Table 5 Comparison of the proposed hybrid architecture with original EFS-Net, modified EFS-Net, original Swin-T (Swin-UperNet), and simplified Swin-T in complexity, speed, and recognition.
Criteria | Different components of the proposed architecture | ||||
|---|---|---|---|---|---|
Original EFS-Net | Modified EFS-Net | Original Swin-T (Swin-UperNet) | Simplified Swin-T | Proposed architecture | |
Mean IoU | 77.12% | 79.92% | 86.71% | 81.81% | 98.44% |
Mean accuracy | 79.4% | 83.12% | 89.5% | 87.02% | 99.5% |
Mean precision | 78.83% | 80.10% | 88.43% | 82.11% | 98.98% |
Mean recall | 78.01% | 81.01% | 87.61% | 82.71% | 98.52% |
Mean specificity | 78.68% | 81.98% | 88.19% | 83.96% | 99.1% |
Mean F-score | 78.41% | 80.05% | 88.01% | 82.40% | 98.74% |
Mean F-boundary | 71.23% | 79.99% | 78.2% | 69.15% | 82.01% |
Mean testing time (s) | 0.010 | 0.009 | 0.021 | 0.0105 | 0.011 |
GFLOPs | 2.2 | 2.9 | 4.5 | 3.1 | 6.2 |
Parameters no. (M) | 3.27 | 4.68 | 31 | 21.5 | 25 |
FPS | 100 | 111 | 48 | 95 | 91 |
Memory-usage (GB) | 4.1 | 4.8 | 32 | 21.5 | 20 |
