Searching for low-surface-brightness galaxies with compact neural networks A parameter-efficient approach to first-pass selection of diffuse galaxy in HSC-SSP imaging

Abstract

Low-surface-brightness galaxies (LSBGs) trace a diffuse and observationally challenging regime of galaxy formation that remains highly susceptible to selection effects in optical surveys. We present a hybrid LSBG detection pipeline that integrates compact convolutional neural networks (cCNs) as a morphological validation stage within a multi-step detection framework applied to HSC-SSP imaging. Permissive low-threshold source detection with SEP is followed by cCN-based morphological filtering, physically motivated consistency checks, and parametric surface-brightness modelling with galfitm. The cCN is deliberately constrained to ${\sim}10^4$ trainable parameters to match the spatial scales of diffuse emission and ensure computational efficiency at survey scale. The cCN reduces the initial detection set of ${\sim}77\times10^6$ objects by more than four orders of magnitude while recovering essentially all literature LSBGs that pass the detection stage. The resulting Gold catalog of 5156 candidates occupies the characteristic loci of LSBGs in colour-colour and structural parameter space, in close agreement with previously published HSC-SSP based samples. Cross-matching with SDSS and DESI spectroscopic redshifts confirms that the pipeline identifies genuine nearby diffuse galaxies. Compact neural networks provide a robust and computationally scalable morphological filter for LSBG searches, effectively bridging classical detection pipelines and large general-purpose deep architectures. The pipeline is detection-limited rather than selection-limited, identifying the initial source extraction stage as the primary target for future improvement.