Research Article
Identifying Capsule Defect Based on an Improved Convolutional Neural Network
Table 2
Details of the proposed CNN model used in the experiments.
| Layer | Kernel size | Stride | Padding | Output size | Output number |
| Convolution 1 | 55 | 1 | Yes | 100100 | 32 | Pooling 1 | 22 | 2 | No | 5050 | 32 | Convolution 2 | 55 | 1 | Yes | 5050 | 64 | Pooling 2 | 22 | 2 | No | 2525 | 64 | Convolution 3 | 33 | 1 | Yes | 2525 | 128 | Pooling 3 | 22 | 2 | No | 1212 | 128 | Convolution 4 | 33 | 1 | Yes | 1212 | 128 | Pooling 4 | 22 | 2 | No | 66 | 128 | Fully connected 1 | | | | 11 | 1024 | Fully connected 2 | | | | 11 | 512 | Fully connected 3 | | | | 11 | 12 |
|
|