The Research On Target Recognition Of EEG Signals From Satellite Images Based On Rapid Serial Visual Presentation

With the advancement of Electroencephalogram(EEG)research technology,the automatic recognition of EEG signals has become an important research direction.To address the problems of small data sets and inconsistent acquisition standards,low recognition rate,poor generalization ability and long training period of the current satellite image EEG signal recognition task,this paper investigates the target recognition method of satellite image EEG data based on Rapid Serial Visual Presentation(RSVP)paradigm,and the main research content and work are summarized as follows.RSVP paradigm EEG experiments were designed and EEG datasets were constructed.There are currently few publicly available datasets on target recognition tasks,and different device acquisition standards prevent the construction of a universal method.To address this issue,EEG experiments were designed based on the RSVP paradigm using satellite images,and EEG data were acquired using a Neuro Scan 64-channel scanning system,followed by data pre-processing and noise removal using Independent Components Analysis(ICA).Finally,the EEG signal data sets of 15 subjects with a total of 3150 trials were acquired and constructed in the EEG laboratory of the Military Academy of Sciences,and the significant differences in the P300 components of the EEG signals evoked by target and non-target stimuli were observed and analyzed.The CNN-GRU EEG signal feature extraction algorithm is proposed.It solves the problem that traditional machine learning methods in EEG target recognition tasks rely on manual feature extraction and have low recognition accuracy.First,the EEG effective features of each channel are automatically extracted using a one-dimensional Convolutional Neural Network(CNN),in which the convolutional and pooling layer structures can adaptively extract frequency and spatial feature information for each channel’s EEG signal.The Gate Recurrent Unit(GRU)is then used to further extract multi-channel fusion features,where the gating mechanism can tap the time-dependent characteristics of the EEG data,allowing the whole CNN-GRU network to automatically extract more representative time-space-frequency feature information in the satellite image EEG target recognition task.The GRU-Attention EEG signal feature selection algorithm based on the attention mechanism is proposed.It solves the problem of long training time and poor generalization ability of models in existing deep learning methods.Firstly,GRU is used for feature extraction,which can extract effective features and reduce the number of parameters of the network,thus reducing the training time.Then,we use the Attention module modified by Dropout for feature selection,which can freeze some of the neural network nodes to further reduce the training time,and the Attention module can weight the extracted features to filter out the more representative and generalized P300 features.By selectively learning the filtered features,the training period of the whole GRU-Attention network is greatly reduced and the generalization ability is improved in the EEG target recognition task of satellite images.