Spatial-Frequency Feature Learning and Classification of Motor Imagery EEG Based on Deep Convolution Neural Network

Table 3

CNN model architecture for 10-fold cross-validation (BCI competition III dataset IVa). Conv refers to convolution layer, Flatten refers to flatten layer, and FC refers to fully connected layer.

	Kernel size	Kernel number	Padding	Activation	Output shape
Conv_1	49 × 1	6	Valid	Elu	1 × 6 × 6
Conv_2	1 × 3	12	Valid	Elu	1 × 2 × 12
Flatten	—	—	—	—	24
FC_1	—	—	—	Relu	50
FC_2	—	—	—	Relu	100
FC_3	—	—	—	Relu	200
Softmax	—	—	—	—	2