EEG Signal Classification Based On Extreme Learning Machine Algorithm

Brain-computer interface(BCI)can construct a direct communication and control channel between the brain and external environment by computers,which is different with the normal pathway depending on peripheral nerves and muscle tissue.The BCI systems can not only help to improve the self-care ability for the handicapped,but also be applied in many areas,such as rehabilitation training,game,wearable devices and military.Therefore,the study of BCI system has great application value and practical significance.Electroencephalogram(EEG)signals can effectively reflect the station of human physiological functions,emotions and other information.Feature extraction and classification with high accuracy and fast speed is a key of the EEG-based BCI system.Extreme learning machine(ELM)has been demonstrated to have excellent generalization performance and fast learning speed,and thus it has potential advantages in BCI applications.To further improve the performance of basic ELM,it is improved by combining deep learning and semi-supervised learning,and then applied for classification of EEG signals.The main work can be seen as follow:(1)This paper summarizes the research status of BCI and ELM,and points out the shortcomings of the state-of-the-art ELM-based algorithms including hierarchical ELM(H-ELM)and semi-supervised ELM(SS-ELM).Then some improved algorithms are proposed and applied for EEG classification,yielding a good classification performance.(2)Due to the limited of single hidden layer structure,feature learning is not enough in SS-ELM,and the H-ELM algorithm ignores unlabeled samples which may include a lot of useful information.To address these issues,a novel method of hierarchical semi-supervised ELM(HSS-ELM)is proposed in this paper.Firstly,the deep architecture of H-ELM is employed for feature learning automatically,and then these new high level features are classified using the SS-ELM algorithm which can exploit the information from both labeled and unlabeled data.Compared with SVM,ELM,SAE,H-ELM and SS-ELM,the proposed method can achieve a good trade-off between high accuracy and fast speed on benchmark datasets and BCI Competition IV Dataset 2a.(3)Due to the ignorance of the risk of unlabeled samples,SS-ELM may perform worse than ELM in some situations.To address this issue,a safe semi-supervised ELM(Safe-SSELM)algorithm is proposed by designing a novel risk-based mechanism for SS-ELM,which utilizes the Wasserstein distance to compute the risky degrees of unlabeled data,and it is suitable for safe semi-supervised classification in multi-class situations.Experimental results on benchmark datasets and BCI Competition IV Dataset 2a show the effectiveness of the proposed method.