Heart Signal Analysis Using Multistage Classification Denoising Model

Table 2

Architecture of each layer used in the proposed multistage model.

Layer	Learnable properties	Number of learnable
PCG input 1 × 2942 × 1	—	0
Convolution-1	Weight 1 × 33 × 1 × 8	272
32 filters 1 × 33, stride (1, 1)	Bias 1 × 1 × 8
Maxpooling: (2 × 2), stride Convolution-2	Weight 1 × 21 × 8 × 4	676
16 filters 1 × 13, stride (1, 1)	Bias 1 × 1 × 14
Maxpooling: (2 × 2), stride (1, 1)
Convolution-3	Weight 1 × 11 × 4 × 4	180
16 filters 1 × 13, stride (1, 1)	Bias 1 × 1 × 4
Maxpooling: (2 × 2), stride (1, 1)
Flatten		0
LSTM layer (64 units)	Input (192 × 11768)	2271936
	Recurrent weight (192 × 64)
	Bias 192 × 1
Fully connected layer (5 classes)	Weights 5 × 64	325
	Bias 5 × 1