Orateur
Description
Galaxy clusters are powerful and key cosmological probes that trace the evolution of large-scale structures of the universe and inform models of cosmic evolution, but their identification in the increasingly large imaging datasets produced by current and upcoming surveys requires detection methods that are both accurate and scalable. The Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) will provide deep, wide-field optical imaging over an unprecedented survey area, enabling the detection of thousands of galaxy clusters up to high redshifts and motivating the development of new approaches for identifying galaxy clusters directly from astronomical images.
Building upon the YOLO-CL framework, originally based on YOLOv3, I have developed an updated galaxy-cluster detection pipeline using the more recent YOLOv10 object-detection architecture. The model detects and localizes galaxy clusters directly in multiband optical images, without requiring a preliminary galaxy catalog. Applied to Sloan Digital Sky Survey (SDSS) imaging using redMaPPer clusters as reference detections, the method achieves high completeness and purity, demonstrating the effectiveness of modern object-detection architectures for this task.
In parallel, I am investigating the incorporation of astrophysical prior information through class-activation-map-based auxiliary supervision, encouraging the network to focus on physically relevant galaxy populations and providing greater interpretability of its detections.
This effort aims to contribute a scalable and interpretable, high-fidelity cluster detection pipeline for next-generation cosmological analyses and to assess its robustness as we move from controlled cluster samples toward realistic next-generation survey data.