Motor Imagery EEG Decoding Method Based on a Discriminative Feature Learning Strategy

Abstract

With the rapid development of deep learning, more and more deep learning-based motor imagery electroencephalograph (EEG) decoding methods have emerged in recent years. However, the existing deep learning-based methods usually only adopt the constraint of classification loss, which hardly obtains the features with high discrimination and limits the improvement of EEG decoding accuracy. In this paper, a discriminative feature learning strategy is proposed to improve the discrimination of features, which includes the central distance loss (CD-loss), the central vector shift strategy, and the central vector update process. First, the CD-loss is proposed to make the same class of samples converge to the corresponding central vector. Then, the central vector shift strategy extends the distance between different classes of samples in the feature space. Finally, the central vector update process is adopted to avoid the non-convergence of CD-loss and weaken the influence of the initial value of central vectors on the final results. In addition, overfitting is another severe challenge for deep learning-based EEG decoding methods. To deal with this problem, a data augmentation method based on circular translation strategy is proposed to expand the experimental datasets without introducing any extra noise or losing any information of the original data. To validate the effectiveness of the proposed method, we conduct some experiments on two public motor imagery EEG datasets (BCI competition IV 2a and 2b dataset), respectively. The comparison with current state-of-the-art methods indicates that our method achieves the highest average accuracy and good stability on the two experimental datasets.

A Temporal-Spectral-Based Squeeze-and- Excitation Feature Fusion Network for Motor Imagery EEG Decoding

“…In order to realize MI-EEG decoding, many machine learning algorithms have been proposed [5][6][7][8][9]. For example, the Common Spatial Pattern (CSP) was a powerful algorithm in discriminating MI-EEG signals [6], which constructed spatial filters and extracted time-frequency features effectively.…”

“…Saha et al further [10] used CSP with Joint Approximate Diagonalization (JAD) and applied wavelet decomposition as the feature extraction method, which generated the subband energy and entropy of EEG. However, these methods above highly focused on the energy features of EEG, which failed to obtain features with high discrimination from raw EEG signals subjectdependently, and thus limited the decoding performance of MI-EEG [9].…”

“…Additionally, for BCI IV 2a dataset, some recent studies [5,8,9,12] report results by using 10-fold cross-validation on merged data (576 trials) for each subject. Therefore, in order to make a fair comparison with these conventional methods, Table VIII gives the comparison between the proposed TS-SEFFNet and recent studies on BCI IV 2a by training on merged data (576 trials) for each subject.…”

“…Deep ConvNet [28] 2017 92.50 0.900 Hybrid Net [28] 2017 91.80 0.891 Residual Net [28] 2019 88.90 0.852 FBCSP [19] 2019 90.90 0.879 CP-MixedNet [19] 2019 PSO-Rough Set [5] 2016 80.99 0.743 Brain Network [8] 2017 79.70 0.730 3D-CNN [12] 2019 75.02 0.644 Discriminative Feature [9] 2021 81.85 0.758 Our method 2021 84.49 0.794 where the bold values indicate the proposed method. frequency bands.…”

BECT Spike Detection Based on Novel EEG Sequence Features and LSTM Algorithms

“…Recently, machine learning based neural networks have been widely used for EEG analysis, such as epilepsy and seizure detection [26][27][28][29], spike detection [30][31][32], brain-computer interfaces [33][34][35], etc. Webber et al [30] implemented spike detection algorithms through spike candidate selection and artificial neural networks (ANN) based classification.Özdamar et al [31] directly adopted ANN to learn on the raw EEGs, and explored the influence of the feature input dimension and network structure parameters.…”