Segmentation and Tracking of Eruptive Solar Phenomena with Convolutional Neural Networks

Solar eruptive events are complex phenomena, which most often include coronal mass ejections (CME), CME-driven compressive and shock waves, flares, and filament eruptions. CMEs are large eruptions of magnetized plasma from the Sun’s outer atmosphere or corona, that propagate outward into the interplanetary space. Over the last several decades a large amount of remote solar eruption observational data has become available from ground-based and space-borne instruments. This has recently required the development of software approaches for automated characterisation of eruptive features. Most solar feature detection and tracking algorithms currently in use have restricted applicability and complicated processing chains, while complexity in engineering machine learning (ML) training sets limit the use of data-driven approaches for tracking or solar eruptive related phenomena. Recently, we introduced Wavetrack - a general algorithmic method for smart characterization and tracking of solar eruptive features. The method, based on a-trous wavelet decomposition, intensity rankings and a set of filtering techniques, allows to simplify and automate image processing and feature tracking. Previously, we applied the method successfully to several types of remote solar observations. Here we present the natural evolution of this approach. We discuss various aspects of applying Machine Learning (ML) techniques towards segmentation of high-dynamic range heliophysics observations. We trained Convolutional Neural Network (CNN) image segmentation models using feature masks obtained from the Wavetrack code. We present results from pre-trained models for segmentation of solar eruptive features and demonstrate their performance on a set of CME events based on SDO/AIA instrument data.

💡 Research Summary

This research presents a groundbreaking approach to the automated segmentation and tracking of solar eruptive phenomena, such as Coronal Mass Ejections (CMEs) and flares, by bridging the gap between traditional algorithmic methods and modern deep learning. A primary bottleneck in applying machine learning to heliophysics has been the scarcity of high-quality, manually labeled training datasets. To overcome this, the researchers utilized a specialized algorithm called ‘Wavetrack’—which employs a-trous wavelet decomposition and intensity ranking—to automatically generate precise feature masks. These masks serve as the ‘ground truth’ for training Convolutional Neural Networks (CNNs), effectively automating the labeling process and mitigating the need for labor-intensive manual annotation.

The study adopts the U-Net architecture for the segmentation task, specifically chosen for its parameter efficiency and its ability to preserve spatial information through skip connections. This architecture is particularly effective for processing high-dynamic-range (HDR) solar images, such as those from the SDO/AIA instrument, as it can simultaneously capture fine-scale structural details and large-scale features like the Coronal Bright Front (CBF). Beyond simply implementing a standard U-Net, the research team conducted a systematic hyperparameter optimization process, adjusting the network’s depth, width, and filter sizes to tailor the model to the specific characteristics of different solar observation data, such as AIA 193Å differential images.

One of the most significant advantages of this approach lies in its inference efficiency. While the training phase relies on the complex Wavetrack algorithm, the resulting trained models can generate segmentation masks for new observational data almost instantaneously, without the need for the original complex processing chain. This streamlined inference capability is crucial for the large-scale batch processing of massive solar databases and holds immense potential for real-time, on-board processing in future space-borne missions. The study demonstrates that the trained CNN models achieve performance levels comparable to, or even exceeding, the original algorithmic method. This suggests that machine learning can successfully internalize and generalize the logic of expert-designed algorithms, providing a scalable and robust framework that can be easily extended to different wavelengths and various solar observation instruments in the future.