Assessment of Sea Surface Signatures for Naval Platforms Using SAR Imagery (AssenSAR)
Lead Research Organisation:
University of Bristol
Department Name: Electrical and Electronic Engineering
Abstract
In space imaging, enhanced image quality is key to the detection and characterisation of difficult and transient targets. For example, accurate evaluation of the sea surface conditions can help with the detection and characterisation of ship wakes. These provide key information for tracking (illegal) vessels and are also useful in classifying the characteristics of the wake generating vessel.
Until recently, one of the main factors hampering research into sea surface modelling was the lack of sufficient data of high enough quality, able to accurately describe the sea surface. Remote-sensing technologies have however shown remarkable progress in recent years and the availability of remotely sensed data of the Earth and sea surface is continuously growing. Several European missions (e.g., the Italian COSMO/SkyMed or the German TerraSAR-X) have developed a new generation of satellites exploiting synthetic aperture radar (SAR) to provide spatial resolutions previously unavailable from space-borne remote sensing. The UK is currently developing the first of a constellation of four satellites that will constitute the NovaSAR mission. This represents a milestone for Earth-observation capabilities but also requires the development of novel image modelling, analysis, and processing techniques, able to cope with this new generation of data and to optimally exploit them for information-extraction purposes.
Indeed, the mathematical modelling and understanding of wakes and other sea surface signatures can be greatly enhanced through image analysis and information extraction from SAR imagery. Hence, this project is concerned not only with the development and validation of new sea surface models, but also with the design of very advanced methods for enhancing SAR image quality and for subsequent information extraction.
The results of this project will be important in the detection and tracking of illegal vessels involved in smuggling goods or humans. They will also be indicative in terms of understanding and classifying the characteristics of the wake generating vessel. As a consequence, the work will directly benefit the design of stealthy vessels that can avoid such detections, reducing the risk to naval operations.
Until recently, one of the main factors hampering research into sea surface modelling was the lack of sufficient data of high enough quality, able to accurately describe the sea surface. Remote-sensing technologies have however shown remarkable progress in recent years and the availability of remotely sensed data of the Earth and sea surface is continuously growing. Several European missions (e.g., the Italian COSMO/SkyMed or the German TerraSAR-X) have developed a new generation of satellites exploiting synthetic aperture radar (SAR) to provide spatial resolutions previously unavailable from space-borne remote sensing. The UK is currently developing the first of a constellation of four satellites that will constitute the NovaSAR mission. This represents a milestone for Earth-observation capabilities but also requires the development of novel image modelling, analysis, and processing techniques, able to cope with this new generation of data and to optimally exploit them for information-extraction purposes.
Indeed, the mathematical modelling and understanding of wakes and other sea surface signatures can be greatly enhanced through image analysis and information extraction from SAR imagery. Hence, this project is concerned not only with the development and validation of new sea surface models, but also with the design of very advanced methods for enhancing SAR image quality and for subsequent information extraction.
The results of this project will be important in the detection and tracking of illegal vessels involved in smuggling goods or humans. They will also be indicative in terms of understanding and classifying the characteristics of the wake generating vessel. As a consequence, the work will directly benefit the design of stealthy vessels that can avoid such detections, reducing the risk to naval operations.
Planned Impact
AssenSAR is highly relevant to space imaging and hence to the UK's satellite industry (one of the Eight Great Technologies) where enhanced image quality can be exploited to detect and characterise difficult and transient targets.
The results of this project will be important in the detection and tracking of illegal vessels involved in smuggling goods or humans. They will also be indicative in terms of understanding and classifying the characteristics of the wake generating vessel. The work will thus inform the design of stealthy vessels that can avoid such detections, reducing the risk to naval operations.
The range of commercial and societal impact objectives are as follows:
Capability: We will help to shape UK capability across the satellite industry through better understanding of remote sensing imagery and its utility in information extraction in marine environments. Our researchers will be trained with the technical and enterprise skills needed to deliver the impact of this research across the wider community.
Commercial: Close coupling with partner organisations will facilitate pull-through into advanced products, opening new opportunities for innovation in both marine surveillance and remote sensing imaging. By liaising with the UK Satellite Applications Catapult, it is expected that the innovation potential of this project will be nurtured and eventual commercialisation of research will be facilitated. There are significant foreseeable opportunities for the creation of new starts-up to exploit the outcomes of this project.
Operational Practice: We will concentrate on the design of new denoising and super-resolution protocols for situations where automated processing is used to support human decision making. Obvious examples are in the viewing and analysis of remote sensing imagery and information extraction therefrom.
The results of this project will be important in the detection and tracking of illegal vessels involved in smuggling goods or humans. They will also be indicative in terms of understanding and classifying the characteristics of the wake generating vessel. The work will thus inform the design of stealthy vessels that can avoid such detections, reducing the risk to naval operations.
The range of commercial and societal impact objectives are as follows:
Capability: We will help to shape UK capability across the satellite industry through better understanding of remote sensing imagery and its utility in information extraction in marine environments. Our researchers will be trained with the technical and enterprise skills needed to deliver the impact of this research across the wider community.
Commercial: Close coupling with partner organisations will facilitate pull-through into advanced products, opening new opportunities for innovation in both marine surveillance and remote sensing imaging. By liaising with the UK Satellite Applications Catapult, it is expected that the innovation potential of this project will be nurtured and eventual commercialisation of research will be facilitated. There are significant foreseeable opportunities for the creation of new starts-up to exploit the outcomes of this project.
Operational Practice: We will concentrate on the design of new denoising and super-resolution protocols for situations where automated processing is used to support human decision making. Obvious examples are in the viewing and analysis of remote sensing imagery and information extraction therefrom.
Publications
Pappas O
(2018)
Superpixel-Level CFAR Detectors for Ship Detection in SAR Imagery
in IEEE Geoscience and Remote Sensing Letters
Karakus O
(2020)
Convergence Guarantees for Non-Convex Optimisation With Cauchy-Based Penalties
in IEEE Transactions on Signal Processing
Karakus O
(2020)
Ship Wake Detection in SAR Images via Sparse Regularization
in IEEE Transactions on Geoscience and Remote Sensing
Karakus O
(2020)
Correction to "Ship Wake Detection in SAR Images via Sparse Regularization"
in IEEE Transactions on Geoscience and Remote Sensing
| Description | 1) In order to detect ship wakes, one important first step is to detect the wake-generating vessel. That provides the starting point from which the search for sea wakes can begin. In this context, the AssenSAR team have developed a novel, superpixel-based topology for the detection of ships at sea in synthetic aperture radar (SAR) remote sensing imagery. The method is based on the so-called Constant False Alarm Rate (CFAR), which are detectors that utilise various statistical models for sea clutter in conjunction with thresholding schemes. Superpixels on the other hand are groups of connected pixels with similar colors or gray levels. The architecture based on the combination of superpixels and CFAR can be utilized with most prior sea clutter statistical models or thresholding schemes. The performance of this modified SP-CFAR algorithm was demonstrated on TerraSAR-X and SENTINEL-1 images, with superior results in comparison to classical CFAR for various background distributions. 2) Following accurate detection of ship targets in SAR images, the team have also developed a method for the detection of corresponding ship wakes. Since ship wakes mostly appear as lines in SAR images, we have used line detection methods for their identification. Unlike classical approaches that use direct line detection methods, for example based on the Radon transform, our proposed method uses an inverse problem formulation that relies on very advanced mathematical concepts related to non-convex optimisation. We have conducted extensive experiments involving a high number of ship wakes visible in SAR images acquired using different satellite platforms. The performance of our proposed method was demonstrated to be superior to that of previously proposed techniques with a significantly higher detection performance. 3) An important theoretical development during the life of the project has been the formalisation of a novel penalty function, based on the Cauchy distribution that can be used to regularise the solution to various inverse problems. Convergence guarantees have been proven and published and the resulting algorithm has been employed to solve a number of computational SAR imaging problems, including despeckling, autofocusing, or super-resolution. Notably, the same penalty function has been shown to be applicable in a completely different domain, i.e. it was used for detection of markers of COVID-19 in lung ultrasound images. 4) The visibility of sea wakes in SAR images is enabled by the interaction of the electromagnetic waves with the water surface. This is called Bragg resonance scattering and relates to both characteristics of the sea state (amplitude of waves and their direction) and SAR signal (wavelength, incidence angle of signal, etc). We investigated five different sea wave spectra and, besides hydrodynamical effects (wave elevation models), we also simulated corresponding synthetic aperture radar images. We considered different SAR parameters, including frequencies (X, C, and L-band), incidence angles, signal polarizations (VV, HH), image resolutions (2.5, 5 and 10 m), four different platform configurations (two airborne and two spaceborne), and velocity bunching effects (azimuthal cut-off, shifting and smearing). We also determined the boundary conditions for ship wake detectability in SAR images. Most importantly, the MATLAB implementation of this extremely versatile simulator is available as open source and can become a valuable resource for anyone interested in oceanography, or marine and maritime engineering. |
| Exploitation Route | Applications of our ship and ship wake detection algorithms can be related to a diverse set of problems such as navigation and equipment positioning; radar from manned or unmanned platforms and satellites for object detection, classification, tracking, or avoidance; remote sensing of oceanographic phenomena such as fronts and eddies; law enforcement and monitoring for possible drug trafficking and illegal immigration. One interesting potential application that emerged following our interactions with the Satellite Applications Catapult (SAC) is in illegal fishery identification. In particular, we have started talking to OceanMind, a spin-out from SAC, who provide insights and intelligence into fishing compliance to government authorities and seafood buyers, in order to enable enforcement and compliance to protect the world's fisheries. Our methods could form the basis of machine learning algorithms that would classify detected ships not associated to automatic identification systems and could hence potentially perform illegal fishing activities. |
| Sectors | Aerospace, Defence and Marine,Digital/Communication/Information Technologies (including Software),Environment,Healthcare,Government, Democracy and Justice,Security and Diplomacy,Transport |
| URL | https://assensar.blogs.bristol.ac.uk/ |
| Description | Based on scientific findings from AssenSAR, the University of Bristol has been successful in securing funding in the form of a knowledge transfer partnership grant (KTP13509) to develop, in collaboration with Toshiba Europe Limited (TEUR), a novel radar technology based on sea wave prediction to improve offshore floating wind turbine efficiency, maintenance, and safety. This will reduce their cost of energy to benefit the electricity consumer and enable their wider deployment in deeper waters (e.g.: in the Celtic Sea). Specifically, AssenSAR involved the development of hydrological models of the sea surface along with a simulator of synthetic aperture radar (SAR) images corresponding to those models. Image analysis tools have also been developed to extract information from real SAR data and verify the developed models. From an application point of view, the focus of AssenSAR was on the detection and identification of moving platforms (ships). However, the project provided a proof-of-concept for the analysis and synthesis of SAR images of the sea surface that the KTP project can build upon and apply to the stabilisation of floating offshore wind turbines. |
| First Year Of Impact | 2023 |
| Sector | Energy |
| Impact Types | Economic |
| Title | AssenSAR Image Simulator |
| Description | This software implements a simulation framework for SAR imagery of the sea surface, which includes the superposition of sea waves and ship wakes. The methodology is based on well-proven concepts: the linear theory of sea surface modelling, Michell thin-ship theory for Kelvin wake modelling, and ocean SAR imaging theory. The modelling of the sea surface is based on the sea wave spectrum. A two-scale model is used to simulate intensity SAR images of disturbed sea surface with Kelvin wakes, including tilt and hydrodynamic modulations, while the velocity bunching model is also employed. SAR image simulator includes different parameters, such as the wind speed and direction, fetch length (sea state factors), ship parameters, platform parameters (airborne and satellite SAR platforms), and SAR scanning parameters, all of which contribute to the visualization of sea waves and ship wakes in SAR images. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2021 |
| Provided To Others? | Yes |
| Impact | This forms the ground for a new KTP award to the PI, in collaboration with Toshiba Europe Limited, as described in the "Narrative Impact" section. |
| URL | https://data.bris.ac.uk/data/dataset/el0p94vgxjhi2224bx78actb4/ |
| Title | Cauchy Proximal Splitting (CPS) |
| Description | These Matlab functions implement a proximal splitting method involving a non-convex penalty function based on the heavy-tailed Cauchy distribution. A function for calculating the proximal operator of the Cauchy prior is provided, and two examples are included to illustrate how to perform cost function optimisation with a forward-backward (FB) -based algorithm that implements the corresponding Cauchy proximal splitting (CPS) method. The two signal processing examples include 1D signal denoising in the frequency domain and image de-blurring. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2020 |
| Provided To Others? | Yes |
| URL | https://data.bris.ac.uk/data/dataset/15y437loa26cr2nx8gnn3l4hzi/ |
| Title | CauchySAR_v1 |
| Description | This software includes MATLAB source code for solving SAR imaging inverse problems using the Cauchy proximal splitting (CPS) algorithm. CPS is a method for the optimization of a cost function that includes a non-convex Cauchy-based penalty. The convergence of the overall cost function optimization is ensured through careful selection of model parameters within a forward-backward (FB) algorithm. The package includes solutions to three standard SAR imaging inverse problems, i.e. super-resolution, image formation, and despeckling, as well as ship wake detection for maritime applications. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2021 |
| Provided To Others? | Yes |
| Impact | Nothing yet but it is part of a portfolio of tools that have been developed as part of this project and could serve the purpose of performing image analytics in SAR data. |
| URL | https://data.bris.ac.uk/data/dataset/2ld1on1vrlzba2o05jd5jhdhd0/ |
| Title | GG-Rician SAR Image Modelling |
| Description | This package includes the MATLAB source code for the modeling of SAR images with the Generalized-Gaussian Rician (GG-Rician) distribution. The GG-Rician distribution is a novel statistical model specifically developed for the characterization of synthetic aperture radar (SAR) images. It is based on the Rician distribution to model the amplitude of the complex SAR signal, the in-phase and quadrature components of which are assumed to be generalized-Gaussian distributed. The extension to modeling SAR images in intensity format is also included. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2021 |
| Provided To Others? | Yes |
| Impact | Nothing yet, but the code can be part of any future algorithms that rely on the statistical characterisation of SAR images to perform data analytics. |
| URL | https://data.bris.ac.uk/data/dataset/3nqhdd4qvorwx28hjh8bh6g3r8/ |
| Title | QuantLUS - CPS v1.0 |
| Description | This source code package includes the MATLAB source codes for implementing a novel method for line artefacts quantification in lung ultrasound (LUS) images of COVID-19 patients. The method is formulated as a non-convex regularisation problem involving a sparsity-enforcing, Cauchy-based penalty function, and the inverse Radon transform. Moreover, a simple local maxima detection technique in the Radon transform domain, associated with known clinical definitions of line artefacts, is employed. The method accurately identifies both horizontal and vertical line artefacts in LUS images. In order to reduce the number of false and missed detection, the method includes a two-stage validation mechanism, which is performed in both Radon and image domains. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2020 |
| Provided To Others? | Yes |
| URL | https://data.bris.ac.uk/data/dataset/z47pfkwqivfj2d0qhyq7v3u1i/ |
| Title | Ship Wake Detection in SAR Images via GMC Regularization |
| Description | This code implements a method for detecting ship wakes in synthetic aperture radar (SAR) images of the sea surface. The method is based on a linear model assumption for the wakes and hence the Radon transform is employed, within an inverse problem formulation, for detecting the wakes. The cost function associated with the image formation model includes a sparsity enforcing penalty, i.e., the generalized minimax concave (GMC) function. Despite being a nonconvex function, the GMC penalty allows the overall cost function to remain convex. The proposed solution is based on a Bayesian formulation, whereby the point estimates are recovered using a maximum a posteriori (MAP) estimation. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2020 |
| Provided To Others? | Yes |
| URL | https://data.bris.ac.uk/data/dataset/f2q4t5pqlix62sv5ntvq51yjy/ |
| Title | SynthWakeSAR dataset |
| Description | The synthetic ship classification dataset of synthetic aperture radar (SAR) images of the sea surface (SynthWakeSAR) includes 10 real ship models for a total of 46080 simulated images containing visible ship wakes. The dataset was created using the "AssenSAR Image Simulator", also available via the University of Bristol. The simulated SAR images consist of normalized radar cross-sections (NRCS) with tilt and hydrodynamic modulations and velocity bunching. The size of each scene is 0.96 × 0.96 km and the SAR resolution is 3.3 m in both azimuth and range directions. The size of the images is 227x227x1 as required by the standard convolutional neural network architecture AlexNet. The dataset is provided in three variants (i) as noise-free images NF, (ii) as noisy images (N) with simulated speckle noise, and (iii) as despeckled SAR images (D). This dataset is intended primarily for developing ship classification algorithms based on their wake signature in SAR images. This dataset is published under a GNU General Public License v3.0. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2022 |
| Provided To Others? | Yes |
| Impact | This dataset has been released very recently to have made a notable impact yet, but the corresponding paper, published in MDPI Remote Sensing is one of the most read article in the special issue where it has been published, indicating its potential for becoming a highly cited and used resource by the ocean engineering community. |
| URL | https://data.bris.ac.uk/data/dataset/30kvuvmatwzij2mz1573zqumfx/ |
| Title | AssenSAR Image Simulator |
| Description | This software implements a simulation framework for SAR imagery of the sea surface, which includes the superposition of sea waves and ship wakes. The methodology is based on well-proven concepts: the linear theory of sea surface modelling, Michell thin-ship theory for Kelvin wake modelling, and ocean SAR imaging theory. The modelling of the sea surface is based on the sea wave spectrum. A two-scale model is used to simulate intensity SAR images of disturbed sea surface with Kelvin wakes, including tilt and hydrodynamic modulations, while the velocity bunching model is also employed. SAR image simulator includes different parameters, such as the wind speed and direction, fetch length (sea state factors), ship parameters, platform parameters (airborne and satellite SAR platforms), and SAR scanning parameters, all of which contribute to the visualization of sea waves and ship wakes in SAR images. |
| Type Of Technology | Software |
| Year Produced | 2021 |
| Open Source License? | Yes |
| Impact | This constitutes the basis for further work done on automatically identifying ship targets, done in collaboration with SSTL as part of a follow-on EPSRC Impact Acceleration Award. |