Table 10 Performance comparison of different optimizers and learning rate configurations on standard images and occluded images.
S. No. | Model Name | Ablation components | Standard bird objects | Occluded bird objects | |||
|---|---|---|---|---|---|---|---|
Optimizer | Learning Rate | A (%) | FS (%) | A (%) | FS (%) | ||
1 | DenseNet121 | ADAM | 1e-3 | 90.38 | 90.39 | 81.01 | 77.37 |
2 | DenseNet121 | ADAM | 1e-4 | 94.8 | 94.82 | 89.24 | 90.01 |
3 | DenseNet121 | ADAM | 1e-5 | 93.3 | 93.41 | 86.08 | 83.54 |
4 | DenseNet121 | RMSProp | 1e-3 | 75.35 | 76.58 | 60.57 | 60.24 |
5 | DenseNet121 | RMSProp | 1e-4 | 81.25 | 81.47 | 70.89 | 69.78 |
6 | DenseNet121 | RMSProp | 1e-5 | 95.07 | 95.13 | 91.77 | 91.17 |
7 | DenseNet169 | ADAM | 1e-3 | 88.61 | 88.86 | 79.75 | 78.47 |
8 | DenseNet169 | ADAM | 1e-4 | 94.65 | 94.71 | 89.24 | 88.75 |
9 | DenseNet169 | ADAM | 1e-5 | 93.49 | 93.52 | 87.97 | 87.56 |
10 | DenseNet169 | RMSProp | 1e-3 | 75.08 | 75.86 | 60.87 | 59.66 |
11 | DenseNet169 | RMSProp | 1e-4 | 81.52 | 81.78 | 70.89 | 71.77 |
12 | DenseNet169 | RMSProp | 1e-5 | 94.96 | 94.94 | 92.14 | 90.16 |
13 | ResNet101V2 | ADAM | 1e-3 | 89.92 | 90.01 | 83.54 | 83.57 |
14 | ResNet101V2 | ADAM | 1e-4 | 95.65 | 95.7 | 91.41 | 91.59 |
15 | ResNet101V2 | ADAM | 1e-5 | 95.91 | 95.62 | 91.77 | 91.85 |
16 | ResNet101V2 | RMSProp | 1e-3 | 74.58 | 75.91 | 61.87 | 61.42 |
17 | ResNet101V2 | RMSProp | 1e-4 | 82.41 | 82.66 | 62.66 | 63.07 |
18 | ResNet101V2 | RMSProp | 1e-5 | 96.04 | 96.02 | 91.41 | 91.59 |
