Table 2 Ablation experimental results of the proposed DL-based method.
Experiment | Models | Accuracy (%) | Precision (%) | Recall (%) | Speed (FPS) |
|---|---|---|---|---|---|
1 | Baseline Model | 80.21 | 78.55 | 77.15 | 25.25 |
2 | Baseline + Model Pruning | 78.54 | 76.31 | 76.84 | 69.36 |
3 | Baseline + Model Pruning + ResCBAM Attention | 84.13 | 83.73 | 83.22 | 58.71 |
4 | Baseline + Model Pruning + Multi-Feature fusion | 85.47 | 84.94 | 83.32 | 59.87 |
5 | Baseline + Model Pruning + Online feature distillation | 88.25 | 87.65 | 86.51 | 57.81 |
6 | Full model(Proposed) | 90.05 | 89.22 | 88.94 | 57.74 |
