My implementation of using the UNET model to identify buildings in satellite imagery:
For now the code remains mostly undocumented, but if you are greatly interested, the project report that I have written for the development and methodology behind my model for my PHYS381 lab course is included.
This project was started for developing a method of quantifying urban expansion of cities. This will continue in development in the near future, as a deliverable software package to end-users.
Some overviews and quick guide to understanding a confusion matrix: The code for the model itself is in UNET.ipynb. The trained model is in the model.h5 file. Other files in this repo was used as utilities for pre-processing model input. The paper outlining the model and methodology is titled PHYS381_project_report.pdf.