Galaxy Classification Using Dilated-Separable Convolution Neural Network

Galaxy morphology is an essential aspect to take into account when studying the formation and evolution of cosmic galaxies. Galaxy morphology classification using a convolutional neural network based on a deep learning framework is made feasible by the rapid growth of deep learning and artificial intelligence technology. The classification of galaxy images of Galaxy Zoo 2 dataset into five main categories-cigar, edge-on, in-between, spiral, and round etc., has been done in this research using a novel deep learning-based method. Performance of the all the deep learning models evaluated using precision, recall, f-1 score, and accuracy. Proposed dilation-separable convolution neural network model successfully classified the images with an accuracy of 89.74 %, which outperformed conventional approaches such as ResNet50, DenseNet121, DenseNet169, MobileNet, MobileNetV2, MobileNetV3 etc. with less number of parameters of 80,469 and floating point operations of 1.94 G. In future work, performance of proposed lightweight model will be trained on large number of samples.