Speaker
Description
Segmentation and characterization of solar coronal structures are essential for advancing our understanding of the solar atmosphere and accurately identifying key regions such as active regions and coronal holes which are precursors to phenomena like solar flares and coronal mass ejections (CMEs). In parallel, it is crucial to incorporate onboard such artificial intelligence (AI) algorithms into future space missions to automate tasks, overcome challenges such as communication delays when ground-based human intervention is required, intelligently activate instruments, discard non-essential data to conserve storage and resources, and prioritize critical data for downlink.
In this study, we investigate various complementary approaches to automate this process. First, we employ a previously presented deep learning-based U-Net architecture tailored for segmenting and characterizing solar coronal structures. Since this framework is notoriously heavy to be deployed even for testing, we explore ways to address this problem more efficiently without compromising performance. To this end, we first reformulate the task as a detection problem and explore the widely used You Only Look Once (YOLO) model, whose nano edition requires 95% fewer trainable parameters than U-Net. We also propose an even more lightweight Convolutional Neural Network (CNN) capable of achieving comparable accuracy. Additionally, we design a framework that combines basic computer vision techniques with traditional machine learning methods to segment and characterize solar coronal structures. This approach relies on a carefully selected set of hand-crafted features and unsupervised clustering methods such as K-means and t-SNE, facilitating the distinction of coronal features like active regions, coronal holes, and bright spots. All methods are evaluated and compared against each other and against state-of-the-art techniques using metrics such as the Dice score and Intersection over Union (IoU). Furthermore, we assess them in terms of trainable parameters, number of operations, and inference time across various hardware configurations.
The method that best balances segmentation and detection performance with computational efficiency will be selected for integration into a prototype designed to support future space exploration missions. Employing such a framework onboard can automate solar activity monitoring and facilitate solar phenomena prediction and observance such as solar flares and coronal mass ejections (CMEs). This work is part of the project AutomaticS in spAce exPloration “ASAP” funded by the European Union under HORIZON Research and Innovation Action (GA no.101082633).
| Do you plan to attend in-person or online? | In-person |
|---|
