Table 3.
Main hyperparameter settings of rice grain detection model and occlusion restoration model.
| Model | Hyperparameter | Setting |
|---|---|---|
| Rice grain detection model based on Faster R-CNN | Img_size | 1425 x 1700 pixels |
| NMS threshold | 0.74 | |
| Batch size | 1 | |
| Optimizer | SGD | |
| Learning rate | 0.02 | |
| Momentum | 0.9 | |
| Weight decay | 0.0001 | |
| RPN proposal number | 4000 | |
| Maximum epoch number | 100 | |
| Rice grain occlusion restoration model based on improved Pix2Pix | Img_size | 256x256 pixels |
| Batch size | 64 | |
| Optimizer | Adam | |
| Learning rate | 0.0002 | |
| Maximum epoch number | 500 |