Oral Presentation

Searching Direct-Imaging Exoplanets through Denoising Convolutional Neural Network

Author(s): Pattana Chintarungruangchai (NTHU), Ing-Guey Jiang (NTHU), Jun Hashimoto (NAOJ, NINS-ABC), Yu Komatsu (NAOJ, NINS-ABC), Mihoko Konishi (Ōita University)

Presenter: Pattana Chintarungruangchai (National Tsing Hua University)

The data of exoplanet imaging is usually analyzed by angular differential imaging (ADI) technique with principal component analysis (PCA). ADI takes a series of image frames with short exposure time, and combine them to increase S/N. When there are less/more frames, S/N is smaller/larger and the exoplanets are less/more clear. We try to make a 2D convolutional neural network (CNN) that can make exoplanets more clear for those results with less frames. The resulting picture with more frames is the objective picture and those pictures made from fewer frames are the input of CNN. CNN can do machine learning and establish the connection between the objective picture and input picture. Finally, CNN can output pictures which are as clear as those with more frames.