Research Article
Deep-Learning-Based Bughole Detection for Concrete Surface Image
Table 2
Dimensions of layers and operations.
| Type | Patch size/stride | Output size | Depth | #1 × 1 | #3 × 3 | #5 × 5 | Pool proj. | Params |
| Input | | 28 × 28 × 3 | 0 | | | | | | Convolution | 5 × 5/2 | 24 × 24 × 20 | 1 | | | | | 1.5 K | Batch norm | | 24 × 24 × 20 | 0 | | | | | | Max. pool | 3 × 3/2 | 12 × 12 × 20 | 0 | | | | | | Inception (2) | | 12 × 12 × 64 | 2 | 16 | 32 | 8 | 8 | 10.3 K | Batch norm | | 12 × 12 × 64 | 0 | | | | | | Max. Po1ol | 2 × 2/2 | 6 × 6 × 64 | 0 | | | | | | Inception (3) | | 6 × 6 × 512 | 2 | 192 | 208 | 48 | 64 | 214 K | Batch norm | | 6 × 6 × 512 | 0 | | | | | | Avg pool | 6 × 6/1 | 1 × 1 × 512 | 0 | | | | | | Convolution | 1 × 1/1 | 1 × 1 × 10 | 1 | | | | | 5.1 K | Softmax | | 1 × 1 × 1 | 0 | | | | | 0.01 K |
|
|