Electronics Essay

337 Words2 Pages
Nonlinear and non-stationary nature of biomedical signals, such as Electroencephalogram (EEG) that is of critical importance in understanding brain functions, dictates necessity for proper and accurate analysis techniques. Hilbert Huang Transform (HHT) is one of the algorithms suitable for such nonlinear and nonstationary signal analysis. In present project, a new HHT algorithm was implemented with a novel boundary condition for spline-interpolated envelopes during the decomposition. An improved approach to estimate the instantaneous frequency from the HHT’s analytical signal output was used in obtaining the time-frequency data. A new method of estimating power-related features from the HHT time-frequency data was developed and used in analyzing various EEG signals, such as Epileptic EEG, SSVEP BCI EEG, Alcoholic and Control VEP EEG. Various features were extracted, analyzed, and compared using the modified Kruskal-Wallis test that minimizes the effect of sub-group selection during the evaluation of the intra-group and inter-group H-statistics. The power-based features extracted by HHT from the EEG signals were utilized next in classification. k – Nearest Neighbor method based on Manhattan distance with the Leave-One-Out-Cross validation scheme was predominantly used in this project for classification of EEG signals in three independent experiments. First, Epileptic EEG signal analysis was performed and the signals were successfully classified with the accuracy of 95% for the realistic classification problem. Next, SSVEP BCI EEG signal analysis was conducted and the classification results obtained were on par to the output of advanced algorithms that use preprocessed filtered input signal with an accuracy around 70% for a three class classification problem and 92% for a two class classification problem. Finally, VEP EEG collected from alcoholics

More about Electronics Essay

Open Document