Table 3 The steps performed by our algorithm for TWFB + DGFMDM.

Step 1

Divide the time window and filter bank, where the time window is 0–1 s, 0.5–1.5 s, 1–2 s, 1.5–2.5 s, 2–3 s, 2.5–3.5 s, 3–4 s. Divide the filter band is 8–12 Hz, 9–13 Hz, 10–14 Hz, 11–15 Hz, 12–16 Hz, 13–17 Hz, 14–18 Hz, 15–19 Hz, 16–20 Hz, 17–21 Hz, 18–22 Hz, 19–23 Hz, 20–24 Hz, 21–25 Hz, 22–26 Hz, 23–27 Hz, 24–28 Hz, 25–25 Hz, 26–30 Hz.

Step 2

Select the time window and filter bank by backtracking search optimization algorithm.

Step 3

The covariance matrix is calculated in the optimal time window and frequency window to form the covariance matrix set M.

Step 4

The local tangent space arrangement algorithm is used to reduce the feature dimension of the covariance matrix in set M.

Step 5

Input the reduced dimension characteristic matrix into discriminant geometry filter, then recognition are carried out by Riemann minimum distance.