A Motion Imagery EEG Signal Recognition Algorithm Based on Power Spectral Density combined with Particle Swarm Optimization Algorithm Optimized Support Vector Machine

For the purpose of addressing the issue of low classification accuracy resulting from low signal-to-noise ratios in electroencephalogram (EEG) signals, in this paper, a classification algorithm for motion imagery EEG signals (PSD-PSO-SVM) utilizing power spectral density analysis (PSD) combined with particle swarm optimization method (PSO) improved support vector machine (SVM) is proposed. A first step of the algorithm is to extract features of the EEG signal in the frequency domain by PSD, and the energy spectral densities of 0-30Hz in delta, theta, alpha, and beta frequencies are selected as their frequency domain features. Following that, a support vector machine is utilized for the classification of the extracted features. A solution is provided for the problem that performance in classification of traditional SVM is greatly influenced by its k-function parameters. It is the parameters of the kernel function of the SVM that are optimized by using the global optimisation-seeking capability of PSO to achieve optimal performance in classification. Finally, the algorithm validity was assessed by analyzing both the laboratory dataset (IMUT data) and the open dataset III developed for the 2003 competition on brain-computer interfaces (III BCI 2003). These findings show that PSD-PSO_SVM can perform classifications up to a level of 80.63%. It is proposed in this article that the proposed algorithm is compared to other optimization SVM algorithms, such as genetic algorithm optimization SVM (GA_SVM), and shows that PSO_SVM outperforms GA _ SVM with an average classification accuracy of 5.63%. Thus the PSD-PSO_SVM was shown to have good classification performance.