Retrieval of Images with Coronal Loops from Solar Image Databases

We present a system for the automatic retrieval of images with coronal loops from the solar image database captured by the SOHO/EIT satellite [1]. Our image retrieval system provides relevant data to astrophysicists who need such data to perform studies that shed light on the mysteries behind solar activities and to analyze the relations between coronal loops and other solar events. As part of building this system, we investigated various image preprocessing techniques, image based features, and classifiers to detect coronal loops automatically and to indicate their locations on the images. Despite many challenges related to the coronal loop characteristics, we obtained promising results, namely 85% precision and 83% recall in loop retrieval. 1) Problem and Motivation The Sun, the source of our life, is a highly energetic star where several gigantic energy revealing events occur. Some events such as coronal mass ejections or the solar wind affect the Earth and might cause damage to polar grids and satellites. Figure 1 illustrates several satellites monitoring the Sun and how a coronal mass ejection affects the Earth. Several satellites have been deployed to closely monitor the solar events, understand their dynamics, and take precautions to reduce their damage on Earth. These satellites, which include SDO, SOHO, TRACE, and YOHKOH, have been taking pictures of the Sun regularly and storing the images in public databases [1]. SOHO is the oldest satellite and has taken more than 30000 images that are stored in the SOHO online database [1]. One of the key solar events is coronal loops which are immense arches of hot gas on the surface (corona) of the Sun as shown in Figure 2. Coronal loops are linked to some other events such as the solar flare triggering or coronal heating problem. In order to make progress in studying coronal loops, scientific analysis requires the data observed by the instruments such as SOHO/EIT [1]. Currently, astrophysicists prepare the data set containing coronal loops by looking at every single image separately and putting markings around the coronal loops manually. The manual search for coronal loops is not only subject to human error but is also time-consuming and tedious. The main contribution of our study is developing an image retrieval tool that can (i) separate images with coronal loops from images without coronal loops; (ii) extract a subset of images that contain loops from a larger collection, and locate the detected coronal loops on these images. We are especially interested in coronal loops that are visible outside the solar disk. For this study, we focused on coronal loops that have strong arches and are easily discernable from the background. Figure 2 SOHO/EIT Image with a Coronal Loop Region on the outside of the Sun Figure 1 The interaction between the Sun and Earth along with the designated satellites [10] 2) Background and Related Work Coronal loop detection from solar images has sparked attention from different aspects [2, 3, 4]. The main aim of these studies is highlighting the loop structures on the given solar image regions by using curve tracing [2], ridge detection [4] or the wavelet transform [3]. Most of these algorithms, however, were tested and compared on images that were already known to contain loops. The main difference between our work and these previous studies is that our methods are not aware of the existence of loops in the given images, while existing methods already know that a given image contains a loop. Thus, our automated methods must first have to learn the characteristics of coronal loops and then distinguish regions with loops from the regions without loops.