Interferometric Synthetic Aperture Radar (InSAR) is a satellite-based imaging technique which has been used to learn about earth’s surface and sub-surface movements. It can measure earths displacements by comparing phase information from the SAR images taken at different points in time. But, due to high level of noises, the wrapped phases are distorted, because of which, it cannot be used for its intended purposes. Understanding the coherence of the images becomes rather important in this situation to denoise the image and extract useful information from it. In our project, we focus on applying Super Pixeling using both traditional Computer Vision methods as well as Graph Neural Networks to understand how better we can estimate image coherence between two SAR images. By applying Super pixeling in two stages, using specific clustering techniques, we would be able to expose non-local similarities and measure the coherence from a non-local perspective.

InSAR-Starter-Kit

1. Download sample data from the google drive
2. Save the data to you local directory - $PAHT_DIR$
3. Run sample code via

python example.py --ddir $PATH_DIR$