SS-SWT and SI-CNN: An Atrial Fibrillation Detection Framework for Time-Frequency ECG Signal

Table 1

The detailed architecture of the proposed SI-CNN.

CNN parameters	Volume
1st convolutional layer kernel size	1 × 3 × 32
2nd convolutional layer kernel size	1 × 3 × 32
1st max-pooling layer kernel size	1 × 3 × 32
1st batch normalization layer	—
1st dropout layer rate	0.25
3rd convolutional layer kernel size	1 × 3 × 32
4th convolutional layer kernel size	1 × 3 × 32
2nd max-pooling layer kernel size	1 × 3 × 32
2nd batch normalization layer	—
2nd dropout layer rate	0.25
5th convolutional layer kernel size	1 × 3 × 64
6th convolutional layer kernel size	1 × 3 × 64
3rd max-pooling layer kernel size	1 × 3 × 64
3rd batch normalization layer	—
3rd dropout layer rate	0.25
7th convolutional layer kernel size	1 × 3 × 64
8th convolutional layer kernel size	1 × 3 × 64
4th max-pooling layer kernel size	1 × 3 × 64
4th batch normalization layer	—
4th dropout layer rate	0.25
9th convolutional layer kernel size	1 × 3 × 128
10th convolutional layer kernel size	1 × 3 × 128
5th max-pooling layer kernel size	1 × 3 × 128
5th batch normalization layer	—
5th dropout layer rate	0.25
Global average pooling layer	—
6th dropout layer rate	0.25
The number of neurons in the fully connected layer	128
7th dropout layer rate	0.25
The number of neurons in the softmax layer	2