[Image Credits:Youtube]

Arun Aniyan and Kshitij Thorat
SKA South Africa & Rhodes University
[email protected]

This is repository contains the Code and Model to do the morphological classification of radio galaxies with deep convolutional neural network.

If you make use of the code and models, please cite it as

Arun Aniyan, & Kshitij Thorat. (2017). ratt-ru/toothless: First release of toothless [Data set]. Astrophysical Journal Supplement Series. Zenodo. http://doi.org/10.5281/zenodo.579637

The model is implemented with the popular deep learning package Caffe.

Before you run the code the following software requirements need to met :

• Python 2.7.x
• Caffe
• PIL
• Astropy
• Photoutils
• Scikit-image

The repository contains the following folders required for executing the code

• Models - Contains the trained caffe models
• Prototxt - Contains the network structure for deployment
• Labels - Contains the labels
• Sample-Images - Contains image to test the model

The sample image folder contains few example images in fits format. The code also accepts preprocessed images in png or jpg format.

Please download the models from these links :

• FRI vs FRII
• FRI vs Bent-Tail
• FRII vs Bent-Tail

The downloaded model files have to be copied to the Model directory.

Once all software requirements are met the code can be run as follows with a sample image as

$ python Codes/fusion-classify.py Sample-Images/3C194.fits

The code will output the prediction with its probability and will show the total time for execution.