RSVP Target Detection Based On Temporal-Spatial Convolutional Neural Network

The rapid development of the Internet has brought massive information to modern society.In many fields related to image processing,a great challenge faced by image analysts has become how to quickly and efficiently detect the target image of interest from massive images.Although the computer vision system can complete the target recognition quickly and accurately,the processing effect of some low-quality pictures still can not reach the ideal state.As a new way of human-computer interaction,brain-computer interface can convert the user’s brain activity mode into a simple message or code.The rapid serial visual presentation(RSVP)paradigm based brain-computer interface system provides the possibility to realize efficient target image detection by using human visual system.The powerful information processing ability of human visual system and the fast analysis ability of modern EEG signal processing algorithm can play a role together.Therefore,improving the recognition accuracy of EEG signals in the brain-computer interface system based on RSVP has become the key to the effective work of the system.Traditional machine learning algorithms and deep learning models have achieved considerable results in the field of image and speech processing,which has also attracted the attention of researchers in the field of EEG signal processing.EEG has the inherent characteristics of low signal-to-noise ratio,non-stationary and strong specificity,which is the biggest problem faced in the analysis and processing of EEG.To solve these problems,the existing algorithms and models have proposed variety of solutions to achieve efficient classification of EEG signals,and considerable results have been achieved.However,when analyzing and processing EEG signals,the existing algorithms and models ignore the phase locking characteristics and electrode distribution position information of EEG signals under RSVP paradigm,which is helpful to classify EEG signals.Starting from the key characteristics of EEG in RSVP paradigm,this paper proposes a convolutional neural network based on time-domain phase preserving by learning from the design framework of the current mainstream deep learning model.In the design of the network,the phase information of the signal in time domain is fully considered,and the feature representation of EEG signal with better representation ability is extracted.In order to verify the performance of the proposed model,this paper introduces two public datasets and one local dataset.The performance of the proposed convolution neural network is compared with the existing excellent algorithms and models on these three datasets.The results show that the classification accuracy of the proposed model is significantly higher than that of the existing methods.On this basis,a convolution neural network based on timedomain phase preserving spatial brain topology is proposed.The temporal-spatial convolution neural network fully excavates the electrode distribution position information of the signal in the spatial domain,and extracts the temporal-spatial representation of EEG signal for classification.The superior performance of the model is verified on two datasets.In addition,the classification performance comparison experiment is further analyzed and discussed,and the effects of different model parameters on model performance are discussed.The results not only prove the efficient feature extraction ability of the proposed model,but also agree with the research conclusions in the field of neuroscience,which verifies the effectiveness of this method again.