Nowadays, global vessel monitoring is made possible by combining information acquired by different sensors that are generally divided into cooperative and non-cooperative data sources. In this study, we demonstrate the usefulness of combining/fusing Synthetic Aperture Radar (SAR) images, as a source of non-cooperative data, and Automatic Identification System (AIS) data flows, as cooperative data, for a comprehensive analysis of the maritime traffic. SAR images occupy a privileged place in vessel monitoring due to their quasi all-weather and day/night observation capacity. The C-band Sentinel-1 sensor acquires systematically SAR images over areas of maritime interests and presents several assets such as the open access policy data or the availability of complex dual-polarized data. The availability of large datasets of dual-polarized SAR images has led to new scientific studies and the development of new SAR vessel detection algorithms based on dual-polarized features. Several research studies have shown that the fusion of polarimetric channels introduces a gain with respect to the use of single polarization images in diverse remote sensing applications, including ship detection. Over areas of maritime surveillance interest, Sentinel-1 is systematically acquiring data in the VV-VH polarization configuration. In this study, we have explored several polarimetric descriptors derived from the VV-VH configuration, namely the complex VV-VH coherence and descriptors derived from the Eigenvalue-Eigenvector decomposition of the VV-VH covariance matrix. Several case studies that illustrate the vessel signatures for different dual-polarized descriptors are first presented. The VV-VH coherence, defined in this study as the un-normalized correlation between the two polarimetric channels, presents high correlation values for vessel signatures and low values for the sea clutter. This allows increasing the target to clutter ratio with respect to intensity domains of each polarimetry channel. We have performed a complete validation of the detection performances of VV-VH coherence issued from the processing of several S1 images and the use of AIS data as ground truth. The AIS positions have been interpolated by considering the heading and speed between each two consecutive positions, in order to predict the location of a vessel at the acquisition time of a SAR image. For the experimental results, we have focused on the Sentinel-1 Interferometric Wide Swath mode with a resolution of 20m, acquired over different areas including both coastal and open sea zones. The overall cross-correlation between the AIS and SAR data sources presents a significant agreement between the two data sources. However, an analysis of uncorrelated AIS and SAR targets is also presented for a better understanding of the coupling mismatching generated by different factors such as localization and temporal inaccuracies.

Iceberg monitoring is required to ensure navigation and offshore operational safety in ice prone Arctic and sub-Arctic waters. The risk of iceberg collision with ship or offshore structure can be immensely reduced by reliably detecting the iceberg well ahead of time so that safety measures can be taken. Satellite Synthetic Aperture Radar (SAR) imagery has been a reliable technology for operational iceberg monitoring in the Arctic for its all weather and all season imaging capability. In SAR images, ship targets often become ambiguous with iceberg targets and extensive efforts have been made to reduce false alarms in the iceberg-ship detection algorithm. Use of machine learning to develop algorithms for distinguishing iceberg and vessel targets requires large validated data sets that are often costly, time consuming and, in some cases, inaccessible. Generating electromagnetic (EM) backscatter models of iceberg and ship targets can be a vital step in developing a robust iceberg/ship classification algorithm. This abstract outlines work steps in developing iceberg backscatter model of icebergs and assessing the influence of geometric, environment parameters on SAR response.

An EM backscatter modelling tool called GRECOSAR was used for the modelling work and compared with ground truth data. The imaging scene consists of iceberg targets surrounded by the ocean surface. The 3D profiles of the icebergs were obtained using LiDAR and multi-beam sonar data collected during field programs off the coast of Newfoundland and Labrador, Canada. While profiling the iceberg targets, SAR images from satellites were captured and compared with the simulated SAR images. Comparisons were made in terms of total radar cross section (TRCS), SAR signatures and polarimetric decompositions of the targets generally indicate credible simulations.

Simulations were performed at six orientations of the target in the 180˚ range. Simulations were also performed at varying sea conditions (i.e., wind speeds and directions). Wind direction was varied over 90˚ extent to observe the level of backscatter variability of the icebergs in the simulator. Furthermore, simulated SAR images were generated for low and high wind condition. Our study finds that the macro structure of the iceberg dominates its polarimetric behaviour. Variability of the TRCS and signatures were seen as the scene parameters change highlights the usefulness of EM modelling in developing robust iceberg/ship classifiers.

Icebergs tend to experience a significant amount of weathering from sea wash, sun exposure and sea ice abrasion that changes the surface and sub-surface dielectric permittivity of the icebergs. Icebergs were modelled with high dielectric conditions to mimic melt iceberg surfaces as seen during field work also as GRECOSAR can simulate single layer propagation. A mathematical model was developed to quantify the effect of melt water in the backscatter. A simplified two layer iceberg model is proposed. The first layer consists of a thin homogenous water layer representing the ice melting condition. The second layer consists of an inhomogeneous ice medium with significant air bubble content. The two layers are enclosed by the air and sea-water medium at the top and bottom layer respectively. The surface scattering is from the air-water, water-ice and ice-sea water surfaces and the volume scattering is from the air bubbles inside the iceberg. These models are developed based on the integral equation model (IEM). As a next step, simulated datasets of icebergs and ships at varying imaging parameters will be generated and passed on to an iceberg/ship classifier and compared with classifier performance with real dataset. A good agreement between the classifier’s performances will further indicate simulated data set captures all the physical properties of icebergs and completes the validity of the EM modelling.

Aquaculture are a very valuable asset for many coastal countries and in the future they will play an important role in food security. Satellite remote sensing can improve the temporal and geo-spatial analysis of such marine facilities. Detecting platforms used for fish and shellfish farming provides a way to monitor assets and check they do not get damaged by storms. It also allows to identify illegal deployment of structures in areas which should not host farms.
In this work, we want to evaluate the potential of a new methodology that uses PolSAR data. This is called intensity Dual-Pol Ratio Anomaly Detector (iDPolRAD) and it was initially proposed to detect icebergs embedded in sea ice. Extensive work has been carried out on detecting ships using space-borne Synthetic Aperture Radar (SAR) systems. However, the identification of smaller and non-metallic targets is still challenging especially when the sea conditions are rough. This work presents the first analysis of the iDPolRAD performance with aquaculture structures.

The algorithms:
The algorithm is based on the observation that the most of the maritime targets exhibit a different polarimetric behaviour compared to the sea. Specifically, the cross polarization channel and the ratio between cross- and co-polarizations (here referred to as depolarization ratio) increases. One of the reason is that complex targets (e.g. shellfish platforms) will provide scattering which will resemble Volume scattering or reflections from planes (mostly wet surfaces) with random orientations. They are therefore expected to have a polarimetric backscattering that is different from the one of the sea which is surface scattering. Two boxcar filters are applied over the HV and HH intensity images, exploiting two different window sizes: a smaller window or test window Wtest and a larger window or training window Wtrain. The detector can be writes as the difference of the “HV filtered with the test window” minus “HV filtered with the train window”, all divided by the “HH filtered with the train window”.
The previous operator is built as a ratio between intensities and therefore it is scale invariant. This is a very valuable property for a polarimetric observable, however scale invariance may be disadvantageous for some detection tasks. For instance, if the signal is very low and close to the noise floor, the detector may become noisy. An easy way out is by multiplying the detector by an intensity or magnitude image. In this context, the cross polarization channel should be preferred because it shows a higher contrast between icebergs and clutter. Multiplying the intensity of HV by the previous operator forms the iDPolRAD. If a pixel of the HV intensity image presents an anomaly in volume or oriented reflections, then it is multiplied by a large number. If it presents a homogeneous area, then it is multiplied by zero and if it presents a decrease in volume or oriented reflections, then it becomes negative. This enhances the contrast between anomalies in volume/oriented reflections and clutter.

The data:
In this work we tested the iDPolRAD on a large amount of dual-polarization Sentinel-1 and 5 HH/HV PINGPONG Cosmo-SkyMed images. The test site is in the coastal area near Vigo, Spain. This is an area intensely exploited for the production of mussels and hundreds of platforms cover the coastal area.

Preliminary results:
The iDPolRAD seems able to increase the contrast between the sea background and the platforms allowing the identification of more platforms. An extensive comparison of the dual-pol and single-pol algorithms will be presented.

Novelty:
The topic of aquaculture structure detection is not very well studied in literature, although it is valuable to monitor presence of illegal aquaculture structures. The iDPolRAD was never used to detect aquaculture structures. The comparison of Sentinel-1 and COSMO-SkyMED for aquaculture studies is also novel.

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.

With rapid growth of aquaculture as one of the main food sectors the task of monitoring aquaculture sites becomes of primary importance for regulating authorities and aquaculture controlling bodies. Many of these tasks, such as estimating accurate position of aquaculture sites, controlling the number of sites in operation and preventing intrusion into marine protected areas cannot be solved efficiently without using remote sensing methods. Application of remote sensing technology helps to reduce inspection time and improve the efficiency of monitoring aquaculture in remote areas.

This study focuses on ESA’s Sentinel-1 SAR data to detect finfish cages and shellfish farms, and estimate the coordinates and dimension of each structure. For a UK case study, the Sentinel-1 images are processed automatically using custom image processing methods and algorithms. The study goes further than applying standard constant false alarm rate (CFAR) target detection algorithms to SAR images, by implementing time-series analysis to detect fish cages as static objects, experimenting with polarisation signatures and investigating the feasibility of detection. To discriminate aquaculture structures from small rocks and other coastal features we use a high resolution land/water mask generated from Ordnance Survey DEM data.

The developed methodology was tested in Scotland for the area of approximately 100km (a tile on the Ordnance Survey grid). Approximately 150 images were processed in total, covering 3 months of Sentinel-1A data in June-August, 2016. Performance was validated experimentally by comparison of detection results with positions of aquaculture sites identified visually in the Google map (see Figure 1): the system was able to correctly identify about 70% of aquaculture structures (fish cages and mussel farms), with 20% false alarms associated with small rocks and moored boats.

It is envisaged that the developed methodology will provide significant benefits to studies of aquaculture environmental impact. Locational information for aquaculture structures is vitally important for considering risk/growth while planning sustainable expansion of the industry. Linking data acquired within this study to environmental impact analysis (using modelling or remote sensing) would lead to an enhanced capability to estimate the impact of current aquaculture cages and to select future sites.

Over the last 60 years the increase in chemical and plastic pollution have a significant pressure on the ocean health [1]. The present extent of plastic pollution according to several research papers has already filled all five subtropical ocean gyres - in the Indian, North Atlantic, North Pacific, South Atlantic and South Pacific oceans [2]. Plastic pollution signifies an enormous threat to marine life and ecosystems, because plastic fragments can be ingested by marine animals and fish and transferred via food chain to humans. The fragments can leach their chemical components/ingredients into surrounding seawater and to release sorbed pollutants, depending on bioavailability [3].

The aim of this research is to investigate the applicability of Sentinel-1 SAR data and technology to detect and map ocean surfactants and films, with the hypothesis that they are related to microplastic pollution. The research objective is to test Sentinel-1 data to identify areas with plastic (microplastic) pollution across research sites located in the North Pacific, North Atlantic and North Indian oceans, including area of Hudson Bay in Canada as reference. The North Pacific and North Atlantic oceans are chosen because of high microplastic count, mentioned in [4], [5], [6]. According to [7], the Hudson Bay has the lowest microplastic count, in less than 100 counts per km2 and Indian Ocean (inside the ocean gyre) with less than 10,000 counts per km2.

The Sentinel-1 dataset used in this study is VV-polarized, Ground Range Detected (GRD), Level-1 product with multi-looked intensity and acquisition modes of Extra-wide (EW), Interferometric Wide swath (IW) and Wave (WV). In addition, the ancillary data on chlorophyll-a, wind speed and shipping density are used to illuminate areas of possible oil spills and ships waste discharge. The pre-processing steps for Sentinel-1 data followed steps of removing thermal noise, multi-looking, Box car filtering and ellipsoid correction. The surfactants, sea-slicks and bio-films were visible on Sentinel-1 images as a dark “ring-like” effects and structures with clear and distinct dark signatures in areas of North Pacific and North Atlantic oceans. The main hypothesis is that these “ring-like” and linear structures occur because of the microbial colonization of microplastics related with low shipping activities and low chlorophyll-a concentrations.

The VV-polarized Sentinel-1 images are showing distinct dark signatures in areas of North Pacific and North Atlantic oceans. These signatures show dependency on a certain range of winds, starting from 0.12 to maximum of 4.44 ms–1, measured by Aquarius and CERSAT scatterometer. No significance with presence of chlorophyll-a are found. The areas with dark signatures are located in areas of “solutions of microplastics count”, the areas of high concentrations of microplastics over the North Atlantic and North Pacific oceans, according to [7].

The research is still on-going, with the final aim to develop a machine learning algorithm (using a supervised classifier and convolutional neural network) able to separate the backscattering from surfactants, sea-slicks and bio-films. The machine learning algorithm will be aided by textual image analysis, results of modelling of the surfactants accumulation mechanism based on oceanographic processes (eddies) and knowledge on microplastics and surfactants behaviour from lab experiments. The final results will be verified against in-situ data from previous scientific studies.

References:
[1] C. M. Duarte, “Global change and the future ocean: a grand challenge for marine sciences,” Frontiers in Marine Science, vol. 1, no. 63, 2014-December-02, 2014.
[2] V. Hidalgo-Ruz, L. Gutow, R. C. Thompson, and M. Thiel, “Microplastics in the marine environment: a review of the methods used for identification and quantification,” Environmental science & technology, vol. 46, no. 6, pp. 3060-3075, 2012.
[3] B. Jovanović, “Ingestion of microplastics by fish and its potential consequences from a physical perspective,” Integrated environmental assessment and management, vol. 13, no. 3, pp. 510-515, 2017.
[4] D. Cressey, "The plastic ocean," NATURE PUBLISHING GROUP MACMILLAN BUILDING, 4 CRINAN ST, LONDON N1 9XW, ENGLAND, 2016.
[5] A. Cózar, F. Echevarría, J. I. González-Gordillo, X. Irigoien, B. Úbeda, S. Hernández-León, Á. T. Palma, S. Navarro, J. García-de-Lomas, and A. Ruiz, “Plastic debris in the open ocean,” Proceedings of the National Academy of Sciences, vol. 111, no. 28, pp. 10239-10244, 2014.
[6] J. Boucher, and D. Friot, “Primary microplastics in the oceans: a global evaluation of sources,” IUCN, Gland, Switzerland, 2017.
[7] S. Erik van, W. Chris, L. Laurent, M. Nikolai, H. Britta Denise, A. v. F. Jan, E. Marcus, S. David, G. Francois, and L. Kara Lavender, “A global inventory of small floating plastic debris,” Environmental Research Letters, vol. 10, no. 12, pp. 124006, 2015.

Wind is a sustainable and alternative resource for producing energy and it has a good reputation of being a green form of electricity. Within this context, wind turbines are widely used at onshore and offshore sites to convert the energy of moving air into electrical power. For this reason, wind turbines are a critical infrastructure whose monitoring is an important issue for both economy and environment protection. Within this context, remote sensing can be an important tool to guarantee an effective and relatively cheaper monitoring. Optical images have the great advantage of being simple to interpret and they are easily obtainable. However, optical radiation is severely affected by cloud cover, solar illumination, and other adverse meteorological conditions. These problems can be solved using radar sensors, which guarantee all-day and almost all-weather acquisitions, together with a wide area coverage. In particular, the Synthetic Aperture Radar (SAR) can be very useful for intertidal zone monitoring purposes, because of its fine spatial resolution.
The objective of this study is quantifying the added-value of polarimetric information to detect metallic targets. On this purpose, a very challenging scenario is considered that consists of mud flat area where wind turbines are present. To make the analysis fair, we selected detection algorithm that are able to work with both full- and partial-polarimetric information, i.e.; Polarimetric Notch Filter (PNF) and the change detection approach proposed in [1] and [2], respectively.
Experiments, undertaken on actual SAR data collected over the intertidal zone near Jiangsu, China, by the C-band RadarSAT-2 and Sentinel-1 missions show that the proposed methodologies, well detect the wind turbines inside mud flat areas. Furthermore, a detailed analysis shows that polarimetric information always guarantees performance better than the single–polarization counterpart.
[1] A. Marino, (2013), “A Notch Filter for Ship Detection With Polarimetric SAR Data", IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 6(3), 1219-1232.
[2] A. Marino; S.R. Cloude, and J. M. Lopez-Sanchez, (2013), “A New Polarimetric Change Detector in Radar Imagery”, IEEE Transactions on Geoscience and Remote Sensing, 51(5), 2986 -3000.