Final year (Senior year) project during my undergraduate degree
Classification of galaxies into 37 classes falling under 11 questions.
Inspired by this paper: https://arxiv.org/abs/1503.07077
Kaggle Competition: https://www.kaggle.com/c/galaxy-zoo-the-galaxy-challenge
The orientation of the galaxy in a image should not affect the classification. Same image is rotated to generate multiple samples, which will eliminated the bias caused by its rotational orientation.
Majority of pixels in every image are dark, add no value to the features, we can localize the object in each image to increase the information/pixels ratio.