Research on feature extraction method based on epileptic EEG signal

Epilepsy is a common brain disease, and the epileptic region in the human brain can be located non-invasively through EEG. The frequency and rhythm of epileptic EEG signals are obviously different from those of normal EEG signals. When epileptic seizures occur, EEG will fluctuate temporarily, so as to detect epilepsy. In order to improve the diagnostic accuracy of epileptic EEG, this paper proposes a feature extraction method of epileptic EEG based on mutual information and Shannon entropy. Firstly, the genetic algorithm is used to optimize variational modal decomposition (VMD) to find the most suitable number of decomposition modes k and penalty factors, and then the original epileptic EEG signal is decomposed by VMD. Secondly, the decomposed components are optimized by mutual information, which VMD components have strong correlation and their temporal or spatial relationship, and the VMD components with high mutual information are found, and the Shannon entropy is calculated to compare the complexity and information between different VMD components, so as to optimize feature selection. Finally, the appropriate features are selected from the calculation results and classified by support vector machine (SVM). Compared with the traditional single feature extraction method, the proposed method has advantages. The experimental results show that the average accuracy rate is 85.87%.