基于多层时变功能脑网络特征的运动想象识别OA北大核心CSTPCD

Aiming at the problem that the dynamic change of signal structure with time and the separation and integration process of network were ignored in the recognition of motor imagination electroencephalography(EEG)signals,a method of feature extraction of motor imagination based on multi-layer time-varying functional brain network was proposed.In this method,the effective fragments of motor imagination were extracted and put into EEGLAB for signal preprocessing.According to the sliding window method,appropriate length and step length were set,and the signals were divided into continuous and partially overlapping time windows.The EEG data intercepted by the time window were generated into multiple brain networks,and the multi-layer time-varying network model was constructed based on the phase-locking values between nodes.First,the core network layer was determined through the network topology analysis of different layers of the multi-layer time-varying network and the inter-layer similarity metric,and the node degree and clustering coefficient were extracted to describe the functional connection in the network space.Then,multi-layer participation coefficient and multi-layer clustering coefficient were combined to describe the dynamic changes and separation and integration characteristics of EEG networks,and the two characteristices were combined to form the feature vector of multi-layer time-varying brain functional network to complete the task of motor imagination recognition.Results of support vector machine(SVM)identification show that the classification accuracy of the proposed network feature vector is as high as 89.14%,which is 6.61%higher than that of the single layer network feature used for comparison.