Feature Extraction Of Week EEG Signal Based On Two-dimensional Transform

As a special bio-electrical signal that reflects cerebral activity, electroencephalogram (EEG) is a significant basis for the diagnosis and treatment of brain diseases. The characteristics of weak amplitude, strong randomness, non-stationary and nonlinear make it difficult to analyze EEG effectively. Feature extraction plays very important roles in EEG signal analysis and the effect of the extracted feature is of great significance for research on brain function. Therefore, feature extraction of weak EEG is a hot topic in the field of biomedical engineering all the time. At present, reported by national and international documents, the processing method of feature extraction of EEG is given priority to traditional one-dimensional EEG processing and nonlinear dynamic analysis. The former method is simple but the reflected characteristic of non-stationary is not obvious. Although the extracted feature of weak EEG by the latter method is prominent enough, it is hard to achieve the rapid extraction because of the complexity of algorithm.This study is mainly to design a new method for weak EEG hidden feature extraction and evaluate its performance. In this method, one-dimensional EEG is mapping to two-dimensional signal based on the theory of two-dimensional transform. This article was focused on the feature extraction for the reconstructed two-dimensional EEG, using two-dimensional fast Fourier transformation (2D-FFT) and two-dimensional discrete cosine transformation (2D-DCT). Epileptic EEG as example was analyzed. The results show that using the method for extracting feature of short sequence of EEG, the rapid detection of pre-ictal and accurate classification of epilepsy stages were implemented.In this paper, the hidden feature of weak EEG was extracted rapidly and effectively by the proposed method with simple calculation. The new method is significant to the diagnosis of brain disease and research of brain science. The results also provide a new idea of feature extraction for other bio-electrical signals.