Table 4 Performance comparison between state-of-the-art techniques and this work in terms of sensitivity (%), specificity (%), accuracy (%), and F1 score (%).
From: Anomaly detection in cropland monitoring using multiple view vision transformer
Methods | Sensitivity | Specificity | Accuracy | F1 score |
|---|---|---|---|---|
U-Net31 | 81.2 | 82.4 | 83.7 | 81.9 |
Mask R-CNN32 | 81.9 | 82.5 | 82.9 | 80.6 |
ExtremeNet33 | 81.7 | 82.3 | 83.6 | 82.3 |
TensorMask34 | 82.2 | 82.9 | 84.3 | 81.8 |
Visual transformer24 | 89.5 | 85.6 | 87.3 | 86.2 |
ViT17 | 88.5 | 86.4 | 87.1 | 87.1 |
MViT19 | 87.9 | 87.5 | 88.1 | 87.9 |
PiT21 | 87.6 | 88.2 | 89.4 | 88.3 |
PVT20 | 89.3 | 89.2 | 90.1 | 89.5 |
UViT22 | 88.6 | 89.7 | 91.5 | 90.3 |
Swin transformer18 | 89.1 | 87.2 | 88.0 | 87.4 |
Proposed work (10-fold) | 92.8 | 93.1 | 93.5 | 94.1 |
Proposed work (15-fold) | 91.7 | 92.5 | 92.9 | 93.4 |
Proposed work (20-fold) | 90.4 | 91.6 | 92.3 | 92.8 |
