Imagined Digit Recognition Based On Brain-Computer Interface Electroencephalogram

Brain Computer Interface(BCI)is a study field that uses the Electroencephalogram(EEG)to record human activity.This study was started in 1929 when a Scientist named Hans Berger invented the EEG to enable human psychology study.EEG is the equipment used to record human brain by attaching several metal plates called electrodes on human scalp.These electrodes will catch the electric field of human brain in micro-volt and enable the reading in computer.The location to put the electrodes is based on the 10-20 electrode positioning system.The naming is 10-20 electrode positioning system,because initially,the number of electrodes used are only 10-20,but nowadays it can reach more than 256.In this thesis the EEG has 14 electrodes for the recoding.Based on the 10-20 electrode positioning system,they are named:AF3,F7,F3,FC5,T7,P7,01,02,P8,T8,FC6,F4,F8,AF4.Ultimately the goal of BCI is to enable people with disability to be able to work like other normal people.One of the examples in the BCI for disabled people is the BCI for wheel chair control.However,the older definition for the BCI ultimate goal is changing with the arrival of modern computing.Artificial Intelligence is taking part in most technology.Therefore,the goal of BCI is also changing for a better technological advancement.Nowadays,people are trying to emerge the technology to their body for daily usage.Such technologies are being developed and known as the biotechnology.BCI is one of it,and planned to be used for multi-purpose control.As some research fields in BCI are being developed,i.e.the P300 simulation and motor imaginary,there are still limits in some other new BCI research field.One of them is the recognition of human imagination through EEG.There have been several studies done for the imagination recognition,but in this decade,the result is far from satisfying for real world implementation.Hence,this thesis goal is to push beyond the limit for human imagination recognition by enabling the differentiation between two digits.The experiment is divided into two parts for comparison.The first part is by using the traditional machine learning classifiers and another one is by using the deep learning classifier.For the first part,EEG recording for imagining a single digit,either 0 or 1,is done for 100 times(30 seconds of recording per trial).After that,some preprocessing is done to the concatenated 100 samples including band-pass filter.Then feature extraction is done by using ERP and PSD analysis,CSP,etc.And finally,classification is done by having SVM,LDA,ANN-MLP and Logistic Regression as classifiers.In ANN-MLP outperforms the other classifiers with the highest rate 66.88%.In the second part,deep learning is applied by using the designed CNN architecture.The CNN is known best for its performance in image classification.Therefore,ERP topo-maps are plotted for the 10 seconds middle span of 30 seconds recorded time.The ERP topo-maps are then fed into the CNN architecture for classification.Out of 15 subject who took part in the experiment,the highest classification rate is 99.65%.