The study focuses on using models that integrate deep learning and Persistent Homology to classify images of stars and galaxies obtained from po-werful telescopes. The main result of this paper is that the persistent images captured the essential information from the dataset we used. We get slightly better evaluation results, if we use just persistent images obtained from the original images instead of the original images. Also, in this paper we evaluate models that incorporate Persistent Homology and Deep Learning that we have proposed earlier. From the evaluation, we can say that usage of topological characteristics of data improves the classification.