A Multi-Stage Deep Learning Approach for Stellar Mass Prediction of Galaxies

In astronomical research, the stellar mass of galaxies is a crucial physical parameter that directly influences a galaxy's luminosity, dynamical properties, and evolutionary process. Tra-ditional methods for calculating stellar mass rely on spectral data, which require complex techniques to obtain and limit their broader application. With advancements in astronomical observation technology and deep learning, it has become possible to estimate galaxy parameters from photometric images using deep learning. This paper proposes a multi-band deep learning approach, integrating the improved Rep Vit framework with the CBAM attention mechanism and a lightweight Stem module, and designing a new Head module. We introduce the Stellar Mass Prediction Network, Multi-Stage Attention-Enhanced Convolutional Network (MACNet), which effectively captures complex features in galaxy images. Experimental results validate the superior performance of this method for stellar mass prediction, demonstrating that the proposed approach outperforms existing methods and provides new possibilities for predicting galaxy parameters.