right, and left sleep postures.
• The developed ResNet model eliminated the need for complex feature extraction, improving performance.

Conclusions:

• The proposed smart mat system offers a comfortable, high-resolution, and cost-effective solution for sleep posture monitoring.
• The system demonstrates high accuracy and reliability, outperforming most state-of-the-art methods.
• The technology shows significant potential for applications in sleep research, pressure ulcer prevention, and other healthcare areas.