Deep galaxy classification of galaxies based on deep convolutional neural networks
1. Kafr ElSheikh Workshop – 6/12/2017
DEEP GALAXY: CLASSIFICATION OF GALAXIES BASED ON DEEP
CONVOLUTIONAL NEURAL NETWORKS
Nour Eldeen Mahmoud Khalifa
Lecturer of Information Technology – FCI, Cairo University
SRGE Member
SRGE 2017
1
Dr. Mohamed Hamed N. Taha Prof. Aboul Ella Hassanien Prof. I. M. Selim
3. Introduction
3
Galaxy
A system of millions or billions of stars, together
with gas and dust, held together by gravitational
attraction.
Kafr ElSheikh University Workshop – 6/12/2017
4. Introduction
4
How Many Galaxies
100 billion and 200 billion galaxies in the observable
Universe that human-made technology can detect.
Astronomers at the University of Nottingham now say the number of galaxies in the observable
Universe is 2 trillion, more than 10 times as many as previously thought.
Kafr ElSheikh University Workshop – 6/12/2017
5. 5
Workshop on Swarm Intelligence and Application (SIA), FCI, CU - Sat 23-09-2017
7. Introduction
7
Types of Galaxies
Elliptical, Spiral, and Irregular
Kafr ElSheikh University Workshop – 6/12/2017
8. Introduction
8
Types of Galaxies
Hubble Classification
Kafr ElSheikh University Workshop – 6/12/2017
9. Problem Statement
9
Enable Computer to classify the types of galaxies
automatically.
Why !!
Education Purposes.
Classifying billions of galaxies. (Trillions in near future)
Kafr ElSheikh University Workshop – 6/12/2017
10. Proposed Approach
10
Year Publication Accuracy
2004 Machine learning and image analysis for morphological galaxy
classification
91%
2007 A Hierarchical Model for Morphological Galaxy Classification 91.64%
2017 Galaxy Image Classification using Non-Negative Matrix
Factorization
93%
Kafr ElSheikh University Workshop – 6/12/2017
11. Proposed Approach
11
Deep Learning “Convolutional Neural Networks”
Second Appearance.
The Rise of Deep Learning.
Why now !!
More Powerful Gpu’s and Cpu’s
Large Datasets.
Kafr ElSheikh University Workshop – 6/12/2017
16. Experimental Results
16
Dataset
This catalog dataset consists of more than 11,000
images and contains samples from different Hubble
types galaxies.
Trained over 1346 image only.
Kafr ElSheikh University Workshop – 6/12/2017
17. Experimental Results
17
Implemented using a software package (MATLAB).
The implementation was CPU specific. All
experiments were conducted on a server with Intel
Xeon E5-2620 processor (2 GHz) and 96 GB Ram.
Kafr ElSheikh University Workshop – 6/12/2017
21. Experimental Results
21
Year Publication Accuracy
2004 Machine learning and image analysis for morphological galaxy
classification
91%
2007 A Hierarchical Model for Morphological Galaxy Classification 91.64%
2017 Galaxy Image Classification using Non-Negative Matrix
Factorization
93%
2017 Deep Galaxy: Classification of Galaxies based on Deep
Convolutional Neural Networks
97.272%
Kafr ElSheikh University Workshop – 6/12/2017