Table 6.
Quantitative evaluation of different spatial attention methods. The "Attention" column refers to the attention mechanism used, and the "Models" column indicates ResNeXt50 with the respective spatial attention module inserted.
| Dataset | Attention | Models | Accuracy (%) | Recall (%) | Kappa (%) | Precision (%) | F1-Score (%) |
|---|---|---|---|---|---|---|---|
| TrashIVL-5 | CBAM [26] | ResNeXt50+SAM | 96.12 ± 0.15 | 95.62 ± 0.17 | 94.96 ± 0.19 | 95.63 ± 0.20 | 95.66 ± 0.25 |
| NAM [12] | ResNeXt50+SAM | 96.28 ± 0.26 | 95.76 ± 0.34 | 95.78 ± 0.33 | 95.83 ± 0.32 | 95.17 ± 0.33 | |
| IPAM | ResNeXt50+SAM | 96.31 ± 0.21 | 95.74 ± 0.27 | 95.21 ± 0.27 | 95.83 ± 0.26 | 95.96 ± 0.24 | |
| TrashIVL-12 | CBAM [26] | ResNeXt50+SAM | 93.71 ± 0.26 | 93.00 ± 0.38 | 92.91 ± 0.29 | 92.71 ± 0.11 | 92.81 ± 0.23 |
| NAM [12] | ResNeXt50+SAM | 93.98 ± 0.30 | 93.44 ± 0.46 | 93.21 ± 0.34 | 93.28 ± 0.41 | 93.22 ± 0.41 | |
| IPAM | ResNeXt50+SAM | 94.01 ± 0.11 | 93.53 ± 0.37 | 93.24 ± 0.11 | 93.40 ± 0.07 | 93.22 ± 0.12 |