Authors :
Aditi Ravishankar; Aishwarya K Karanth; Ameena, Adhishreya P; Apoorva N Hegde
Volume/Issue :
Volume 6 - 2021, Issue 7 - July
Google Scholar :
http://bitly.ws/9nMw
Scribd :
https://bit.ly/3yXFNjQ
Abstract :
There are millions of huge collections of stars,
gas, dust and stellar remnants all held together by gravity
in our vast universe. These collections, or galaxies, help in
deciphering the structure and history of the universe in
general. The classification of these galaxies based on morphological parameters is a relevant requirement in understanding their formation and evolution. Manual identification of the categories to which each belongs to can be tiresome, time consuming and error prone. The objective of
our work was to automate the process of finding the features that characterize a galaxy using convolutional neural
networks, a cardinal concept in the image data space,
whilst comparing the accuracy of the classification with
and without prior processing of the image dataset. The
Galaxy Zoo dataset was used for the same and it was preprocessed by applying median filtering and contrast limited adaptive histogram equalization. The final classification model was a CNN based on the VGG-16 architecture
with some modifications. We considered all 37 features as
per the decision tree by Willet et. al. and with a multiregression approach, obtained a model with a validation
loss of 0.0102 (mean square error) on processed images as
the best performing model. The model was then deployed
onto a client-side interface using Flask to predict the features of the galaxies in real-time
Keywords :
Deep Learning, Image Processing, Convolutional Neural Networks
There are millions of huge collections of stars,
gas, dust and stellar remnants all held together by gravity
in our vast universe. These collections, or galaxies, help in
deciphering the structure and history of the universe in
general. The classification of these galaxies based on morphological parameters is a relevant requirement in understanding their formation and evolution. Manual identification of the categories to which each belongs to can be tiresome, time consuming and error prone. The objective of
our work was to automate the process of finding the features that characterize a galaxy using convolutional neural
networks, a cardinal concept in the image data space,
whilst comparing the accuracy of the classification with
and without prior processing of the image dataset. The
Galaxy Zoo dataset was used for the same and it was preprocessed by applying median filtering and contrast limited adaptive histogram equalization. The final classification model was a CNN based on the VGG-16 architecture
with some modifications. We considered all 37 features as
per the decision tree by Willet et. al. and with a multiregression approach, obtained a model with a validation
loss of 0.0102 (mean square error) on processed images as
the best performing model. The model was then deployed
onto a client-side interface using Flask to predict the features of the galaxies in real-time
Keywords :
Deep Learning, Image Processing, Convolutional Neural Networks
