Synthetic loop generation and finding LFFF parameter 'Alpha' using Deep Learning
Recent advances in the field of neural networks have made convolutional neural networks (CNNs) a conventional algorithm for many computer vision tasks including image recognition and object detection. Because modeling the coronal magnetic field of the Sun is an important objective in heliophysics, this study extends the use of CNNs to the application of coronal magnetic field modeling. We employ a simple one-parameter model of linear force-free magnetic fields (LFFFs) to model active regions of multiple dipolar configurations including the Active Region (AR) 11117. We use state-of-the-art architectures such as ResNet and Inception networks, and develop our customized network “SolarNet” to determine the associated LFFF parameter alpha from a set of pseudo-coronal loop images, which are generated using the modeled active regions. Our results show very high accuracy of determining the LFFF parameter alpha, thereby demonstrating the effectiveness of the generic and customized deep CNN architectures to understand the coronal magnetic field.
[Astronomy and Computing] (https://www.sciencedirect.com/science/article/pii/S2213133718301148?via%3Dihub)