Track Circuits Fault Diagnosis Method Based on the UNet-LSTM Network (ULN)

Table 3

LSTM model parameters of the applied method.

Name	Filter	Kernel size/stride/padding	Unit	Input size	Output size
Downsample
Conv-1	4	3/2/1		1 × 600	4 × 300
Conv-2	8	3/2/1		4 × 300	8 × 150
Conv-3	16	3/2/1		8 × 150	16 × 75
Feature extraction block
Conv-4	16	3/1/1		8 × 150	16 × 150
Conv-5	32	5/1/2		16 × 150	32 × 150
Conv-6	64	7/1/3		32 × 150	64 × 150
Cat				112 × 150	8 × 150
Upsample
ConvTranspose1d-1	8	1/1/0		16 × 75	8 × 150
ConvTranspose1d-2	4	1/1/0		16 × 150	4 × 300
Conv-7	1	3/8/0		8 × 300	1 × 38
LSTM
LSTM_1			16	1 × 38	16 × 38
FC			7	1 × 38	1 × 7