From: Identification of citrus diseases based on AMSR and MF-RANet
Network | Normal citrus (%) | Huanglong disease (%) | Corynespora blight of citrus (%) | Fat spot macular disease (%) | Citrus scab (%) | Citrus canker (%) | Average testing accuracy (%) | Training time | Model size (MB) |
---|---|---|---|---|---|---|---|---|---|
ResNet50 | 86.24 | 84.26 | 86.65 | 86.27 | 83.70 | 83.24 | 85.06 | 5:48ʹ12ʺ | 26 M |
ResNet50(E) | 90.76 | 89.67 | 88.19 | 87.40 | 87.65 | 86.37 | 88.34 | 4:53ʹ40ʺ | 26 M |
ResNet50 + L | 89.47 | 89.39 | 89.02 | 86.57 | 86.31 | 87.48 | 88.04 | 4:48ʹ33ʺ | 26 M |
ResNet50 + R | 93.21 | 92.10 | 92.76 | 91.83 | 90.96 | 91.00 | 91.82 | 5:31ʹ10ʺ | 25.2 M |
ResNet50 + A | 92.99 | 92.04 | 92.71 | 92.26 | 90.29 | 91.65 | 91.99 | 5:29 ʹ04ʺ | 24.8 M |
MF-RANet | 97.55 | 96.77 | 96.22 | 95.96 | 93.93 | 94.04 | 96.87 | 3:47ʹ46ʺ | 24 M |