Research On Solar Image Segmentation Method Based On Deep Learnin

The study of solar filaments has long been one of the hot topics in solar physics research,as they can be used to track the solar atmospheric magnetic field and are extremely important for the study of the solar magnetic field.Recent studies have found that Coronal Mass Ejections(CMEs)and solar flares are closely related to filament eruptions,which produce flares,coronal mass ejections(CMEs)and release large amounts of magnetic energy,ejecting solar energetic particles and magnetized plasma that cause strong perturbations in the Earth’s space environment and cause drastic changes in the surrounding and even large-scale solar atmosphere.The CMEs and the release of large amounts of magnetic energy from the solar energetic particles and magnetized plasma will trigger strong perturbations in the Earth’s space environment and cause drastic changes in the surrounding and even large-scale solar atmosphere.Coronal mass ejection is an intense solar eruption activity that can affect the Earth’s space environment and human life,and it is an obvious event that accompanies the release of material from the Sun by the corona.Therefore,the analysis of the filaments will not only help to explore the activity of the filaments themselves,but also help to understand the physical nature of the CMEs and flares,so the study of the solar filaments is of positive importance to protect the Earth and the human environment.In this paper,we investigate the solar filament segmentation method using high quality observational data from the observatory.This paper focuses on how to effectively segment the solar filament images,and the main work has the following two aspects:(1)An AA-UNet solar image segmentation algorithm based on UNet network is proposed,which adds the attention mechanism to the downsampling structure and UNet network for effective combination.By focusing the attention on the solar filaments and then testing on the solar filaments segmentation dataset,the experimental results show that the accuracy of the solar image filaments segmentation can be improved.The effectiveness of the algorithm is verified by comparing it with the proposed algorithm for solar filament segmentation.(2)The dual-path network approach,based on the consideration of high-level features and local fine features,is chosen and a gated axis attention mechanism is added to obtain higher accuracy and improve the segmentation effect of tiny dark bars and solve the problem of broken dark bars on the solar dark bar dataset.Here,the gated position-sensitive axis attention mechanism is added to enable it to better focus and learn the image features to obtain more accurate segmentation results.The image segmentation algorithm based on deep learning technology effectively combines artificial intelligence with solar filament segmentation,which provides a feasible idea to guide physicists in solar activity research and promotes the rapid development of artificial intelligence technology in the field of solar activity research.