MATTCH - TRE ALTAMIRA POLIMI ESA

Mattch
Machine Learning Methods for SAR-derived Time Series Trend Change Detection

TRE ALTAMIRA
POLITECNICO DI MILANO
ESA

Expand knowledge about Machine Learning techniques applied to Earth Observation (EO) big data.

MATTCH PROJECT

Using ESA Sentinel-1 data, we envisage the creation of innovative solutions, spurring new services to end-users while contributing to the EO market growth

The MATTCH project aims to apply Machine Learning (ML) techniques to advanced InSAR (Interferometric Synthetic Aperture Radar) analyses over large areas, with the goal of identifying among the huge number of InSAR measurements the ones exhibiting a change in their motion trend or, more generally, an “anomalous behaviour”. This “data screening and data mining” step is extremely important to support the End Users Community in the exploitation of frequently updated (every few days) and highly populated (millions of measurement points) information layers.

EO DATA ACCESS

EO DATA ACCESS
Free access to Sentinel-1 images. Every 6 days, acquired worldwide.

freccia
PROCESSING

PROCESSING
SqueeSAR® processing chain on a cloud computing base.

freccia
MATTCH

MATTCH
Deep Learning architectures applied to the processing chain.

freccia
USERS

USERS
“Hotspots” showing a change in the dynamic of motion.

Achievements

Driven by the link between TRE Altamira and the Final Users community, Polimi-DEIB has developed a Machine Learning approach, which starts from a Recurrent Neural Network architecture to build Long Short-Term Memory integrated layers.

The prototype has been engineered and ingested into TRE-A’s processing environment on a cloud-based architecture, demonstrating to be comparable in terms of results - detection of measurement points exhibiting changes in trend - and significantly better - in terms of computational cost - with respect to other statistics-based approaches in use.

As a final step, the ML method has been applied to real data. This is the case of the Tuscany Region monitoring project using Sentinel-1 acquisitions, in continuous update. The “new” results were delivered to the Final User (University of Florence, Earth Science Department) and their feedback allowed a fine tuning of the algorithm parameters.

Schematic of SqueeSAR® monitoring using Sentinel-1 data. After a baseline processing at time T0, using all the images available from the archive, a new image is acquired every 6 days, triggering an update of SqueeSAR® results. The updated deformation data - each time consisting of hundreds of thousands/millions of measurement points (MP) - are delivered to the End User. Each MP is described by a displacement time series, with a new sample after each update (red dots in the diagrams on the right).
Sentinel-1 Descending SqueeSAR® results over Tuscany - 03/03/2020 delivery - on the left. MATTCH results on the right: the red dots represent the 145 hotspots exhibiting a trend change, out of more than 800,000 measurement points.
MATTCH results computed on the 03/03/2020 delivery of the TOSCANA project: examples of time series detected by MATTCH.
ESA

TRE ALTAMIRA

TRE ALTAMIRA website

TRE ALTAMIRA is the largest InSAR group worldwide. With over 19 years’ experience, we are globally recognised as leader in measuring ground motion from space with millimetre precision. We provide displacement measurements and mapping solutions from satellite radar (SAR) data that are used in a variety of sectors, including oil & gas, mining, civil engineering and geohazards. Founded in 2000 (TRE) and 1999 (ALTAMIRA), we bring together a multidisciplinary team with excellent knowledge of SAR and InSAR applications. Our core competence is radar imagery processing and we are the reference InSAR provider for NASA (National Aeronautics and Space Administration), CSA (Canadian Space Agency), ESA (European Space Agency), JAXA (Japan Aerospace Exploration Agency), CNES (French Space Agency), DLR (German Centre for Aeronautics and Space) and the World Bank.

POLITECNICO DI MILANO

POLIMI DEIB website

Founded in 1863, Politecnico di Milano (POLIMI) is the largest school of architecture, design and engineering in Italy. Its Dipartimento di Elettronica, Informazione e Bioingengeria (DEIB) is one of the largest European ICT departments. Research is the main focus of DEIB, pursued according to the highest international quality standards. The six department sections cluster consolidated competences in systems and control, computer science and engineering, electronics, telecommunications, bioengineering and electrical engineering. They have a broad network of partnerships with the best international institutions, which makes the Department one of the fundamental players in the worldwide scenario of scientific and technological innovation. DEIB is participating in the project with researchers and professors from the Artificial Intelligence and Robotics Laboratory (AIRLab).

EXTRAS

Recurrent Neural Networks for Trend Change Detection in InSAR Time SeriesPDF
Presented during the Earth Observation Phi-Week, 9-13/09/2019, ESA-ESRIN Frascati (Rome), and ASAR 2019 organized by CSA, 1-3/10/2019, Saint-Hubert, Quebec (Canada)

Authors: Francesco Lattari (1), Matteo Matteucci (1), Alessio Rucci (2), Christine Bischoff (2), Marco Basilico (2), Emanuele Passera, Marco Bianchi (2) │ (1) Politecnico di Milano, Milano, Italy, (2) TRE ALTAMIRA s.r.l., Milano, Italy

Download