Classification Method of Motion Image EEG Signals Based on LMD-FBCSP and IGASVM

In order to solve the problem that traditional signal analysis methods cannot deal with nonlinear and non-stationary EEG signals effectively, which leads to inaccurate feature extraction and poor classification effect, this paper proposes an EEG signal classification method based on Local Mean Decomposition (LMD), Filter Bank Common Space Pattern (FBCSP) and Improved Genetic Algorithm (IGA) to optimize Support Vector Machine (SVM). By means of LMD, the pre-processed EEG signals are adaptive divided into multiple product functions (PF) and a residual quantity, from which the most discriminant PF components are selected to form a matrix, and the corresponding spatial features are extracted using the FBCSP. The features containing the largest amount of information are selected and combined into the feature matrix, which is sent to the IGA optimized SVM for classification, and the average classification accuracy is 91.0714%. The results show that the method has certain practicability and reliability.