基于时空特征学习Transformer的运动想象脑电解码方法

宋耀莲, 殷喜喆, 杨俊

Transformer based on temporal⁃spatial feature learning for motor imagery electroencephalogram signal decoding

Yaolian Song, Xizhe Yin, Jun Yang

表1 本文方法与其他方法测试不同受试者的正确率与Kappa系数

Table 1 Accuracy and Kappa coefficient of our method and other methods for different subjects

Subject C2CM[21] S3T[2] EEGNet[20] FBSF⁃TSCNN[22] Our method Acc Pk Acc Pk Acc Pk Acc Pk Acc Pk Average 74.46% - 82.59% - 64.60% 0.734 62.70% 0.720 84.16% 0.844 A01 87.50% - 91.67% - 79.20% 0.844 81.00% 0.858 94.50% 0.930 A02 66.28% - 71.67% - 33.50% 0.501 46.80% 0.601 79.76% 0.836 A03 90.28% - 95.00% - 84.10% 0.881 83.80% 0.878 93.70% 0.913 A04 66.67% - 78.33% - 50.90% 0.631 52.30% 0.642 84.39% 0.788 A05 62.50% - 61.67% - 54.30% 0.657 31.60% 0.486 58.67% 0.692 A06 45.49% - 66.67% - 47.00% 0.602 42.60% 0.569 70.80% 0.772 A07 89.85% - 96.67% - 82.20% 0.866 77.30% 0.830 94.14% 0.920 A08 83.33% - 93.33% - 73.20% 0.799 75.50% 0.816 90.88% 0.880 A09 79.51% - 88.33% - 77.00% 0.827 73.60% 0.802 90.67% 0.873