FIGURE 1.

The flowchart of (a) the proposed ST‐GCN framework and (b) the T‐CNN framework for comparison. $X$ is sized M*N (M = 25, representing the length of the mini‐epoch channel; N = 23, representing the 23 channels) and the input $W$ is sized N*N. The (i,j)th entry of the adjacency matrix $W$ denotes spatial coupling correlation strength between the ith and jth of all the 23 different channels, the detailed calculation of which is presented in section 2.2. $K_t$ is the size of the temporal 1D‐Conv filter, set as 3 here. $C_l\left(l=1,2,\dots ,6\right)$ is the number of filters in each layer. The estimated computational complexity of ST‐GCN and T‐CNN are 54M and 52M, respectively