WFC3 IR Blob Classification with Machine Learning
2021-10-25, 20:15–20:30, Grand Ballroom

IR blobs are small, circular, dark artifacts in the Hubble Space Telescope's WFC3 IR images caused by particulates that are occasionally deposited onto a flat mirror that is nearly optically conjugate to the IR detector. Machine learning can potentially reduce the effort currently devoted to visually inspecting blobs. We describe how machine learning (ML) techniques have been implemented to develop software that will automatically find new IR blobs and notify the WFC3 Quicklook team. This report describes the data preparation, development of the ML model, and criteria for success. The results of our latest test cases demonstrate that the model finds blobs reliably, with the model correctly classifying blob and non-blob images 94% and 88% of the time, respectively. We also report tips and lessons learned from our experience in machine learning as a result of this project.


Understanding and improving machine learning, Modernizing and maintaining telescopes, Other

See also: The accompanying publication (Instrument Science Report) to our talk, presenting the data processing steps and model architecture in more detail.