Model-based Identification and 3D reconstruction of solar coronal mass ejections using deep neural networks

SÁNCHEZ, M.; IGLESIAS, F.A.; CISTERNA, F.; MACHUCA, Y.; LLOVERAS, D.; MANINI, F.; LÓPEZ, F.M.; CREMADES, H.; ASENSIO-RAMOS, A.

San Juan

Congreso; 65a Reunión Anual de la Asociación Argentina de Astronomía; 2023

Coronal mass ejections (CMEs) are critical drivers of space weather, with significant implications for technology and society. Due to the challenges in forecasting CME occurrences, assessing their geoeffectiveness is of paramount importance. A key task is the accurate identification and assessment of CME 3D morphology in coronagraph images. Recent advancements in Deep Neural Networks (DNNs) show important progress in image recognition and segmentation tasks, however, there is no large dataset of segmented CMEs usable for supervised training. To address this gap, we introduce a synthetic dataset of CME coronagraph images, blending actual quiet coronagraph images with synthetic CMEs generated using the Graduated Cylindrical Shell geometric model (GCS). In this poster, we present preliminary results from two DNN-based models. The first model is the fine-tunned MaskR-CNN, trained to identify and segment the outer envelope of CMEs in single differential coronagraph images, producing a GCS-like mask. The second model estimates the simplified 3D structure of the CME outer envelope from 2 or 3 simultaneous differential coronagraph images, employing a modified AlexNet model to infer the GCS parameters that best fit the CME outer envelope in the input images.