An Effective Galaxy Classification Using Fractal Analysis and Neural Network

Abstract

Astronomy is always in a quest of revealing the mysteries of our Universe. There is a vast amount of astronomical data collected and this information comes from stars, galaxies and other celestial objects. While exploring this type of astronomical data, we can identify some complex self-similar patterns. Such self-similar patterns are shown in our own galaxy and are called fractals. This research work has been developed for finding such self-similarity that can be measured from galaxy clusters and this feature can be learned through a suitable neural network. This research work gives an insight about calculating the fractal dimension of galaxy images using a box counting algorithm and training the images using LeNet - 5. The box counting fractal dimension is a specified range of values for each particular class of galaxy. By using the fractal dimension as a primary feature of different classes of galaxy and with the help of LeNet-5 network model classifying the galaxy images into ten specified classes according to its morphological properties. The model produced an accuracy of 74% when implemented with the baseline algorithm. When implemented with LeNet- 5 it produced an accuracy of 96%. The precision recall and f1-Score value of the LeNet-5 model was also calculated. The precision recall and f1-Score value for class 1, class 2, class 4 and class 6 were higher than those of the other classes.