Table 3.
Comparison of improving module performance.
| Model | Average Accuracy/% | Average Precision/% | Average Recall/% | F1/% | Detection time/s | |
|---|---|---|---|---|---|---|
| 1 | Inception-Resnet | 88.10 | 90.26 | 88.10 | 89.17 | 2.59 |
| 2 | Inception-Resnet (+depthwise separable convolution) | 89.68 | 91.33 | 89.67 | 90.49 | 2.56 |
| 3 | Inception-Resnet (+ECANet) | 95.96 | 92.46 | 91.26 | 91.86 | 2.68 |
| 4 | Inception-Resnet (+feature fusion) | 93.65 | 94.28 | 93.65 | 93.96 | 2.62 |
| 5 | Inception-Resnet (+depthwise separable convolution + ECANet) | 92.06 | 93.05 | 92.06 | 92.55 | 2.77 |
| 6 | Improved Inception-ResNet | 96.03 | 96.27 | 96.03 | 96.15 | 2.44 |