Table 4.
Average testing accuracy of comparative methods with multi-sensory data.
| Fusion Level | Model | Feature Learning from Raw Data | Manual Feature Extraction | ||
|---|---|---|---|---|---|
| Time-Domain Features | Frequency-Domain Features | Handcraft Features | |||
| Data-level fusion | DCNN | 99.28% | 66.08% | 87.63% | 90.23% |
| BPNN | 53.28% | 65.95% | 87.89% | 91.22% | |
| SVM | 51.62% | 67.32% | 87.28% | 90.67% | |
| Feature-level fusion | DCNN | 98.75% | 86.35% | 92.34% | 94.08% |
| BPNN | 64.74% | 86.81% | 92.15% | 94.04% | |
| SVM | 56.27% | 86.74% | 94.62% | 95.80% | |
| Decision-level fusion | DCNN | 93.65% | 84.65% | 90.23% | 92.19% |
| BPNN | 77.62% | 84.47% | 91.19% | 93.42% | |
| SVM | 76.17% | 86.32% | 90.98% | 93.44% | |