Coherence change detection workflow for damaged buildings detection

Hello everyone,

Today, I’d like to share with you something I’ve been working on, which is a coherence change detection workflow based on Sentinel-1 SLC SAR images for damaged buildings detection.

The methodology developed for this work can be split into 2 parts. First, SLC image processing and the generation of the Coherence image stack with ESA’S SNAP and QGIS, followed by the anomaly detection in the coherence timeseries done with a python script which returns the pixels identified as “damaged buildings”. (see image below)

The workflow will compare the coherence timeseries of every pixel to a linear regresion representing the coherence of an undamaged building.
As damaged buildings have a distinct coherence timeserie caracterized by a sudden drop in coherence when the building will sustain any damaged (see image below).

I have included detailed instructions on how to reproduce my work in this pdf.
CCD_for_damaged_buildings.pdf (270.2 KB)

The python script for anomaly detection in the coherence timeseries can be found here

I am still in the process of evaluating this method, but there is room for improvement.
So far, I noticed that the buildings size, density and orientation plays a big role in detecting whether they have been damaged or not.

I can answer your questions here, or you can email me at : mourad_ikhelif@proton.me

4 Likes

Very nice application - thank you for sharing!

1 Like