Combination of the Motor Imagery EEG (MI-EEG) imaging and Deep Convolutional Neural Network is a prospective recognition method in brain computer interface. Nowadays, the frequency or time-frequency analysis has been applied to each channel of MI-EEG signal to obtain a spatio-frequency or time-frequency image, and even the images from several channels are infused to generate a combined image. However, the real position information of channels or electrodes is lost in these MI-EEG images, and this is contradictory to the activation area of MI-tasks. In this paper, the MI period and the frequency band covered by μ and β rhythms are divided into ten time windows and three sub-bands, respectively. Then, for each electrode, Fast Fourier Transform (FFT) is employed to transform each time window to spectrum, and its inverse FFT is calculated for each sub-band. The time-domain powers of ten time windows are averaged for the same sub-band. So, three average powers are generated as the time-frequency features of each electrode of MI-EEG. They are further arranged to the electrode coordinate figure by using Clough-Tocher interpolation algorithm, and a complicated image, in which the time-frequency features are correctly located at the real position of each electrode, is obtained to embody the MI-EEG in detail. Furthermore, a VGG network is modified to perform effective recognition for MI-EEG image, and it is called mVGG. Extensive experiments are conducted on three publicly available datasets, and the 10-folds cross validation accuracies of 88.62%, 92.28% and 96.86% are achieved respectively, and they are higher than that of the state-of-the-art imaging methods. Kappa values and ROC curves demonstrate our method has lower class skew and error costs. The experimental results show that the effectiveness of proposed MI-EEG imaging method, and it is well-matched with mVGG.

Motor imagery EEG (MI-EEG), which reflects one’s active movement intention, has attracted increasing attention in rehabilitation therapy, and accurate and fast feature extraction is the key problem to successful applications. Based on wavelet packet decomposition (WPD) and SE-isomap, an adaptive feature extraction method is proposed in this paper. The MI-EEG is preprocessed to determine a more effective time interval through average power spectrum analysis. WPD is then applied to the selected segment of MI-EEG, and the subject-based optimal wavelet packets (OWPs) with top mean variance difference are obtained autonomously. The OWP coefficients are further used to calculate the time-frequency features statistically and acquire the nonlinear manifold structure features, as well as the explicit nonlinear mapping, through SE-isomap. The hybrid features are obtained in a serial fusion way and evaluated by a k-nearest neighbor (KNN) classifier. The extensive experiments are conducted on a publicly available dataset, and the experiment results of 10-fold cross-validation show that the proposed method yields relatively higher classification accuracy and computation efficiency simultaneously compared with the commonly-used linear and nonlinear approaches.

Robotic-assisted rehabilitation system based on Brain-Computer Interface (BCI) is an applicable solution for stroke survivors with a poorly functioning hemiparetic arm. The key technique for rehabilitation system is the feature extraction of Motor Imagery Electroencephalography (MI-EEG), which is a nonlinear time-varying and nonstationary signal with remarkable time-frequency characteristic. Though a few people have made efforts to explore the nonlinear nature from the perspective of manifold learning, they hardly take into full account both time-frequency feature and nonlinear nature. In this paper, a novel feature extraction method is proposed based on the Locally Linear Embedding (LLE) algorithm and DWT. The multiscale multiresolution analysis is implemented for MI-EEG by DWT. LLE is applied to the approximation components to extract the nonlinear features, and the statistics of the detail components are calculated to obtain the time-frequency features. Then, the two features are combined serially. A backpropagation neural network is optimized by genetic algorithm and employed as a classifier to evaluate the effectiveness of the proposed method. The experiment results of 10-fold cross validation on a public BCI Competition dataset show that the nonlinear features visually display obvious clustering distribution and the fused features improve the classification accuracy and stability. This paper successfully achieves application of manifold learning in BCI.

The breakthrough of electroencephalogram (EEG) signal classification of brain computer interface (BCI) will set off another technological revolution of human computer interaction technology. Because the collected EEG is a type of nonstationary signal with strong randomness, effective feature extraction and data mining techniques are urgently required for EEG classification of BCI. In this paper, the new bionic whale optimization algorithms (WOA) are proposed to promote the improved extreme learning machine (ELM) algorithms for EEG classification of BCI. Two improved WOA-ELM algorithms are designed to compensate for the deficiency of random weight initialization for basic ELM. Firstly, the top several best individuals are selected and voted to make decisions to avoid misjudgment on the best individual. Secondly, the initial connection weights and bias between the input layer nodes and hidden layer nodes are optimized by WOA through bubble-net attacking strategy (BNAS) and shrinking encircling mechanism (SEM), and different regularization mechanisms are introduced in different layers to generate appropriate sparse weight matrix to promote the generalization performance of the algorithm.As shown in the contrast results, the average accuracy of the proposed method can reach 93.67%, which is better than other methods on BCI dataset.