EEG Signal Classification for Motor Imagery Tasks

An electroencephalogram (EEG) is an experiment to determine the activity of electrical brain signals using small, metal electrodes which are affixed to the brain scalp. These EEG signals are one of most complex signals that represents the brain activity. Dataset is collected from the Physionet data-base which comprises the motor Imagery functions of Left and right fist, both fists and both feet. These recordings might be distorted by the contamination of traces such as blinking of eyes or muscle movements etc. In order to segregate the relevant neural signals from the recordings several feature extraction techniques are available. In this paper, the method called the common spatial patterns (CSP) has been implemented as feature extraction technique. This CSP technique includes one-hot encoding and z-score normalization. CSP is the characteristic eradication approach with the usage of spatial filtering technique that differentiates the EEG signalsof two classes. One-hot encoding is used to transform categorical statistics into integer information. The Deep learning technique used for classification is two-Dimensional convolutional neural network (CNN).Convolutional Neural Network contains three hidden layers. The filter size of first convolutional layer is 32, the kernel size is 3*3, the filter size of second convolutional layer 64 and the kernel size is 3*3 and The filter size of third convolutional layer is 128 and the kernel size is 3*3.We have got the training accuracy of 93.8% and the validation accuracy of 92.9%.