Abstract：

This report will explore the applications of machine learning in photo-z estimation, galaxy image enhancement, and galaxy morphology classification. In the area of photo-z estimation, we will introduce a series of studies that demonstrate how machine learning models can accurately estimate the photo-zs of galaxies by analyzing photometric data. This includes improving traditional machine learning methods, developing photo-z templates suitable for CSST observations, and a novel method for imputing missing data, all of which significantly enhance the accuracy of redshift estimation.

Regarding galaxy image enhancement, we propose the Pix2WGAN hybrid deep learning model to achieve high-quality transformation of cross-survey astronomical images. This model can upgrade SDSS and DECaLS images to a level close to that of HSC images, significantly enhancing the ability to recognize complex astronomical structures such as galactic spiral arms and tidal tails.

In the field of galaxy morphology classification, addressing the issue of traditional methods' reliance on large amounts of labeled data, we introduce the GC-SWGAN semi-supervised deep learning model. This model, trained with limited labeled data and abundant unlabeled data, achieves good classification results and significantly reduces the need for labeled data. It has been successfully applied to detect special structures in galaxies, such as ring structures.