EEG decoding method based on AOVMD multi-feature and CPOSVM

In order to improve the accuracy of electroencephalogram (EEG) decoding and classification and further improve the performance of Brain-Computer Interface (BCI), this paper introduces a novel EEG signal decoding model utilizing the Artemisinin Optimization (AO) to search the optimal penalty factor $\alpha$ of variable mode decomposition (VMD), Dot-extract fusion and Crested Porcupine Optimizer (CPO) to search the penalty coefficient C and kernel function parameter g of support vector machine. Using Artemisinin optimization algorithm, the fitness criterion was determined by employing the minimal envelope entropy to search and determine the optimal factor $\alpha$ of VMD. The pre-processed EEG was decomposed into multiple Intrinsic Mode Function (IMFs) and a residual. The IMF excluding 8-30 Hz frequency band was removed, and the remaining IMF was reconstructed. The FBCSP was utilized to capture spatial characteristics from the reconstructed signal, followed by the extraction of entropy and Power spectral density attributes from the EEG. The fused features were sent into the SVM classifier optimized by Crown Hedgehog algorithm for decoding, and the average classification accuracy was 95.83%. The outcomes demonstrate that the EEG decoding model exhibits practicality and reliability.