Alberto Morea
Sessions
Since the deployment of the first satellite equipped with a Synthetic Aperture Radar (SAR) into orbit in 1978, the use of SAR imagery has been a vital part of several scientific domains, including environmental monitoring, early warning systems, and public safety.
SAR could be described as "non-literal imaging" since the raw data does not resemble an optical image and is incomprehensible to humans.
For this reason, raw data is typically processed to create a Single Look Complex (SLC) image, which is a high-resolution image of the scene being observed. The processing of raw data to create a SLC image involves several steps, including range compression, Doppler centroid estimation and azimuth compression.
Processing raw data requires a significant amount of computer power; as a result, it is almost never practical to do it on board. As a direct consequence, the data is transmitted back to Earth to be processed.
The objective of next-generation studies [1] is to optimize Earth Observation (EO) data processing and image creation in order to deliver EO products to the end user with very low latency using a combination of advancements in the on-board parts of the data chain.
In this work, we focus on a sea scenario and propose to eliminate any pre-processing by training a Deep Convolutional Neural Network (DCNN) to directly recognize bright targets on raw data.
This indeed might substantially shorten the delivery time thus improving the efficiency of satellite-based maritime monitoring services.
In this regard, the availability of training data represents one of the critical issues for the development of machine learning algorithms. In fact, the efficacy of the final machine learning-powered solution for a specific application is ultimately determined by the quality and amount of the training data.
However, to date, there are no training SAR raw data available in scientific literature with regard to the specific topic of sea scenario monitoring. Furthermore, their generation from real data is a time-consuming task.
In this work we propose and investigate physically and statistically based approaches to simulate a marine scenario and generate realistic synthetic training SAR raw datasets.
We then trained and evaluated a state-of-the-art DCNN on the generated synthetic dataset and successively on real raw data extracted from ERS imagery archive. It is one of the first
experiments proposed in the SAR literature and results are quite encouraging, as they reveal that a well-trained DCNN can correctly recognize strong scattering objects on SAR raw data.
[1] M. Kerr, et al. “EO-ALERT: a novel architecture for the next generation of earth observation satellites supporting rapid civil alerts”, in 71st International Astronautical Congress (IAC), 2020.
Acknowledgments
This work was carried out in the framework of the APP4AD project (“Advanced Payload data Processing for Autonomy & Decision”, Bando ASI “Tecnologie Abilitanti Trasversali”, Codice Unico di Progetto F95F21000020005), funded by the Italian Space Agency (ASI). ERS data are provided by the European Space Agency (ESA).
Multi-temporal SAR Interferometry (MTInSAR) techniques allow detecting and monitoring millimetric displacements occurring on selected point targets that exhibit coherent radar backscattering properties over time. Successful applications to different geophysical phenomena have been already demonstrated in literature. New application opportunities have emerged in the last years thanks to the greater data availability offered by recent launches of radar satellites, and the improved capabilities of the new space radar sensors in terms of both resolution and revisit time. Currently, different space-borne Synthetic Aperture Radar (SAR) missions are operational, e.g. the Italian COSMO-SkyMed (CSK) constellation and the Copernicus Sentinel-1 (S1) mission.
Each CSK satellite is equipped with an X-band SAR sensor that acquires data with high spatial resolution (3x3 m2), thus leading to a very high spatial density of the measurable targets and allowing the monitoring of very local scale events. Thanks to the nationwide acquisition plan “MapItaly”, CSK constellation covers the Italian territory with a best effort revisit time of 16 days since 2010.
S1 mission is instead operational since 2014 and acquires in C-band at medium resolution (5x20 m2) with a minimum revisit time of 12 days (only 6 days between 2016 and 2021, when the full S1 constellation was operational), thus allowing to monitor ground instabilities back in time almost all over the Earth. Moreover, all data acquired by the S1 mission are provided on an open and free basis by the European Space Agency (ESA) and the European Commission (EC), for promoting full utilization of S1 data, with the aim of increasing the scientific research, growing the EO markets and fostering the development of continuous monitoring services, such as the European Ground Motion Service (EGMS) and the Rheticus® Displacement Geo-information Service.
The EGMS is based on the MTInSAR analysis of S1 radar images at full resolution, updated annually, and provides consistent and reliable information regarding natural and anthropogenic ground motion over the Copernicus Participating States and across national borders.
Rheticus® offers monthly updates of the millimetric displacements of the ground surface, through the MTInSAR processing chain based on the SPINUA© algorithm (“Stable Point Interferometry even in Un-urbanized Areas”). Rheticus® is capable to process SAR images acquired by different SAR missions, including CSK and S1. Thanks to the technological maturity as well as to the wide availability of SAR data, these ground motion services can be used to support systems devoted to environmental monitoring and risk management. This work shows the results obtained in the framework of the SeVaRA project (“Environmental Risk Assessment Service”), coordinated by Omnitech srl. The goal of SeVaRA is to implement an innovative system for calculating an aggregate environmental risk index, derived from several parameters related to hydrogeological instability phenomena and/or Weather-related extreme events. In particular, the present work is focused on the analysis of the “Deformation Sub-System”, that has been designed for the computation of risk indices related to structural and ground instabilities (landslides). The first step consists in the Hazard Map computation, which requires the following input data:
- Susceptibility Map (i.e., the European Landslide Susceptibility Map, provided by the Joint Research Centre European Soil Data Centre)
- National mosaic of landslide hazard zones, provided by ISPRA (River Basin Plans PAI)
- Cumulated precipitations (derived by cumulating ground measurement data collected by weather stations, if available, or by interpolating hourly rainfall data provided by the Global Satellite Mapping of Precipitation service, GSMaP, offered by the JAXA Global Rainfall Watch)
- Land Cover Change (i.e., the CORINE Land Cover inventory)
- Seismic events inventory, provided by INGV, to account for earthquake-induced landslides
- MTInSAR ground displacement time series.
The last input is essential for detecting instable areas, whose MTInSAR displacement trend exhibits a significant velocity in the whole observation period and/or an acceleration in the acquisition dates of the last year. The SeVaRA “Deformation Sub-System” has been primarily designed to be interfaced with the Rheticus® Displacement Service, but it supports also products offered by the EGMS service as well as by other MTInSAR services available on the EO market. The final step consists in the computation of the landslide risk index, obtained by combining the previous hazard index with the vulnerability and the exposure of the area of interest. The results of this study over specific areas of interest will be presented and commented.
Acknowledgments
Study carried out in the framework of the SeVaRA project, funded by Apulia Region (PO FESR 2014/2020).
This study presents a novel approach to monitor oil spills and ships using Synthetic Aperture Radar (SAR) raw data and deep learning techniques. The proposed methodology involves several steps including pre-processing (focusing, filtering and land sea mask), semantic segmentation, and classification using a deep convolutional neural network (DCNN) model, as well as real-time (FFT-based) processing to ensure a fast response.
To train the DCNN model, the study combined three datasets: CleanSeaNet, TenGeoP-SARwv, and GAP_OilSpill_DB. The first two datasets are publicly available, while the third dataset was specifically built by the authors by integrating known and documented case studies from news articles and cases identified in the sea area in front of the port of Brindisi (Southern Italy), internally validated by expert GAP operators.
Data augmentation techniques were also utilized to improve the model's performance by generating additional training data. The DCNN model uses DeepLab v3+ based on ResNet-18 and is trained on a large dataset of SAR images that includes various types of oil spills, look-alikes, novelty objects, and ships.
The proposed system is optimized to process data on board the satellite to ensure a real-time response. The system transmits images to the ground segment only if there is an event of interest (e.g. a novelty object or an oil spill detected eventually involving the nearest ships).
The study demonstrates that the proposed approach provides a promising solution for real-time monitoring of oil spills, ships and novelty objects using satellite SAR raw data. The use of deep learning and data augmentation techniques can significantly improve the accuracy and speed of detection, which can ultimately lead to better environmental management and oil spill response. .Additionally, the proposed approach can be applied to a variety of SAR datasets and has the potential to be integrated with existing oil spill response systems.
Acknowledgments
This work was carried out in the framework of the APP4AD project (“Advanced Payload data Processing for Autonomy & Decision”, Bando ASI “Tecnologie Abilitanti Trasversali”, Codice Unico di Progetto F95F21000020005), funded by the Italian Space Agency (ASI). ERS, ENVISAT and Sentinel-1 data are provided by the European Space Agency (ESA).