Eyemovment visual dataset and auroral substorm recognition model

Eye movement patterns are considered empirical knowledge in this work. As a result, an auroral substorm eye movement dataset is established by collecting eye movement data from space physicists. Auroral substorm events in this dataset are comprehensive, including various types of auroral substorm sequences. We analyze these eye movement data and generate the eye movement patterns for various auroral substorms. In addition, we generate visual maps, which we learn from experts' eye movement patterns. By tracking how experts visually analyze auroral substorm images, we constructed a dataset of expert eye movements to record expert-level knowledge (experts' eye movements change) across different substorm phases.

In addition, we generate visual maps using an Eye Movement Pattern Prediction (EMPP) module, which learns from the eye movement patterns of experts. We have developed a novel space knowledge embedding module called the Visual-Physical Interactive (VPI) module, which simultaneously incorporates eye movement patterns and scientific knowledge. As a result, we have developed a visual-physical interactive model based on deep learning capable of accurately identifying auroral substorm