A Neural Network-Based Optimal Spatial Filter Design Method for Motor Imagery Classification
Fig 5
Input data and SFN output data for toy data with 4-classes.
(a) log-variance feature for 2-dimensional input data. Note that each point represents an epoch that belongs to class 1 (red circle), class 2 (blue plus), class 3 (green asterisk) or class 4 (yellow cross). (b) Enclosing ellipses represent the input data. (c) SFN spatial filter layer output (f) with generated class borders (black dashed lines) of the classifier layer. (d) Enclosing ellipses represent the spatially filtered input data (y).