| Electroencephalography(EEG)contains abundant information of brain neural activity.Decoding EEG can facilitate the understanding of brain cognitive mechanisms.Due to the high temporal resolution of EEG data,it was widely used in the study of brain cognitive mechanisms.This thesis mainly focused on visual attention cognition,visual attention experimental design and visual search EEG decoding,and developed an attention training system based on visual attention research results.This thesis mainly included the following three aspects:Firstly,neural network technology was used to study the effects of visual attention training on patients with Autism Spectrum Disorders(ASD).The oddball paradigm was used to induce P300 component to study the health status of ASD patients.Neural network model(EEGNet)was used to decode the EEG data generated during visual attention training,and then the saliency map was used to obtain the P300 features of the social and non-social scenarios effectively decoded by EEGNet.And the effect of visual attention training on the characteristic changes of P300 component was explored by saliency map.The results showed that the P300 latencies of ASD patients were significantly shorter after training in both scenarios,while the amplitude value of the P300 peak decreased only in the social scenario.The results suggested that visual attention training can help improve the health status of ASD patients.Secondly,visual attention experiments were designed and a neural network framework with an external attention mechanism was constructed to decode EEG for attentional cognition of visual search.In order to improve the cognitive decoding accuracy of visual search and overcome the problem of large number of trainable parameters in existing attention mechanisms,this research proposed an external attention EEG network(EAEEGNet)for decoding EEG under the visual search task.The effectiveness of the proposed neural network and its components were verified by EEG data of 26 subjects and ablation experiments.Compared with the traditional neural network models,the EAEEGNet demonstrated an excellent performance.Furthermore,the saliency map confirmed that the features extracted by EAEEGNet were consistent with the cognitive mechanism of visual search.Finally,cross-scenario transfer learning experiments were performed on EAEEGNet,and the results showed that EAEEGNet had robust migration learning capability.Thirdly,based on the above research results,a visual attention training system based on brain-computer interface was constructed.Based on the results of the above research on the attention effect in visual search paradigm,the system developed an attention training system on the Windows operating system.The system mainly involved the construction of a visual stimulation module and a data processing module.At the same time,the model training time was shortened when a transfer learning technology was used in the constructed system.Finally,12 users were used to verify the system,and it was found that the users’ attention was significantly improved and the purpose of visual attention training was achieved. |
