Classification of eeg based task on colour visualization and colour imagery
MetadataShow full item record
Electroencephalography (EEG) is a measure of brain waves used to monitor the state of health of the patients in medical applications and other research areas. EEG signals are also used to develop Brain Machine Interface (BMI) system. BMI helps to bring out the intention of users and it is an intelligent interfacing system which acts as a communication channel for sending messages to command the external world. It is one of the most promising communication approach for the differentially enabled people. Over the past two decades, many researchers have concentrated on developing a suitable BMI using variety of EEG signals such as slow cortical potentials, P300 potentials, visually evoked potentials and event related potentials. This thesis discusses the development of colour perception based BMI using non invasive method for the differentially enabled people. Two protocols using visualization and imagination of different colours were investigated. The EEG data was collected from ten subjects using mindset-24 EEG data acquisition instrument with 19 channel electrode cap arrangement. The data is preprocessed and features are extracted from the recorded EEG data. The extracted feature set is then fed to a neural network model to classify the different tasks. From the observed classification results, the spectral energy entropy features using probabilistic neural network has the highest classification performances. In EEG signals, frequency band and channel selection plays an important role in increasing the classification performance and in decreasing the number of input features. In this research work, frequency band and channel selection algorithm is proposed to find the relevant frequency bands and electrode positions (or channel) for the proposed BMI protocols. Experimental results show that the alpha, beta and gamma (αβγ) frequency band combinations gives better classification accuracy and the selected 9 channels using the proposed channel selection algorithm yields a better classification accuracy of above 90% when compared to the conventional method.