Application Research On Feature Extraction And Classification Of EEG Signal With The Method Of ELM

Human’s senior neural activity, such as behavior、thoughts and emotions, is subject to the command and control of the brain. Therefore, there is rich effective information in the human brain. How to effectively obtain and utilize this information has been a hot issue of concern to researchers. EEG(electroencephalogram) is the change of event related potentials in cerebral cortex caused by the interaction of numerous interconnected neurons, which is one of the important means to obtain information of the brain. Brain-Computer Interface(BCI) technology, which allows users to control computers and other external devices by brain activities, is an effective mean to utilize the information of the brain. EEG signal recognition is the key technology of BCI, which includes feature extraction and classification of EEG signals. Some BCI system demands high accuracy and fast learning speed. However, it is difficult to balance the two above demands for traditional algorithm. Therefore, Extreme Learning Machine(ELM), which has better generation performance and faster learning speed, is adopted to recognize EEG signals. The research work of this paper is as below.Firstly, in the aspect of feature extraction, aiming to extract discriminative features of EEG signals with high dimension efficiently, an EEG signals’ feature extraction method based on Extreme Learning Machine Auto-encoder(ELM-AE) is proposed. ELM-AE is based on basic ELM. ELM-AE’s output weight is learned to represent the feature of the EEG input data via singular values. The Compressed ELM-AE is used to extract feature and reduce dimension of the EEG input data. Experiment results show the classification accuracy with the features extracted by ELM-AE is higher than other related methods in the condition that they use the same classifier.Then, in the aspect of classification, as the weights of the connections between the input layer and hidden neurons are randomly selected, the classification accuracy may be affected. An improved ELM method called Constrained Extreme Learning Machine(CELM) is adopted to classify EEG signals. In CELM, the weights of the connections between the input layer and hidden neurons are randomly drawn from a closed set of difference vectors of between-class samples, rather than an open set of arbitrary vectors. There is more discriminative information of input samples in the weights of CELM. Experiment results show the efficiency of applying CELM to EEG data, compared with the state of the art methods. This paper also introduces Multilayer Extreme Learning Machine(ML-ELM) to classify EEG signals. ML-ELM adds more hidden layers to ELM. Those more hidden layers is connected by the output weights produced by ELM-AE. Experiment results show classifying EEG signals with the method of ML-ELM can not only has the same advantage as ELM but also get better performance than ELM.Finally, in this paper, an EEG signal analysis process system is programmed with Matlab, in which encapsulates the methods of this paper and related methods of the reference. The system aims to analyze EEG signals quickly and compare the performance of the related methods of feature extraction and classification easily.