Table 3 The convergence comparison (variance of accuracies (e-5)) of the proposed OFBCNN with the nonoverlapping filter bank CNN and original CNN in multisubject MI recognition
CNN-Dataset-Low-Cut Frequency | Original | NOFBCNN | Sliding OFBCNN | Fixed OFBCNN |
|---|---|---|---|---|
EEGNet-2a-0Â Hz | 9.24 | 36.67 | 13.33 | 3.45 |
Shallow-2a-0Â Hz | 15.32 | 10.00 | 8.77 | 3.42 |
Deep-2a-0Â Hz | 12.11 | 16.67 | 10.00 | 3.53 |
EEGNet-2a-4Â Hz | 23.55 | 23.33 | 16.67 | 13.33 |
Shallow-2a-4Â Hz | 14.82 | 9.33 | 7.74 | 5.81 |
Deep-2a-4Â Hz | 10.85 | 13.15 | 8.67 | 5.00 |
EEGNet-HG-0Â Hz | 7.94 | 40.00 | 25.00 | 4.28 |
Shallow-HG-0Â Hz | 3.90 | 7.71 | 4.35 | 2.25 |
Deep-HG-0Â Hz | 13.33 | 85.00 | 35.00 | 3.88 |
EEGNet-HG-4Â Hz | 8.30 | 25.00 | 25.00 | 20.00 |
Shallow-HG-4Â Hz | 7.05 | 9.13 | 5.79 | 4.98 |
Deep-HG-4Â Hz | 60.00 | 105.00 | 45.00 | 20.00 |
Average | 15.53 | 31.75 | 17.11 | 7.49 |