1. A Literature Review on Data Conversion Methods on EEG for Convolution Neural Network Applications
- Author
-
Jafri Malin Abdullah, Mohd Zaid Abdullah, Shahrel Azmin Suandi, Haidi Ibrahim, Azlinda Azman, and Chi Qin Lai
- Subjects
medicine.diagnostic_test ,business.industry ,Computer science ,Spectral density ,Pattern recognition ,computer.file_format ,Electroencephalography ,Convolutional neural network ,Pearson product-moment correlation coefficient ,Data conversion ,symbols.namesake ,Wavelet ,Robustness (computer science) ,symbols ,medicine ,Artificial intelligence ,business ,computer - Abstract
Convolution neural network (CNN) presents high robustness in computer vision applications. In state-of-the-art methods, CNN is being used in EEG processing for various classification and problem solving. To enable EEG to fit in the CNN architecture, data conversion of EEG has to be done. The ways of data conversion need to be investigated in order to fully utilize the information. From the study, it was found that the simplest way of re-arranging the signal is by creating a two dimensional matrix of channels versus time points. There are approaches that compute Pearson correlation coefficients and fit them into a two dimensional matrix to represent the input signal. There are also methods which extract frequency components and fit them in matrix structure as channels versus frequency components, such as power spectral density. Other approaches includes graph representation and wavelet components.
- Published
- 2019
- Full Text
- View/download PDF