| The biological signal is an important research field of modern digital signal processing.Because of the complexity of life system,biological signals are usually nonlinear,chaotic,non-stationary and belong to nonlinear time series.It is because of these characteristics,the biological signal analysis is different from the general signal processing and it has become an important branch of modern digital signal processing.The proposed algorithm of transforming time series to complex network makes complex network a new tool for nonlinear time series analysis.The proposed algorithm of transforming time series to complex network makes complex network a new tool for time series analysis.After years of development,complex networks have formed a mature theoretical system,which has been widely used in various disciplines.In this paper,based on the existing complex network construction methods,we mainly studies weighted proximity network and weighted horizontal visibility graph(WHVG).Through improving the shortcomings of the current algorithm,we propose new complex network construction methods,which are the improved weighted proximity network and the improved WHVG.The research content and innovation points are as follows:(1)This paper proposed improved weighted proximity network.In the process of constructing traditional proximity network,the number of nodes is small and the information of edges weight is not introduced,which will result in the loss of time series information.After average segmentation of epileptic EEG time series,12 time-domain statistical features are extracted from each sub-time series as nodes of complex network,which increases the number of nodes.Then,Euclidean distance between nodes is calculated,threshold is set,and the connection relationship between nodes is determined.Slope between nodes is used as the edge weight and weight matrix is constructed to complete the construction of epileptic EEG weighted proximity network.The average weighted degree is extracted as classification feature,which is used to classify ictal EEG signals and interictal EEG signals.The highest classification accuracy is 96%.The experimental results show that the performance of the improved weighted proximity network is much better than that of the traditional proximity network algorithm.(2)The traditional WHVG algorithm cannot reflect the dynamic characteristics of time series well.This paper proposed three improved WHVG algorithms.The first is to construct WHVG by using 12 time-domain statistical features as nodes of WHVG and using the difference between nodes as the edge weight.The first algorithm reduces the time complexity.The second is to construct WHVG using the original epileptic EEG time series as the nodes and the slope or the angle between nodes as the edge weight.Thirdly,the original epileptic EEG time series is used as nodes,the slope between nodes is used as the edge weight of the forward network,and the angle between nodes is used as the edge weight of the reverse network.This paper firstly proposed bidirectional WHVG.After transforming epilepsy EEG signals into the improved WHVG,the average weight degree and weight distribution entropy are extracted to classify ictal EEG signal and interictal EEG signals.The highest classification accuracy was 98.5%.The experimental results show that the performance of the improved WHVG is much better than that of the traditional complex network.(3)In this paper,we proposed improved weighted proximity network and improved WHVG.Then,we transformed epileptic EEG signals into complex network through the two proposed method and extracted topological statistical features of corresponding complex network to classify epileptic EEG signals with high accuracy,which is of great clinical significance.By comparing with other related experiments,it is verified that the proposed method can obtain more comprehensive and deeper dynamic characteristics of biological signals,which indicates that the proposed method can be effectively applied to biological signal analysis. |
PDF Full Text Request