Automatic Detection of Light Bridges in Solar Active Regions Based on Deep Learning Techniques

Understanding the nature of the phenomena that occur in the solar surface (photosphere) is greatly benefited from the technical capabilities of the new instruments, located either in terrestrial or space telescopes, providing ever greater spatial and temporal resolution, and therefore, allowing detailed observations of photospheric structures. In particular, new observations have revealed the intricate configuration of sunspots and substructures within them, such as so-called light bridges. Eventually, depending on its morphology and structure, light bridges can alter the evolution of the host sunspot, both in morphology and in other of its physical properties, playing a role on other phenomena such as magnetic reconnection or coronal rain, associated with energy releases in the solar atmosphere. This work presents the development of an identification method of light bridges in sunspots, through the initial implementation of an algorithm for the automatic detection, extraction and characterization of these structures, followed by the application of supervised classification techniques based on Machine Learning (ML) Convolutional Neural Networks (CNN). By using a sample of 265 active regions, over a period of 4 years from 2010 to 2014, with a cadence of 24 hours for each sunspot, from full-disk observations acquired by the Solar Dynamics Observatory (SDO) with the Helioseysmic and Magnetic Imager (HMI) instrument in the FeI (617 Å) line. Detection accuracy of 85.4% is reached, optimizing the model by the iterative variation of the hyperparameters according to the binary classification addressed. The method is planned to be further implemented to probe the dynamics of light bridges and their connection with the evolution of the corresponding active region.