A Study Of Cross-individual Spatial Cognitive Ability Evaluation Based On EEG Signal

Currently,there are many methods for assessing spatial cognitive ability,but few methods focus on individual differences.In EEG signal analysis,the problem of data scarcity and insufficient labeling hinders the learning of the target task.Therefore,it is difficult to identify the spatial cognitive abilities of each subject with a single individual model for different individuals.Cross-individual EEG analysis is better applied to new subjects by learning from data that has been labeled by other subjects.Cross-individual EEG analysis is challenging due to differences in brain structure,function,behavior,and cognitive characteristics between individuals.Firstly,from the perspective of reducing the distribution difference of EEG data between individuals,this paper proposes a multi-source individual matching data alignment method,compares the similarity between different individuals,selects individuals with high similarity as the source user for the target user,and then aligns the source user and the target user with Euclidean spatial data to reduce the distribution difference between individuals,and then improve the accuracy of cross-individual spatial cognitive ability assessment.Secondly,this study starts from the perspective of extracting features that are not related to individuals and tasks,combines deep learning with manifold learning and multihead attention mechanism,and proposes a multi-head attention model based on manifold learning,which can not only use the multi-head attention mechanism to select some features that are independent of individuals but related to spatial cognition,but also use manifold learning to maintain the geometry of EEG during feature extraction,and at the same time update the parameters through deep learning calculation error backpropagation,and achieve a good classification effect on new individuals.Finally,the results show that the multi-source individual matching data alignment algorithm can effectively reduce the distribution difference between the subjects’ EEG data and visualize the effect compared with the algorithm without individual matching and data alignment.Compared with other cross-individual EEG analysis methods,the multi-head attention model based on manifold learning can effectively extract features related to spatial cognitive ability that are not related to individuals,and the attention mechanism inside the model can visualize the classification results,and explain the correlation changes between EEG signal channels before and after spatial cognitive training.