Practice transfer learning on galaxy classification models to try and predict already labeled galaxies. This is using the Galaxy 10 dataset
Galaxy10 is a dataset contains 21785, 69x69 pixels colored galaxy images (g, r and i band) separated in 10 classes. Galaxy10 images come from Sloan Digital Sky Survey and labels come from Galaxy Zoo. The dataset has columns: images with shape (25753, 69, 69, 3), ans, ra, dec, redshift, and pxscale (unit of arcsecond per pixel)
Models included:
- CNN without TL
- VGG 16
- VGG 19
- ResNet50
Contributors: Andrew Pascual, Savanna Guertin
This was a project in order to test different types of neural network models and compare their results. During the project we wanted to compare a CNN model without transfer learning, VGG 16, VGG 19, and ResNet50. Moreover, we wanted to contribute to the issue of tedious manual galaxy recognition by using machine learning these machine learning models.
During the testing we found that the models with transfer learning (VGG 16, VGG 19, and ResNet50) performed much better than the base CNN model without transfer learning. Additionally, the best model when comparing the four came to be the VGG 16 model with transfer learning.
