Hi! I'm Alan Amaro and this is a final project that I make for a college class: TSFC-I: Inteligencia Artificial en la Fisica.
You can see three folders, the notebook's folders contain all the main code source for image analysis and processing, and the central notebook contains all the architecure for the CNN and the U-Net architectures. The Best_Model folder contains only the model in .h5 and .keras extensions, this model has done with only a simple CNN and a FFC.
The U-Net model is the 'best model' that I was able to create, this implements and U-Net architecture in .h5 and .keras extensions. The image is process by the U-Net, and later is given to the first architecture (CNN-FFC). Both models have an acurracy above 90% (U-Net has 96%)
All the data was taken for kaggle.com, in the first notebook I share the link to all data.
Have fun watching or implementing this code to your project!