Event Title

Detecting Ghosting Artifacts in Telescope Images using Machine Learning

Session Number

Project ID: ERSP 3

Loading...

Media is loading
 

Advisor(s)

Dr. Chihway Chang; University of Chicago, Fermilab

Discipline

Earth and Space Sciences

Start Date

22-4-2020 11:30 AM

End Date

22-4-2020 11:55 AM

Abstract

When unwanted artifacts appear in telescope images, whether through imperfect optics, faulty image sensors, cosmic ray hits, or airplanes and satellites, it becomes difficult to process the images for data as the artifacts can confuse data-processing programs. An artifact that is especially difficult to avoid is the ghosting artifact or the ghost, which is caused by scattering and internal reflections of light of the telescope’s mechanical and optical components. Because large cosmological surveys collect thousands of images per night, it’s extremely time-consuming to search for and manually label the images with artifacts. In our study, we used data from the Dark Energy Survey to construct a machine learning model for distinguishing ghosts automatically. This paper details the model’s structure and training process and presents the images used for training and validation. The model’s performance was compared with that of the ray-tracing program currently being used for distinguishing ghosts.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Share

COinS
 
Apr 22nd, 11:30 AM Apr 22nd, 11:55 AM

Detecting Ghosting Artifacts in Telescope Images using Machine Learning

When unwanted artifacts appear in telescope images, whether through imperfect optics, faulty image sensors, cosmic ray hits, or airplanes and satellites, it becomes difficult to process the images for data as the artifacts can confuse data-processing programs. An artifact that is especially difficult to avoid is the ghosting artifact or the ghost, which is caused by scattering and internal reflections of light of the telescope’s mechanical and optical components. Because large cosmological surveys collect thousands of images per night, it’s extremely time-consuming to search for and manually label the images with artifacts. In our study, we used data from the Dark Energy Survey to construct a machine learning model for distinguishing ghosts automatically. This paper details the model’s structure and training process and presents the images used for training and validation. The model’s performance was compared with that of the ray-tracing program currently being used for distinguishing ghosts.