DCT-Tensor-Net for Solar Flares Detection on IRIS Data

• Main Authors: Ullmann, Denis, Voloshynovskiy, Slava, Kleint, Lucia, Krucker, Sam, Melchior, Martin, Huwyler, Cedric, Panos, Brandon
• Format: Conference Proceeding
• Language: English
• Subjects: astrophysics; big data applications; computer aided analysis; data analysis; data preprocessing; data processing; detection algorithm; Discrete cosine transforms; feature extraction; Frequency-domain analysis; image texture analysis; Iris; Satellites; scientific computing; Sun
• ISSN: 2471-8963
• DOI: 10.1109/EUVIP.2018.8611672

Flares are an eruptive phenomenon observed on the sun, which are major protagonists in space weather and can cause adverse effects such as disruptions in communication, power grid failure and damage of satellites. Our method answers the importance of the time component in some scientific video observations, especially for flare detection and the study is based on NASA's Interface Region Imaging Spectrograph (IRIS) observations of the sun since 2013, which consists of a very asymmetrical and unlabeled big data. For detecting and analyzing flares in our IRIS solar video observation data, we created a discrete cosine transform tool DCT- Tensor-Net which uses an empirically handcrafted harmonic representation of our video data. This is one of the first tools for detecting flares based on IRIS images. Our method reduces the false detections of flares by taking into consideration their specific local spatial and temporal patterns.