Analyzing and Processing of Astronomical Images using Deep Learning Techniques

Deep Learning techniques are widely used in various use cases like image detection, pattern recognition, computer vision, and prediction, etc. These days, Convolutional neural networks(CNN) which is an efficient algorithm of Deep Learning are also extensively used in astronomical image processing. In this work, we implemented different architectural models like AlexNet, VGG16, ResNe50, InceptionV3,Xception to classify and predict the Redshift(z) values of unlabeled galaxy images taken from the EFIGI catalog and galaxy-zoo image dataset from the Kaggle website. In addition to these pre-built architectural models, we have introduced a novel, customized CNN classifier and Redshift(z) predictor models to study the behavior of CNN layers and to achieve reasonable accuracy by fine-tuning hyperparameters. Our customized CNN Classifier model achieved a considerably good accuracy of 92.3% with 87.3% validation loss in galaxy classification. Whereas in Redshift(z) prediction, our novel CNN Redshift(z) predictor model achieved a very low loss of 0.000158 when compared to other pre-built architectures.