Solar Filament Detection using Mask R-CNN

Many algorithms have been proposed to detect solar filaments from H-Alpha images automatically, most of them use intensity thresholding method which requires intensity normalization. We propose a fast, automatic, and efficient method to detect filaments on H-Alpha images without carrying out intensity normalization using Mask R-CNN. The proposed method was applied on H-Alpha images obtained from Indonesian National Institute of Aeronautics and Space (LAPAN), resulting 96% precision. The average detection speed is 0.42 seconds per image. The method can be extended to perform detection on other solar features, such as sunspots, flares, and prominences.