Automatic Seizure Detection Using Multi-Input Deep Feature Learning Networks for EEG Signals

Table 3

Layers of the designed network.

	Time domain		TF domain
	Layer info	Output size	Layer info	Output size
Input	—	(1,280, 23)	—	(65, 21, 23)
C11/C21	Conv1D(3) × 16	(1,278, 16)	Conv2D(3, 1) × 16	(63, 21, 16)
	ReLU	(1,278, 16)	ReLU	(63, 21, 16)
	BatchNor	(1,278, 16)	BatchNor	(63, 21, 16)
	MaxP, s:2	(639, 16)	MaxP(2, 1), s:2	(31, 21, 16)
C12/C22	Conv1D(3) × 32	(637, 32)	Conv2D(3, 1) × 32	(29, 21, 32)
	ReLU	(637, 32)	ReLU	(29, 21, 32)
	BatchNor	(637, 32)	BatchNor	(29, 21, 32)
	MaxP, s:2	(318, 32)	MaxP(2, 1), s:2	(14, 21, 32)
C13/C23	Conv1D(3) × 64	(316, 64)	Conv2D(3, 1) × 64	(12, 21, 64)
	ReLU	(316, 64)	ReLU	(12, 21, 64)
	BatchNor	(316, 64)	BatchNor	(12, 21, 64)
	MaxP, s:2	(158, 64)	MaxP(2, 1), s:2	(6, 21, 64)
	—	—	Reshape	(126, 64)
Concatenate	(284, 64)
BLSTM 1	(284, 40)
BLSTM 2	(284, 40)
Flatten	11,360
Dropout	11,360
Dense	128
Dense	1