Figure 5.
The diagram presents a convolutional neural network structure from left to right. The input is 64, 64, 3. Convolution 1, Convolution 2, Convolution 3, Convolution 4, Convolution 5, and Convolution 6 each use 64 filters, a 2 by 2 filter, and rectified linear unit activation. Batch normalisation follows Convolution 1, Convolution 2, Convolution 3, Convolution 4, Convolution 5, and Convolution 6 as Normalisation 1 to Normalisation 6. Pooling 1 follows Convolution 3, and Pooling 2 follows Convolution 6. Both pooling layers use 2 by 2 size. The flatten layer follows Normalisation 6. Dropout 1 has a dropout rate of 0.3. Fully Connected 1 has 2048 neurons and rectified linear unit activation. Dropout 2 has a dropout rate of 0.3. Fully Connected 2 has 1331 neurons and Softmax activation.Structure of convolutional neural network
Source: Authors’ own work