Stellar Blend Image Classification Using Computationally Efficient Gaussian Processes (MuyGPs)

Abstract

Stellar blends are a challenge in visualizing celestial bodies and are typically disambiguated through expensive methods. To address this, we propose an automated pipeline to distinguish single stars and blended stars in low resolution images. We apply different normalizations to the data, which are passed as inputs into machine learning methods and to a computationally efficient Gaussian process model (MuyGPs). MuyGPs with 𝑁𝑡 ℎ root local min-max normalization achieves 86% accuracy (i.e. 12% above the second-best). Moreover, MuyGPs outperforms the benchmarked models significantly on limited training data. Further, MuyGPs low confidence predictions can be redirected to a specialist for human-assisted labeling.