Interpretable Clustering of PS-InSAR Time Series for Ground Deformation Detection

This is the repo for the paper "Interpretable Clustering of PS-InSAR Time Series for Ground Deformation Detection" by Claudia Masciulli, Giacomo Guiduzzi, Donato Tiano, Marta Zocchi, Francesco Guerra, Paolo Mazzanti and Gabriele Scarascia Mugnozza.

The repo is structured as follows:

• data: contains the data used in the paper
• src: contains the source code
• time2feat: contains the code for the Time2Feat model
• featts: contains the code for the FeaTTS model
• jqm_cvi: contains the code for the JQM-CVI package, which implements the Dunn index for clustering evaluation among other metrics.

Both models are available as Python packages on PyPI and are configured as git submodules in this repo.

The suggested way to prepare the code is:

1. Run git submodule init --recursive to initialize time2feat, FeatTS and JQM-CVI submodules.
   • Alternatively, you can clone the repo with git clone --recurse-submodules
2. Install the packages contained in the requirements.txt file into a Python venv or conda environment
3. Move into the jqm_cvi folder and run python setup.py install to install the JQM-CVI package for the Dunn index
4. Move back into the repo folder and export PYTHONPATH=/path/to/Interpretable_PS-InSAR_Clustering to let Python see the time2feat and FeatTS folders as Python modules; finally run python src/simple_pipeline.py from inside the repo folder.
   • Alternatively, directly run PYTHONPATH=/path/to/Interpretable_PS-InSAR_Clustering python src/simple_pipeline.py from inside the repo folder.

For any need related to the code, feel free to open a GitHub issue or email me at: giacomo.guiduzzi at unimore.it.