Faint stellar structure detection

Faint stellar structure detection was part of his work as a research assistant at NARIT in 2026. Machine-learning models built with AutoGluon, run on HPC and PBS job orchestration infrastructure, find faint structures of stars around galaxies, such as streams and shells left behind by past mergers. These structures are hard to see because they hold few stars spread over a wide area.

The M81 and M82 workflow

The method is documented as a public method ledger, KWS–ASTRO–001, in KiWiS. It starts from resolved-star catalogues from the Hubble Space Telescope for the galaxies M81 and M82:

1. Prepare the catalogues, keeping each telescope field's geometry so every star can be traced back to its source.
2. Remove duplicates where neighbouring fields overlap, so the same star isn't counted twice and doesn't inflate the density.
3. Aggregate spatially only then, to look for faint over-densities.

Uneven telescope coverage is shown explicitly on the field maps before any structure is interpreted. Each figure links to its original in the research archive.

Status

The results are not published yet: a paper on this work is in preparation. Until it is out, this page describes the method only.

See also

• Galactic merger-history clustering
• Figure archive
• Machine learning and data