Unsupervised semantic analysis

This repository provides the Python code to reproduce the computational analysis of the paper: "Large-scale unsupervised spatio-temporal semantic analysis of vast regions from satellite images sequences" (Echegoyen et al., 2024).

Table of contents

• Code
• Data and models

Code

We provide python notebooks with experiments of the paper in the directory experiments.

The directory src contains the Tile2Vec code with some mofications and additional functions. As indicated by the authors, the Tile2Vec LICENSE is included in this repository.

The following notebooks must be run in this order:

1. Create_grid_MTS: loads a model and creates a collection of multivariate time series by embedding sequnces of tiles.
2. Experiment_1: runs and plots the clustering of time series in differente ways, calculates qualitiy measures and explores the semantic provided by the clustering

Data and models

An example of sequence with 3 Sentinel-2 images and the embedding models trained for the paper can be dowolad here. The models should be in the directory models and the sequence of satellite images in the directory data/NE-TXN. The rest of images are not provided here due to space constraints, but they can be shared upon request.

Note that the results of these illustrative examples correspond to a sequence of 3 images. In the paper, we analyze a region covered by 4 images and use sequences of 20 images.

Acknowledgements

This work has been supported by Project PID2020-113125RB-I00/MCIN/AEI/10.130 39/501100011033. Aritz Pérez has been supported by Basque Government through the Elkartek program and the BERC 2022-2025 program, and by the Ministry of Science and Innovation: BCAM Severo Ochoa accreditation CEX2021-001142-S/ MICIN/ AEI/ 10.13039/ 501100011033.

Reference

C. Echegoyen, A. Pérez, G. Santafé, U. Pérez-Goya and M.D Ugarte. Large-scale unsupervised spatio-temporal semantic analysis of vast regions from satellite images sequences. Statistics and Computing 34, 71 (2024).