Hybrid Deep Learning Approaches for sEMG Signal-Based Lower Limb Activity Recognition

Table 1

Parameters considered in the proposed study using deep learning models.

Model parameters		CNN	LSTM	GRU	CNN-LSTM	CNN-GRU
Convolution layer 1	Number of filters	16	—	—	16	16
	Kernel size	3	—	—	3	3
Convolution layer 1	Number of filters	8	—	—	8	8
	Kernel size	3	—	—	3	3
Pooling layer	Type of pooling	Max	—	—	Max	Max
	Kernel size	2	—	—	2	2
LSTM	Unit size	—	16	—	16	—
LSTM	Unit size	—	8	—	8	—
GRU	Unit size	—	—	16	—	16
GRU	Unit size	—	—	8	—	8
Fully connected	Unit size	16	16	16	16	16
Fully connected	Unit size	3	3	3	3	3