Algorithm 1 The proposed character recognition algorithm

Input: EEG Data, X with the size of (N × 150 × 30)

Output: Predict Result C

Initialize: i ← 0, P(N) ← 0

for each i ∈ [1, N] do

Normalized Data: Ii ← BN(Xi)

Extract spatial feature, filter size of(1 × 30): Cli ← ReLU(BN((Ii · Ws + bs))

Pooling, apply with stride(2 × 1): Mi ← maxpooling (C1i)

Extract temporal feature, filter size of(20 × 1): C2i ← ReLU(BN((Mi · Ws + bs))

Fully connected:

F1i ← fully connected(C2i), Fli ← Softmax(Fli)

F2i ← fully connected(F2i)

P(i) ← F2i[1]

end for

C ← max(P)