Galaxy Morphology Classification UsingWavemix

WaveMix is a novel, efficient neural network architecture leveraging multi-level 2Ddiscrete wavelet transforms (2D-DWT) to achieve state-of-the-art performance in computer visiontasks with fewer parameters and lower computational cost. By incorporating image priors like scaleinvariance, shift-invariance, and edge sparsity, WaveMix enables scalable, memory-efficientprocessing while preserving input resolution. This study applies WaveMix to galaxy morphologyclassification using pre-trained weights from the Galaxy10 DECals dataset and evaluatesperformance on the Galaxy10 SDSS dataset. With an ensemble and majority voting strategy,WaveMix achieves 89% classification accuracy, outperforming recent methods and demonstratingits potential for large-scale astronomical image analysis.