Figure 3.

Once the signals are acquired, the recordings are segmented into four-second intervals to generate the samples for each class. Samples are prepared as described in Section 2.2 and Section 2.3, with signal transformation to spectrogram images, data dimension re-structuring, and channel downsampling. Finally, the samples are used as input for the proposed models described in Section 2.4 and Section 2.5 for the 3-class classification of MI for rest, left-, and right-hand movement.