Search

Sleep is an essential integrant in everyone’s daily life. Thereby, it is an important but challenging problem to construct a reliable and stable system, that can monitor user’s sleep quality automatically. In this brief, we combine complex network and deep learning to propose a novel Graph-Temporal fused dual-input Convolutional Neural Network (CNN) method to detect sleep stages by using the Sleep-EDF database. Firstly, we segment each single-channel EEG signal into non-overlapping 30s epochs to set up the network. For that, we map each epoch into a Limited Penetrable Visibility Graph (LPVG) and obtain the corresponding Degree Sequence (DS) by calculating the node degree. Finally, the DSs and the 30s EEG epochs are combined as inputs of the novel Graph-Temporal fused dual-input CNN to learn about the graph topology and about the temporal feature representations of the raw data for the purpose of classifying the sleep stages into the two-, three-, four-, five-, and six-state. Notably, the classification accuracy of six-state stage detection is 87.21% and the corresponding Kappa value is 0.80. The results demonstrate the effectiveness of our model structure in detecting sleep states, whereby they further provide a basic strategy for future sleep research. [ABSTRACT FROM AUTHOR]

Gas-liquid two-phase flow widely exists in chemical industries and natural gas industries. But characterizing the nonlinear flow behaviors underlying such flow remains a challenge of great importance. In this brief, we first carry out the vertical gas-liquid two-phase flow experiment in a 50mm-inner-diameter pipe to acquire signals via the four-sector conductance measurement system. Then, we develop a multiresolution multiplex network (MMN) to analyze multichannel measurements. In particular, we employ average edge overlap to quantitatively characterize the derived network corresponding to different flow patterns. Results show that our method allows to effectively uncover the evolution from bubble flow to slug flow. In addition, we calculate the approximate entropy of different flow patterns to verify the findings from our network analysis. All the results demonstrate that our proposed method allows characterizing the nonlinear flow behaviors from multichannel measurements and also provides a novel way for multivariate time series analysis. [ABSTRACT FROM AUTHOR]

Detecting fatigue driving from electroencephalogram (EEG) signals constitutes a challenging problem of continuing interest since fatigue driving has caused the majority of traffic accidents. We carry out a simulated driving experiment for EEG data acquisition. Then, we calculate the wavelet entropy under the alert and fatigue state, respectively, and find that the wavelet entropy gets an acceptable performance on classification. Despite that the traditional entropy-based methods have been successfully applied to detect EEG-based fatigue driving, how to improve the classification remains to be investigated. To solve this problem, we in this paper propose a novel relative wavelet entropy complex network (RWECN) for improving the classification accuracy. In particular, we infer the complex network by regarding each EEG channel as a node and determining the connections of nodes in terms of the relative wavelet entropy between the EEG signals. Then, we extract a series of network statistical measures to characterize the topological structure of the brain networks. We combine the wavelet entropy and RWECN statistical measures to form a feature vector for realizing the classification of different states through the Fisher linear discriminant analysis. The results suggest that our method allows obtaining intrinsic and effective features from fatigue EEG signals and enables to improve the classification accuracy of EEG-based fatigue driving. [ABSTRACT FROM AUTHOR]

The Steady State Visual Evoked Potential (SSVEP)-based Brain Computer Interface (BCI) system has seen extensively applications in many fields, such as physical recovery of handicap persons, obstacle avoidance of intelligent vehicles, entertainment and smart homes. However, subjects easily get fatigued because of the involving long-time operations. The presence of fatigue symptoms typically affect the efficiency of the BCI system, so investigating the effects of fatigue on the SSVEP classification accuracy from the perspective of brain network becomes a challenging issue of significant importance. In this paper, we develop an adaptive optimal-Kernel time-frequency representation (AOK-TFR)-based complex network method for characterizing fatigued behavior using the SSVEP-based BCI system. We apply the traditional Canonical Correlation Analysis (CCA) and Fisher Linear Discriminant Analysis (FLDA) to classify SSVEP signals. We find that the classification accuracy at the fatigue states is significantly lower than that at the normal states. To reveal the reasons, we infer and analyze the AOK-TFR-based functional brain network with SSVEP signals. In particular, we calculate the AOK-TFR of the acquired 30-channel SSVEP signals under both normal and fatigue conditions and then construct a brain network in terms of the two-norm distance between different channels. Our results suggest that the small-world-ness of the network at normal states is prominent, and the main brain regions associated with SSVEP are in the prefrontal cortex and occipital lobe. Our analysis sheds new insights into the understanding and management of the fatigued behavior using the SSVEP-based BCI system. [ABSTRACT FROM AUTHOR]

Motion state monitoring and recognition are the important issues to be dealt to improve the reliability of Autonomous Underwater Vehicle (AUV). In this work, we transform the motion state classification into Multivariate Time Series Classification (MTSC). By combining two kinds of MTSC methods, including the methods based on feature representation transformation and Deep Neural Network (DNN), we propose a new classification method for Multivariate Time Series (MTS). Multivariate monitoring data of AUV are fused to construct complex networks as graphs to represent the motion states. Then, Graph Convolutional Neural Network (GCNN) is used to extract the features of the graphs and classify the graphs. The effectiveness of our method is validated through sea experiments, whose data are from three classes of navigational motions: near the surface, at fixed depth, and influenced by unknown ocean currents at fixed depth. The experimental results show that the graphical representation based on complex networks can effectively describe the motion states. Compared with Support Vector Machine (SVM), the graphical features are extracted automatically by GCNN to get a higher accuracy of classification of the motion states. The experiments also show that the classification accuracy of our method is higher than that of other two DNNs. • Propose a new method to coarsen multivariate time series as complex network graphs on multi time scales. • Based on graph representation, transform the time series classification into graph classification. • Effectively improve the classification accuracy of motion states of AUV. [ABSTRACT FROM AUTHOR]