Development of Deep Learning Model for the Recognition of Cracks on Concrete Surfaces

Table 3

Details of CNN’s architecture.

N°	Layer	Size of activation (i.e., outputs of layer)	Size of weight parameters	Size of bias parameters
1	Input layer	227 × 227 × 3	—	—
2	Convolutional layer	227 × 227 × 16	3 × 3 × 3 × 16	1 × 1 × 16
3	ReLU layer 1	227 × 227 × 16	—	—
4	Fully connected layer 1	1 × 1 × 200	200 × 824464	200 × 1
5	Fully connected layer 2	1 × 1 × 200	200 × 200	200 × 1
6	Batch normalization layer	1 × 1 × 200	1 × 1 × 200	1 × 1 × 200
7	ReLU layer 2	1 × 1 × 200	—	—
8	Fully connected layer 3	1 × 1 × 2	2 × 200	2 × 1
9	Softmax layer	1 × 1 × 2	—	—
10	Classification output layer	—	—	—