Research On Classification Method Of Motor Imagery Based On Modified S-Transform And Convolutional Neural Network

Brain-computer interface is a direct communication and control channel established between human brain and a computer or other devices.Through this channel,people can directly express ideas or manipulate devices without languages or actions.Brain-computer interface systems can expand the new way for normal people to communicate with the outside world,and also can effectively enhance the ability of disabled patients with muscle weakness due to nervous system disorders to communicate with the outside world,thereby improving the living quality of patients.For more than 20 years,with the further understanding of the nervous system function and the development of computer technology,brain-computer interface technology has become a research hotspot in the fields of artificial intelligence,biomedical engineering,and computer technologies.A motor imagery based brain-computer interface system recognizes the subject's brain intentions by analyzing sensorimotor rhythms and issuing control commands.The system consists of signal acquisition interface,signal processing interface,and application interface.The signal processing interface mainly consists of preprocessing,feature extraction,channel selection and classification,which can be used to translate the user's intentions.This thesis conducts research based on the signal processing module of brain-computer interface.This paper not only analyses the changes of sensorimotor rhythm caused by motor imagery but also realizes the recognition of motor imagery tasks.Besides three algorithms are developed to identify brain signals.The main contents are:1.We propose a motor imagery classification method based on modified S-transform and support vector machine.This method uses support vector machine to classify brain signal features based on modified S-transform.In order to reduce the data redundancy and signal interference caused by multiple channels,an optimized wrapped algorithm is developed to select the most appropriate channel combination.The channel selection part reduces the algorithm complexity and improves the classification accuracy.The result of this scheme is satisfactory.2.We propose a classification method based on combined features and probabilistic collaborative representation.The power spectrum density features based on the modified S-transform and the fractal features based on the blanket dimension technology are connected into a new combined features after recording and preprocessing the brain signals.The robust probabilistic collaborative representation is used to classify the extracted combined features,and finally a good classification result is obtained.3.A classification and recognition method based on convolutional neural network is proposed.The convolutional neural network is applied to the classification and recognition of motor imagery actions.The modified S-transform algorithm is used to preprocess the original brain data,and the convolutional neural network performs feature extraction and classification on the preprocessed signals.Both the feature extraction and classification of the brain signals are completed in the convolutional neural network,which avoids the data loss of the interface between the feature and the classifier in the traditional method.This development plan has stronger generalization and development prospects.The research in this thesis is conducive to further promote the development of the motor imagery based brain-computer interface systems.It has a positive effect on realizing a better and faster real-time brain-computer interface system.