Fatigue Driving Recognition Based On Deep Learning And Graph Neural Network

Fatigue driving is one of the major causes of traffic accidents.The performance of fatigue driving identification model under noisy environment is greatly reduced and it is difficult to achieve ideal results in practical application.In order to solve the problem of how to maintain the stability of the fatigue driving recognition model performance when any channel of EEG signals suffers from noises,this paper establishes a fatigue driving recognition framework based on a denoise deep convolutional neural network and a dynamic functional brain network construction method.A residual convolutional EEG denoise network,a functional brain network construction method based on singular entropy and fractal dimension and a spatial convolutional graph neural network for functional brain network classification based on B-spline curve are proposed in the framework.The performance of the framework in fatigue driving recognition and its stability in noise environment are analyzed.The main work and phased research results of the project are as follows:(1)Traditional denoising methods are difficult to adapt to complex situations due to the uncertain type and amplitude of noises in practical applications.In order to solve this problem,this paper improves the shortcomings of denoise auto encoder and proposes a 1D residual convolutional EEG denoise Net(RC-EEGdenosie Net)which based on deep learning technology.RC-EEGdenoise Net firstly uses a residual structure to estimate the distribution of noises by removing EEG signals from input signals,then the clean signals can be obtained by subtracting the estimated distribution of noises from input signals.The experimental results show that RC-EEGdenoise Net achieves better performance than other methods when EEG signals suffer from noise with different types and amplitudes.(2)Traditional functional brain network can’t be classified directly by classifiers,in order to solve this problem,a functional brain network classification method based on graph neural network is proposed.The method firstly construct spatial domain graph convolution based on B-spline curve,then use graph neural network to classify functional brain network which combines with singular value entropy and fractal dimension features.The proposed method is compared with general methods,the experimental results show that the singular value entropy and fractal dimension features used in this paper are better than the general features in fatigue driving recognition,furthermore,the B-spline-curve-based graph neural network can achieve a more accurate and stable recognition result compared with traditional classifiers.(3)The recognition performance of traditional fatigue driving recognition methods will server decline if EEG signals contain lots of noises,in order to solve this problem,this paper constructs a stable fatigue driving recognition framework.This framework firstly use the proposed RC-EEGdenoise Net to denoise EEG signals.then extracting features based on singular value entropy and fractal dimension from clean EEG data and constructs functional brain network.Finally,the graph neural network recognition model is used to directly classify functional brain networks.This paper tests the recognition stability of the proposed framework when different numbers of channels suffer noises of different signal-to-noise ratios,The experimental results show that the framework has a great performance under different situations.