Target Detection

The presence of icebergs represents a danger to navigation and activities in cold waters near and in the Polar Regions. With the aim of improving safe navigation, in the last decades satellite data have been used to detect and track large icebergs. Specifically, the potentials of scatterometers, altimeters and synthetic aperture radar (SAR) systems have been successfully tested. However, small icebergs are still hard to identify especially when these are embedded in sea ice.

In this work, we present the latest developments on a recently proposed iceberg detector, namely the intensity Dual-Pol Ratio Anomaly Detector (iDPolRAD) [1]. The algorithm is based on the use of incoherent (or Ground Detected) dual-pol HH and HV (or VV and VH) images. These are routinely acquired by Sentinel-1 in the Polar Regions.

The detector considers a bank of box cart filters for the HV and HH images averaging over sizes selected as test and training areas. The difference between the HV intensity over the test and the HV intensity over the training areas is divided by the intensity of HH over the training area [1]. Developing the maths, it is possible to demonstrate that the detector returns a high value if the HV intensity and the depolarisation ratio are increasing from train to test area. This happens if we have a positive anomaly (i.e. an increase) in volume or oriented multiple reflections, which forms the main scattering mechanisms of icebergs.

To validate the algorithm, we used Sentinel-1 data acquired on the East Coast of Greenland, near the Kangerlussuaq glacier where a large number of grounded icebergs are visible in the images. The grounded icebergs were used to perform a quantitative validation. The result of the analysis is that the iDPolRAD contrast between sea ice clutter and icebergs increases to about 80 times compared to the contrast of the filtered HV intensity alone.

Reference:

A. Marino, W Dierking, and C. Wesche, “A depolarization ratio anomaly detector to identify icebergs in sea ice using dual-polarization SAR images,” vol. 54, no. 9, pp. 5602–5615, 2016.

The routes of vessels participating to the yacht races around the world (Vendée Globe, Barcelona World Race, Volvo Ocean Race, Jules Vernes Trophy [1-4], and Brest Ultim in 2019 [5]) cross the southern parts of the Atlantic, Indian and Pacific oceans.

The icebergs infesting those areas are a major threat for the safety of the navigators. Iceberg monitoring is thus of vital interest to set up no-go zones, optimize routes and warn sailboats of infested areas. This paper presents the benefit of Sentinel-1 (and more generally SAR imagery) for operational iceberg monitoring service around Antarctica. We specifically provide a feedback on the last 2016-2017 Vendée Globe race

An initial situation map of icebergs was built before the race by using sparse sampled but systematic observations of altimeters (Jason 2, SARAL/Altika). For the first time, all the acquired wave mode (WM) S-1 imagettes were used jointly to produce cumulative iceberg maps over long period (from August to October 2016) enabling the identification of potential iceberg fields. To this end, a dedicated WM processing chain has been developed. More specifically, it enabled the detection and basic characterization of icebergs from both Wave mode 1 and Wave mode 2 acquisitions, respectively at 23° and 36° incidence angle. Whereas the backscattering of icebergs is generally higher than the surrounding ocean sea surface at 36° (double bounce and/or multiple bounces), icebergs observed by WM1 may appear darker. 

During the 2016 VG race, 317 S-1 images in Extra Wide Swath have been specifically planned and acquired by ESA, few days ahead the first boat from mid-November 2016. In addition to these images, 91 IW images have been integrated in the analysis as they were already part of the planned observation strategy, imaging areas with potential iceberg risks (e.g. Kerguelen, Crozet islands…). All these images have been processed with automatic iceberg detection, and then compiled by ice analysts to provide iceberg situation reports to the Vendée Globe race direction. In addition, a drift model has been used to forecast the location of EO-based detected icebergs. Exclusion zones could thus be defined.

In this paper/talk, the following topics will be addressed:

the backscattering phenomenology of icebergs as imaged by Wave Modes is commented, and the newly implemented iceberg detector is explained.

the cumulative iceberg maps provided by S-1 WM acquisitions are compared to the altimeter-based maps.

the activities carried out during the Vendée Globe are explained, with a highlight on the benefit of Sentinel-1 data.

the remarkable drift and fate of A56, the most famous tabular iceberg during the last Vendée Globe.

We would like to thank the ESA Sentinel-1 mission manager Pierre Potin, and all the mission planners for their kind cooperation. In addition, this study is being partly funded by the CNRS/LOPS via the 2018 SARICE project.

[1] Vendée Globe: https://www.vendeeglobe.org

[2] Barcelona Race : http://www.barcelonaworldrace.org/

[3] Volvo Ocean Race : http://www.volvooceanrace.com/fr/home.html

[4] Trophée Jules Vernes: http://www.tropheejulesverne.org/

[5] Brest Ultim http://www.brest-ultim.fr/

The surveillance of the maritime traffic is a major issue for security and monitoring issues. Spaceborne technologies, especially satellite AIS ship tracking and high-resolution imaging, open new avenues to address these issues. The free access to Sentinel Earth Observation data streams (high-resolution Sentinel-1 SAR and Sentinel-2 optical imaging, up to a few TB daily [1]) offers novel opportunities for  the  analysis  and  detection  of ship behaviours, including AIS-Sentinel data synergies.

Regarding the joint usage of spaceborne SAR data and AIS information, its relevance for the characterisation of SAR vessel detection performances using interpolated/extrapolated AIS tracks as ground truth has been already demonstrated [2]. It is operationally used for the monitoring of oil spills at sea and the identification of potential polluter sources [3] in the framework of the European Maritime Safety Agency (EMSA) CleanSeaNet Service [4].

In this context, the SESAME (*) initiative aims to develop new big-data-oriented approaches to deliver novel solutions for the management, analysis and visualisation of multi-source satellite data streams going beyond the current implementation.

Preliminary results highlight the relevance of the synergies between SAR and AIS information in terms of geographic coverage and of detection and characterisation of abnormal activities at sea.    We also demonstrate the feasibility to construct large-scale datasets of SAR echoes acquired in various configuration corresponding to a subset of known vessels with a view to applying machine-learning-based detection and classification strategies.

Considering a four-month dataset of Sentinel-1 A satellite data over Europe from March to June 2017, we collected 5414 SAR images. They were systematically processed using CLS vessel detection algorithm. The detected SAR echoes were then matched with interpolated/extrapolated AIS data to build a unique dataset.  We then explore the development of novel learning-based ship vessel detection and identification strategies.

(*) Acknowledgements :

This work was supported by public funds (Ministère de l'Education Nationale, de l'Enseignement Supérieur et de la Recherche, FEDER, Région Bretagne, Conseil Général du Finistère, Brest Métropole) and by Institut Mines Télécom, received in the framework of the VIGISAT project managed by "Groupement Bretagne Télédétection" (BreTel – Brittany Remote Sensing) and by DGA (French Military Procurement Agency) through the ANR/Astrid Program.

[1]  A.G.  Castriotta,    “Sentinel  data  access  report  2016”, Tech. Rep.

[2] R. Pelich et al,   “Performance evaluation of sentinel-1 data in SAR ship detection,” in IEEE IGARSS, 2015.

[3]  N. Longépé et al,   “Polluter identification with spaceborne  radar  imagery,  ais  and  forward  drift  modeling,”Marine Pollution Bulletin,  vol.  101,  no.  Issue  2,  pp.826–833, 2015.

[4]  European  Maritime  Safety  Agency,     “Clean  sea  net service,” https://portal.emsa.europa.eu/web/csn.

Polarimetric SAR (PolSAR) technology has been proven to be very effective in many areas of Earth Observation (EO) including several maritime applications. Firstly developed for airborne platforms, PolSAR has been later adopted also in space-borne missions, e.g. RADARSAT-2, TerraSAR-X, etc. There are certainly a number of technical and hardware constraints in the space-borne case, such as limited power and antenna size, which limit the radar performances.

Spatial resolution, signal to noise ratio, azimuth ambiguity to signal ratio and spatial coverage of a space-borne SAR system are characteristics that affect the capability to effectively detect and classify ships in the sea environment. Compared to single-pol acquisitions, where only one polarization state is transmitted and only one is received at a designed pulse repetition frequency (PRF), a PolSAR system record the 4 linear combinations of Horizontal (H) and Vertical (V) linear polarized pulses, hence, the system has to operate with an ideally doubled PRF. As consequence: a) the Doppler bandwidth per channel is reduced, which causes a loss in azimuth resolution, b) the ambiguous range region increases, which causes the across-track swath to be narrowed down and c) the azimuth length of the antenna is halved in reception, which causes a wider main lobe and higher side lobes.

In the past, the performances of space-borne PolSAR system for the detection of maritime targets have been assessed be direct comparison among the channels acquired. Although this comparison strategy is quite fair, it has never been possible to compare the ship-sea contrast of a target imaged by PolSAR with the contrast resulting from a conventional single-pol SAR that observes the same target with the same acquisition geometry. This is the fairest comparison one can imagine since it does include on one side the unprecedented scattering details coming from a PolSAR acquisition; however, on the other side, it accounts for the lower spatial resolution that characterizes PolSAR acquisitions. It is obvious that such experiment conducted in space requires two twins’ satellites operating simultaneously (or close to be simultaneous). We have taken the unique opportunity provided by the pursuit monostatic configuration of the TanDEM-X mission to accomplish such experiment. In such a configuration, the twin satellites TerraSAR-X (TS-X) and TanDEM-X (TD-X) can operate independently by each other while flying in a close formation with the two orbits separated of circa 76km (that would correspond to approximately 10sec). Operating the first satellite in the standard StripMap single-pol mode and the second satellite with the experimental Dual Receive Antenna (DRA) quad-pol mode, it has been possible to observe the same scene with the same geometry and almost simultaneously.

This paper summarizes the results obtained in the ship-sea contrast and discrimination, processing real SAR data and collocated ship ground truth information provided by automatic identification system on board the ships present in the selected geographical region.

Recently, the European Board and Coast Guard Agency Frontex has showed the necessity to monitor the traffic of illegal goods in different parts of the Mediterranean Sea [1] . The smugglers typically use very fast and small ships, made by fiberglass and carry out their traffic during the night time. This makes the detection of ships by electro-optical sensors almost impossible and the detection by SAR-based algorithm very difficult, due to weak radar cross section of the used ships. To overcome such difficulties, the Agency suggests the detection of the illegal ships by detecting their wakes, which will be very prominent.

In this ambit, a technique for ship wake detection has been recently proposed and applied to X-band SAR images provided by COSMO/SkyMed and TerraSAR-X [2]-[3]. Moreover, to assess the method’s robustness with respect to wake appearance in C-band images, the algorithm has been also applied over 28 wakes imaged by the Sentinel-1 mission with different polarizations and incidence angles [4].

The detection performance has been analyzed in order to distinguish between true confirmations, i.e., imaged features classified as true; true discards, i.e., not-imaged features classified as false; false confirmations, i.e., not-imaged features classified as true; and wrong positioning, i.e., feature locations mismatched with the true locations. The results showed that the wakes were correctly classified in 78.5% of cases, whereas false confirmations occurred in 18.5% of the components.

The above described technique has been conceived to be applied when the ship position is already known in order to estimate its route.

In this paper, the application of the wake detection technique will be properly adapted to be used as a validation of ship presence. In fact, the algorithm correctly validates the visibility of the wake even when it is not present. This means that the algorithm has the potentiality to recognize when the wake is imaged or not, and, in the former case, to identify the wake vertex and the ship location.  

[1].  On-line Source: http://frontex.europa.eu/assets/About_Frontex/Procurement/2017-OJS191-390537-en.pdf

[2].  Graziano, M.D, Grasso M., D’Errico, M., Performance Analysis of Ship Wake Detection on Sentinel-1 SAR Images, Remote Sens. 2017, 9(11), 1107;

[3].  M.D. Graziano, M. D’Errico, G. Rufino, Ship heading and velocity analysis by wake detection in SAR images, Acta Astronautica, Volume 128, November–December 2016, Pages 72–82.

[4].  M.D. Graziano, M. D’Errico, G. Rufino, Wake component detection in X-band SAR images for ship heading and velocity estimation, Remote Sensing, Volume 8, Issue 6, 2016.