NASA 2024 Decadal Survey: Deep Learning and Computer Vision for Heliophysics

To make meaningful advancements in heliophysics, computer vision techniques based on deep neural networks (DNNs) are a crucial emerging tool. Fundamentally, computer vision is the study of training algorithms to gain high-level insights from imagery and video data, which is especially relevant to a largely observationally-based field such as heliophysics. Machine learning is applied to solar feature detection and phenomenon classification. Utilizing images taken in a multitemporal fashion, computer vision-based approaches can detect solar flares, filament eruptions, coronal jets, etc. and measure properties of our observations over timescales. Data from ​NASA's Solar Dynamics Observatory (SDO), including Atmospheric Imaging Assembly (AIA) and Helioseismic and Magnetic Imagery (HMI) data, are a useful source of imagery for training computer vision models. In this white paper, we briefly highlight current areas of application at the intersection of artificial intelligence and heliophysics and propose how the solar physics and astrophysics scientific communities can join to fill in interdisciplinary technological gaps.