Motor Imagery ECoG Signal Classification Using Sparse Representation with Elastic Net Constraint

In recent years, the brain-computer interface (BCI) technology based on the motor imagery has provided a new method for people to communicate with the outside world. How to effectively extract features and improve the recognition rate of EEG signals is one of the hot problems in this field. This study is based on the motor imagery ECoG signals, in which the common spatial pattern (CSP) algorithm is used for feature extraction, and then the extracted energy features are classified by the classification algorithms. In order to improve the classification accuracy of the ECoG signals, this study introduces the sparse representation-based classification (SRC) algorithm with the elastic network constraint. Then the accelerated proximal gradient (APG) algorithm and the least angle regression (LARS) algorithm are respectively applied to sparse coding for the ECoG signals. The elastic network which combines the L1 norm and the L2 norm not only avoids the over-fitting problem, but also has a higher prediction ability than the Lasso algorithm. The experimental results demonstrate that the proposed method can achieve better classification performance than other algorithms, such as the sparse representation algorithms with L1 minimization, SVM, KNN, Adaboost, and Naive Bayes.