Sleep Posture Transition Recognition Using Millimeter-Wave Radar and Deep Learning: A SleepNet Approach

Although With the increasing concern about the impact of sleep quality on health, inappropriate sleep postures may lead to a variety of health problems such as snoring, sleep apnea, and cervical spine discomfort, thus affecting the overall sleep quality. To address this issue, this study proposes a millimeter-wave radar-based sleep posture transition recognition method. Millimeter-wave radar was chosen for its non-invasiveness, high resolution, penetration into the environment, and privacy-preserving advantages, which make it a unique potential for application in sleep monitoring. By designing the radar's two-dimensional feature mapping channel, generating distance-time thermograms and Doppler-time thermograms, and applying a new deep learning model, SleepNet, the accuracy of gesture transition recognition was significantly improved. After testing 1400 samples from 20 volunteers, the recognition accuracy of the SleepNet model reached 90.39%. Compared with traditional methods, this method has the advantages of non-invasiveness, privacy-friendliness, cost-effectiveness and high accuracy, providing an effective solution for sleep quality monitoring.