A typical method 17 is to represent EEG signals as 2D images by a short-time Fourier transform (STFT) method. In order to apply deep learning method to MI classification, EEG signals need to be represented as a processable form, which is a prerequisite to be satisfied, to meet this premise, EEG are often represented as a two-dimensional array, which taking the number of sampling electrodes as the height and the time step as the width. Moreover, these methods can not be widely used in large population due to the non universality of subjects.Ĭompared with machine learning frameworks 12, 13, 14, deep learning methods does not need to extract features manually, and embeds all calculations, including extracting feature and classification, into a single end-to-end network, which can overcome the disadvantages of traditional machine learning 15, 16.
Therefore, in the above-mentioned methods, the selection of the best filter band is usually subject specific, and it depends heavily on the quality of the hand-made features 11 thus if the suboptimal frequency band is selected in the feature extraction process, the classification performance may not be the best. These methods artificially extracted time–frequency features from EEG signals, and then combine these artificially extracted features into feature vectors, which are then used to train classifiers such as support vector machine (SVM) 8, 9 or decision tree 10 to classify EEG signals. In previous studies, machine learning methods like dynamic connectivity analysis 5, frequency band analysis 4, continuous wavelet transform 6, and Filter Bank Common Spatial Pattern (FBCSP) 7 have been widely proposed for EEG decoding. A large number of motor imagery classification methods have been proposed. Electroencephalogram (EEG) signal is a one of electrical signal widely used in brain computer interface (BCI) systems. Brain computer interface (BCI) based on ERS/ERD phenomenon provides a way for communication between computers and human brain by analyzing the electrical signals generated by brain nervous system 3. The decrease of power spectral ratio is called event-related desynchronization (ERD), and the increase of power spectral ratio is called event related synchronization(ERS) 1, 2. Modern neurophysiological studies shows that the power spectrum of some characteristic frequency components in EEG signals can be changed by actual body movement or imaginary brain movement. Experimental results show that the proposed method is a good method, which can improve classification effect of different motor stages classification. Experimental results indicated that there are also a problem of the different classification difficulty for different classes in motor stages classification tasks, we introduce focal loss to address problem of ‘easy-hard’ examples, when trained with the focal loss, the three-branch 3D-CNN network achieve good performance (relatively more balanced classification accuracy of binary classifications) on the WAY-EEG-GAL data set. To further utilize the spatial and temporal features of EEG signals, we proposed a 3D representation of EEG and an end-to-end EEG three-branch 3D convolutional neural network, to address the class imbalance problem (dataset show unequal distribution among their classes), we proposed a class balance cropped strategy. Recently, various deep learning methods are being focused on finding an easy-to-use EEG representation method that can preserve both temporal information as well as spatial information. Motor Imagery is a classical method of Brain Computer Interaction, in which electroencephalogram (EEG) signal features evoked by the imaginary body movements are recognized, and relevant information is extracted.