Research And Implementation Of EEG Signal Feature Recognition Method For Cognitive Navigation

Spatial navigation is one of the basic cognitive functions that human beings rely on for survival.The process of navigation requires the brain to integrate various sensory information and depends on the participation of higher-level thinking activities such as reasoning,memory,and decisionmaking.Electroencephalography is a bioelectrical signal generated when brain neurons exchange information which contains rich physiological and psychological information,and is an important way to study the state of the brain.Feature extraction and analysis of EEG signals for different navigation tasks are helpful for the understanding of human navigation cognitive function.According to the characteristics of EEG signal being susceptible to interference and low signal-to-noise ratio,the signal is preprocessed from the time domain and frequency domain respectively,which suppresses the noise and reduces the redundancy.In order to extract the features of the navigation EEG signals of different experimental groups,a multi-domainbased feature combination method is proposed.Three feature combinations of signal time-domain statistical features,time-frequency-domain wavelet packet decomposition sub-band energy features and sample entropy are selected respectively,and cognitive state classification experiments are carried out based on support vector machines and random forests.Aiming at the problem that feature construction depends on human prior knowledge,a multi-time scale spatiotemporal composite classification model is proposed based on convolutional neural network.This network directly takes the preprocessed EEG signal as the input,and takes the cognitive state to which the signal belongs as the output,which effectively reduces the addition of prior knowledge.In order to verify the performance of the model,a navigation EEG signal cognitive state classification experiment,an EEG experiment paradigm recognition experiment and a time-scale corrosion experiment were constructed.The experimental results show that the model can effectively improve the classification accuracy and F1-Score.In order to further explore the waveform characteristics of navigation EEG in different cognitive states,a dual-target model of multi-time-scale spatiotemporal composite reconstruction and classification is proposed based on autoencoder network and convolutional neural network.The model starts from vectors and reconstructs EEG signals of different cognitive states.Under the common constraints of the dual objectives,the model can not only achieve good recognition ability,but also mine the EEG waveforms in different cognitive states.Finally,based on the summary of EEG signal processing and feature extraction process,a feature extraction system for EEG signal is designed and developed.The system is mainly composed of a signal preprocessing module;a feature construction module;a deep model training module and a feature display module.