Unsupervised Images Generation Based on Sloan Digital Sky Survey with Deep Convolutional Generative Neural Networks

Convolution neural networks (CNN) has gain huge successful in computer vision applications and image classification in recent years. However, unsupervised learning with CNN attracts less attention than supervised learning. In this work, we use deep convolutional generative adversarial neural networks (DCGANs) to generate images. The model is trained by various star and galaxy images from Sloan Digital Sky Survey. In this paper, we also took a gird search to choose the best value for hyper-parameters and several methods to help to stabilize the training process and promise a good quality of the output. Based on several experiments, we demonstrate that our approach keeps the training procedure stable and the generator and discriminator are both good for unsupervised leaning.