Table 2 Detailed design of CWT-DSCNN-CBAM.

From: Enhanced analog circuit fault diagnosis via continuous wavelet transform and dual-stream convolutional fusion

Detailed design

Layer Type

Filter/Neuron count

Kernel size/Stride

Activation function

Output shape

1DCNN

Input layer

   

(L, 2)

Convolutional layer 1

64

3/1

ReLU

(L-2, 64)

Pooling layer 1

 

2/2

 

(L/2 -1, 64)

Convolutional layer 2

128

3/1

ReLU

(L/2 -4, 128)

Pooling layer 2

 

2/2

 

(L/4 -3, 128)

2DCNN

Input layer

   

(224,224,3)

Convolutional layer 1

64

3 × 3/1

ReLU

(224,224,64)

Pooling layer 1

 

2 × 2/2

 

(112,112,64)

Convolutional layer 2

128

3 × 3/1

ReLU

(112,112,128)

Pooling layer 2

 

4 × 4/4

 

(28,28,128)

Flatten layer

   

(784,128)

Dual-stream fusion

Fusion layer

   

(L/4 + 781, 128)

Flatten layer

   

((L/4 + 781) × 128,)

Fully connected layer 1

120

  

(120,)

Fully connected layer 2

84

  

(84,)

Output layer

N

 

Softmax

(N,)

Back to article page