Real-time solar image classification: Assessing spectral, pixel-based approaches

Abstract

In order to utilize solar imagery for real-time feature identification and large-scale data science investigations of solar structures, we need maps of the Sun where phenomena, or themes, are labeled. Since solar imagers produce observations every few minutes, it is not feasible to label all images by hand. Here, we compare three machine learning algorithms performing solar image classification using extreme ultraviolet and Hα images: a maximum likelihood model assuming a single normal probability distribution for each theme from Rigler et al. (2012), a maximum-likelihood model with an underlying Gaussian mixtures distribution, and a random forest model. We create a small database of expert-labeled maps to train and test these algorithms. Due to the ambiguity between the labels created by different experts, a collaborative labeling is used to include all inputs. We find the random forest algorithm performs the best amongst the three algorithms. The advantages of this algorithm are best highlighted in: comparison of outputs to hand-drawn maps; response to short-term variability; and tracking long-term changes on the Sun. Our work indicates that the next generation of solar image classification algorithms would benefit significantly from using spatial structure recognition, compared to only using spectral, pixel-by-pixel brightness distributions.

The GOES‐R Solar UltraViolet Imager

“…The algorithm currently used to produce operational thematic maps is an improved version of the random forest approach described in Hughes et al. (2019). The current algorithm is still a pixel based random‐forest approach, that is, it classifies each pixel separately with a random forest without knowledge of neighboring pixels and instead only using spectral information from the six SUVI channels.…”

Change point detection and image segmentation for time series of astrophysical images

“…Several observatories have been collect- ing images in extreme ultra-violet (EUV) filters and in X-ray passbands for several decades, and analyzing them to pick out interesting changes using automated routines have been largely unsuccessful. Catalogs like the HEK (Heliophysics Events Knowledgebase Hurlburt et al 2012;Martens et al 2012) can detect and mark features of particular varieties, though these compilations remain beset by incompleteness (see, e.g., Aggar-wal et al 2018;Hughes et al 2019;Barnes et al 2017). In this context, our method provides a way to model solar features without limiting it to a particular feature set to identify and locate regions in images where something interesting has transpired.…”

Possible Advantages of a Twin Spacecraft Heliospheric Mission at the Sun-Earth Lagrangian Points L4 and L5

“…Over the last ∼10 years, a rising number of researchers moved their interest to the promising research fields of Artificial Intelligence and Machine Learning (see review by Camporeale, 2019) applied to solar and i.p. physics data for SW forecasting purposes, for instance to predict the occurrence of solar eruptions and flares (e.g., Ahmed et al, 2013;Benvenuto et al, 2018;Florios et al, 2018), predict geomagnetic storms (e.g., Sexton et al, 2019), and to detect and classify solar events (Martens et al, 2012;Armstrong and Fletcher 2019;Hughes et al, 2019). These methods are really promising, but it is maybe too early to know what will be the new results that these methods will provide in the end.…”