Abstract:
This study investigates the morphological classifications of galaxies in the early universe by employing the capabilities of Morpheus, a deep learning framework, in conjunction with the UNCOVER and CANDELS datasets from the James Webb Space Telescope (JWST) and the Hubble Space Telescope (HST). The primary objective was automatically classifying galactic sources into distinct morphological classes, which would aid future spectroscopic studies and theoretical advancements in understanding galaxy evolution. The initial phase involved replicating Morpheus findings on GOODS-S data, validating the model’s reliability. Subsequent testing on UNCOVER data, particularly utilizing the JWST F444W band, demonstrated the effectiveness in segmenting and classifying high-redshift galaxies with an accuracy of 89.21%, although challenges arose in deblending closely situated sources. To enhance efficiency, a novel model, termed BigMorpheus, was developed, incorporating an attention mechanism that significantly improved the processing speed with an accuracy of 85%. This research highlights the potential of machine learning models, such as BigMorpheus, which can not only classify galaxies but also segment astronomical images and overcome the scalability and efficiency challenges posed by the impending influx of astronomical data from advanced telescopes.
