Table 5 Comparative analysis of our proposed methods and SOTA approaches.
From: Multi-scale convolutional transformer network for motor imagery brain-computer interface
Methods | Preprocessing | Augmentation | Architecture | Parameters | Accuracy % | ||
|---|---|---|---|---|---|---|---|
2a | 2b | 2a | 2b | ||||
Shallow ConvNet 201720+ | STD | S & R | SSCNN | 46.1 k | 10.8 k | 75.15 | 85.30 |
Deep ConvNet 201720+ | STD | S & R | SSCNN | 283.3 k | 268.6 k | 77.75 | 83.28 |
EEGNet 201824+ | STD | S & R | SSCNN | 2.9 k | 2.1 k | 77.39 | 87.41 |
MMCNN 202135 | STD | SW & GN | MSCNN + SE | 90.3 k | 90.3 k | 81.43 | 84.40 |
MBEEGNet 202236+ | STD | S & R | MSCNN | 7.1 k | 4.7 k | 78.69 | 84.48 |
MSNet (proposed) | STD | S & R | MSCNN | 8.6 k | 5.3 k | 79.23 | 86.18 |
Conformer 202340 | BPF & STD | S & R | SSCNN + Transformer | 789.6 k | 759.2 k | 78.66 | 84.63 |
ADFCNN 202443 | BPF & EEMS | - | MSCNN + Transformer | 5.4 k | 3.0 k | 79.39 | 87.81 |
MSCFormer (proposed) | STD | S & R | MSCNN + Transformer | 145.9 k | 144.9 k | 82.95 | 88.00 |
