Automatic detection of CMEs using synthetically-trained Mask R-CNN

Coronal mass ejections (CMEs) are a major driver of space weather. To assess CME geoeffectiveness, among other scientific goals, it is necessary to reliably identify and characterize their morphology and kinematics in coronagraph images. Current methods of CME identification are either subjected to human biases or perform a poor identification due to deficiencies in the automatic detection. In this approach, we have trained the deep convolutional neural model Mask R-CNN to automatically segment the outer envelope of one or multiple CMEs present in a single difference coronagraph image. The empirical training dataset is composed of 10^5 synthetic coronagraph images with known pixel-level CME segmentation masks. It is obtained by combining quiet coronagraph observations, with synthetic white-light CMEs produced using the GCS geometric model and ray-tracing technique. We found that our model-based trained Mask R-CNN infers segmentation masks that are smooth and topologically connected. While the inferred masks are not representative of the detailed outer envelope of complex CMEs, the neural model can better differentiate a CME from other radially moving background/foreground features, segment multiple simultaneous CMEs that are close to each other, and work with images from different instruments. This is accomplished without relying on kinematic information, i.e. only the included in the single input difference image. We obtain a median IoU=0.98 for 1.6*10^4 synthetic validation images, and IoU=0.77 when compared with two independent manual segmentations of 115 observations acquired by the COR2-A, COR2-B and LASCO C2 coronagraphs. The methodology presented in this work can be used with other CME models to produce more realistic synthetic brightness images while preserving desired morphological features, and obtain more robust and/or tailored segmentations.

💡 Research Summary

This paper presents a novel machine learning approach for the automatic detection and segmentation of Coronal Mass Ejections (CMEs) in white-light coronagraph images. CMEs are major drivers of space weather, and accurately identifying their morphology and kinematics is crucial for assessing their geoeffectiveness. Current methods rely on manual catalogs, which are subjective and time-consuming, or traditional image-processing algorithms, which often struggle to distinguish CMEs from dynamic background features like streamers.

The core innovation of this work lies in its two-step methodology: synthetic data generation and supervised training of a state-of-the-art deep learning model. To overcome the scarcity of pixel-level labeled data, the authors generated a large-scale synthetic dataset. They combined real, quiet-Sun coronagraph background images with synthetic CME brightness images created using the Graduated Cylindrical Shell (GCS) geometric model and a ray-tracing technique. This process yielded approximately 113,000 synthetic difference images, each paired with a precise pixel-level segmentation mask, serving as ground truth for training. While the GCS model simplifies the complex, detailed structure of real CMEs, it preserves essential topological properties like connectivity and self-similar expansion.

This synthetic dataset was used to train a Mask R-CNN model, a deep convolutional neural network designed for instance segmentation. Mask R-CNN can detect multiple objects, classify them, and produce a high-quality segmentation mask for each instance within a single image. The trained model was evaluated in two phases. First, on a held-out set of 16,000 synthetic validation images, it achieved a median Intersection over Union (IoU) score of 0.98, demonstrating near-perfect performance on data similar to its training set. Second, and more importantly, its performance was tested on 115 real observations from the LASCO C2, COR2-A, and COR2-B coronagraphs. When compared against independent manual segmentations by human experts, the model achieved a median IoU of 0.77, indicating strong generalization capability to real, complex scenarios.

The study highlights several key advantages of this synthetically-trained Mask R-CNN approach. It can differentiate CMEs from other radially moving background/foreground features using only a single difference image, without heavy reliance on temporal kinematic information. It successfully segments multiple CMEs that appear close to each other in the same frame. Furthermore, the model shows instrument-agnostic potential, working on data from different coronagraphs. The authors conclude that the proposed framework is highly flexible. The methodology is not tied to the GCS model; future work could integrate more sophisticated and physically realistic CME models (e.g., from magnetohydrodynamic simulations) to generate even more realistic synthetic training data. This path promises to yield more robust, accurate, and tailored CME segmentation tools, ultimately contributing to improved 3D reconstruction and space weather forecasting.