UchUchU
astro-ph.IMcs.CV

Object-independent evaluation of small-data astronomical image classifiers: architecture, representation, and acquisition shortcuts

Eraldo Pereira Marinho

Abstract

Reported performance in small astronomical image datasets can reflect object reuse, field-of-view differences, acquisition settings, and catalogue-derived properties as well as astronomical morphology. We designed \dataset{} to measure these effects in a four-class task comprising galaxies, globular clusters, nebulae, and open clusters. The dataset contains 370 catalogue-supported physical objects, each represented by three Digitised Sky Survey FITS bands, and uses a group-constrained split of 2…

Source:arXiv / PDF

Looking for companies in this field?

UchUchU maintains a database of Japanese companies in the space industry.