Machine Learning Applications in Galaxy Parameter Evaluation

The structural parameters of galaxies are the foundation for a deeper understanding of their formation and evolution. Existing galaxy fitting software has limitations such as slow running speed, heavy reliance on manual operation, and sensitivity to initial values. Based on the data from the Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP) survey, we constructed a training and testing dataset. The software GaLNet (Galaxy Light Profile Convolutional Neural Network), developed based on Convolutional Neural Networks (CNN), was applied to implement an algorithm for measuring galaxy structural parameters based on the Sérsic formula. By applying it to real HSC-SSP data and comparing it with the traditional method GALFIT, the average reduced chi-square value of the fitting results was reduced by 37%, and it has clear advantages in accuracy and running speed.