Hybrid Neural Networks for Electroencephalography Motor Imaging Classification

Abstract

The classification of Electroencephalogram (EEG) Motor-Imaging (MI) signals is becoming a hot topic of research in the field of brain-computer interface (BCI). But this study has many challenges, such as The properties of the electroencephalogram (EEG) are often noisy when collected. The EEG signal is complex with many features in space and time domains, so the electricity EEG is more difficult to decipher than other data types such as text, images, archival data, and EEG that differ individually for each sample examined and collected. Recently, Convolutions Neural Networks (CNN) studies have demonstrated that CNN can be used to efficiently extract features from electroencephalographic (EEG-MI) motor images. Especially convolutional neural networks that combine many layers of convolutional neural networks, the combination of convolutional neural networks and transfer learning methods, or convolutional neural networks combined with bugs, short- long-term memory (LSTM), fully-connected (FC) to increase the efficiency of motor image EEG classification. From those studies, in this paper, we propose Hybrid Convolutional Neural Network (HCNN) as a method combined with transfer learning (TL) to extract and classify 4-dass (right arm, left arm, foot, and tongue) EEG-MI features of the competing BCI sample dataset IV 2a. A mixed neural network (HCNN) is a combination of integrated neural network (CNN) and Long-term Short-Term Memory (LSTM) used to classify features in the spatial and temporal domains of signals. EEG-MI signal. To solve the problems of individual differences in EEG, we implement Full Connection (FC) to fine-tune the parameters relative to the training data. Our proposed method improved the results obtained compared to the results we selected for the match.