Skip to content

Master Thesis - Landslide Pipeline

TL;DR: Landslide Pipeline is a tool that lets you select a place and time and run a trained Machine Learning model to identify landslides using satellite images. It is designed to also be accessible for non-developers. 📡 🗺

GitHub

Info

This project was completed as a Master's thesis supervised by Jakob Klotz. The post was written by the student, Armin Muzaferovic.

What is Landslide Pipeline?

In short, the repo contains all the necessary code to select a time and place using a GUI to then be able to detect landslides with the selected parameters. The U-Net model in the background was trained using the landslide4sense baseline model and the provided dataset. GUI was done using CustomTKinter as well as TkinterMap.

How does it work?

When selecting the area on the provided map in the GUI, a request is sent to the ESA Copernicus API with the corresponding parameters. In the query to the API, 12 Layers of the Sentinel2-L1C Satellite and the Copernicus30 DEM Data are requested. After receiving the data and calculating the slope from the digital elevation model, the data is formatted for use with the machine learning model. After user confirmation, the data for each pixel is evaluated to check if there is a landslide detected or not.

Why Landslide Pipeline?

While there are already great models for landslide detection, the next step is to make it easier to generate training data or to check whether a specific region has experienced a landslide. The goal of this implementation was to simplify future research in the field of machine learning with satellite data, making it easier to use available datasets. For example, city administrations could use this tool to check their areas for possible landslide risks.

How to use it?

Simply visit GitHub - muzaferovarmin/LandslideMA and either download the precompiled executable file for your OS or clone the repo and build the Python code yourself! After that, use your free Copernicus Dataspace account (which can be created here) to generate your API credentials to use and you are ready to go!