In the quantum phase transitions, an accurate representation from single-shot experimental momentum-space density images of ultracold quantum gases can be one challenge task due to the noisy images which make data extraction difficult. Introducing some deep learning techniques to denoise the experimental image, mapping two-dimensional phase diagram of Haldane model has shown improvement in attaining accurate results that is however impossible to achieve with conventional methods. In this study two different machine learning models are implemented and compared to study their accuracies using theoretical results as a reference.