DeepSnow is the winning project for the challenge provided by TechnoAlpin for the Hackaton Hack the alps. We were asked to detect a ski slope by analyzing a satellite image.

We gathered data from various sources, such as: NASA, OpenStreetMaps, OpenSnowMap, OpenData. Then, we used a deep learning framework called U-net, using Keras. After that, we used opencv to create a polygon to calculate the area of the slope. We created a simple web-page using React as front-end, where the user can click on what he thinks is a slope: the screenshot of the surrounding area is sent to the back-end (made with Flask) to be analyzed. The output is visible on the map, with the polygon being drawn in the area that was screenshotted. Below you can see an example of the process:

Keras, Opencv, React and Flask.

The project consists of a backend (wepapp in Python) and frontend (ReactJS). Following steps were tested on a Ubuntu 18.04 installation.

The Python backend has a lot of dependencies. A conda environment can be easily setup by using the provided environment file. Install Anaconda Anaconda if not installed yet.

• In a Bash terminal, navigate to the webapp subdirectory
• Create a conda environment using the provided deepsnow_backend.yml file

conda env create -f deepsnow_backend.yml

• Activate the conda environment (in Bash terminal source activate deepsnow)
• Run the python backend: KERAS_BACKEND=tensorflow CUDA_VISIBLE_DEVICES="" python app.py
• Ignore the warnings about the deprecated merge function. The python backend should run now, as can be verified by looking for * Running on http://localhost:5000/ in the logs. The port should be 5000, in case it is not, remember that for later.

The frontend requires NodeJS and the NPM package manager.

• Install both ( in Bash: sudo apt update,then sudo apt install nodejs, then sudo apt install npm)
• Navigate to the frontend folder in a bash terminal
• Initialize NPM by executing npm install
• Start NPM: npm start
• The frontend should now be runnning, probably indicating Compiled with warnings in the terminal.

TODO

@niklaskappler, @thomasverelst, @agemcipe