Table 5.
Comparison of accuracy performance with existing studies.
| Author (Year) |
Np | Ns | Nb | Radar Hardware | Best Model | Accuracy |
|---|---|---|---|---|---|---|
| Zhou, et al. [56] | 3 | 8 | 0 | FMCW radar system | CNN w/Inception-Residual module | 87.2% |
| Piriyajitakonkij, et al. [40] | 38 | 4 | 0 | Xethru X4M03 | SleepPoseNet: a Deep CNN w/MW Learning | 73.7 ± 0.8% |
| Islam and Lubecke [57] | 20 | 3 | 0 | Dual-frequency monostatic CW radar | Decision Tree | Dual: 98.4% |
| Adhikari and Sur [58] | 8 | 5 | 0 | Texas Instrument IWR1443 | Rest Network, a customized Deep Convolutional Neural Network | 95.6% |
| Lai, et al. [44] | 30 | 4 | 1 | Xethru X4M03 | Swin Transformer | 80.8% |
| This study | 70 | 4 | 3 | Xethru X4M03 | MWCNN w/DenseNet121 | 80.9% |
CNN: convolutional neural network; MW: Multiview; Nb: Number of blanket conditions; Np: Number of participants; Ns: Number of sleep postures to be classified; w/: with.