University of Louisville
Principal Curve Extraction from Solar Images
Institution
University of Louisville
Faculty Advisor/ Mentor
Olfa Nasraoui; Nurcan Durak
Abstract
Astrophysicists currently study NASA's extensive databases of solar satellite images, in part because these images provide valuable data about solar events such as coronal loops. They lament that a lot of time is wasted sifting through these databases in order to find the images with interesting coronal loops. Our project automates the process of loop identification, using a computer algorithm to locate all shapes within an image that resemble pieces of ellipses. Our main challenge was that loop pieces are more easily distinguished by the human eye than by a computer in these usually cluttered solar images. Since only the principal coronal loops are desired, they are found by automatically tracing curves in an image as they fade in and out along their length, correctly following them even as they overlap with other curves, and finally delivering only the smoothest, most prominent curve segments as output. The proposed method successfully discovers the full length of coronal loops in real solar images. Our algorithm promises to accelerate the search for relevant data which can in turn support astrophysicists in various scientific analyses related to the coronal heating problem and solar weather prediction. This can in turn accelerate the science return from NASA's missions related to studying the Sun as a star.
Principal Curve Extraction from Solar Images
Astrophysicists currently study NASA's extensive databases of solar satellite images, in part because these images provide valuable data about solar events such as coronal loops. They lament that a lot of time is wasted sifting through these databases in order to find the images with interesting coronal loops. Our project automates the process of loop identification, using a computer algorithm to locate all shapes within an image that resemble pieces of ellipses. Our main challenge was that loop pieces are more easily distinguished by the human eye than by a computer in these usually cluttered solar images. Since only the principal coronal loops are desired, they are found by automatically tracing curves in an image as they fade in and out along their length, correctly following them even as they overlap with other curves, and finally delivering only the smoothest, most prominent curve segments as output. The proposed method successfully discovers the full length of coronal loops in real solar images. Our algorithm promises to accelerate the search for relevant data which can in turn support astrophysicists in various scientific analyses related to the coronal heating problem and solar weather prediction. This can in turn accelerate the science return from NASA's missions related to studying the Sun as a star.