Maritime Traffic Monitoring with TerraSAR-X
The Objectives of the Project are the Following:
1 - Study the potentialities of TerraSAR-X stripMap and scanSAR products to monitor maritime vessel traffic, namely to detect, to estimate cinematic parameters, and to classify moving vessels.
2- Adapt and develop new methods and algorithms aimed at the objectives listed above, taking advantage of the high geometric and radiometric resolutions of TerraSAR-X, in comparison, for example, with the ASAR and RADARSAR sensors.
3 - Build a prototype for for maritime traffic monitoring based on TerraSAR-X data. The system shall be able to provide geographical location of the identified vessels, speed, heading, type, and a degree of confidence of the detected, estimated and classified entities.
SCIENTIFIC ISSUES
1 - Moving vessel detection and cinematic parameter estimation.
Moving targets, even with very low speeds, appear in SAR images defocused and misplaced [6], [7]. For this reason detection and cinematic parameters estimation are complex problems.
Even if we can detect moving targets with a single channel SAR sensor, it is not possible to infer the velocity vector and the correct location from the target phase history. It is the so-called "blind angle ambiguity" [8], [9].
The blind angle ambiguity can be solved by using multichannel SAR [10], [12]. These ideas were successfully illustrated during the SRTM mission, were due to mechanical constraints the two antennas were assembled with 7m along track component [12].
The feasibility of traffic monitoring with TerraSAR-X has been addressed in [6]. In this work authors rely on the Along Track Interferometic Mode [14], which basically means a two channel SAR sensor. Unfortunately, this is an experimental mode that will not be available, at least for now.
As a conclusion, the detection and moving target parameter estimation has to rely on a single channel. This is exactly what is addressed in our works [1],[2],[3],[4],[5]. These works, basically, introduce solutions for the blind angle ambiguity using a single SAR sensor. The key idea is to exploit not only the phase, but also the information contained in the moving target echo amplitude.
The algorithms therein developed are designed for airborne sensors. In this project we propose to adapt them to spaceborne data in StripMap acquisition mode. Since the swath of this mode in only 30km wide, we would like to use ScanSAR data. The transposition of our algorithms [1] and [2] to ScanSAR data is not, however, straightforward, owing to the way data is acquired in this mode. This is an open question that we will address in the course of the project.
2 - Vessel Classification
Classification of naval vessels from SAR data is a challenging problem owing to variable 3-D image acquisition geometry, image blurring due to ship motion, speckle noise, dependence of radar scattering to ship orientation, etc. The enhanced resolution of the TerraSAR-X will surely be a step forward in this pattern recognition subject.
The feature extraction module plays a central role in the success of the classifier. Our approach will be the following:
a) for each detected vessel, refocus the image using the estimated FM rate of the respective detected moving target. This is essential, otherwise the target would appear defocused in a degree depending on its along-track velocity.
b) segment the target using a Bayesian Markov random field (MRF) prior. In this way, the segmentation is carried out not only based on the pixel intensity, but also by enforcing spatial continuity, in a statistical sense.
c) from the segmented objects extract orientation, length, and beam. These parameters are inferred by means of a matched filter type approach. Besides these parameters, we also compute the radar cross section in a number of disjoint regions. This set of features is then applied to a supervised classifier which shall decide among the following classes: category (e.g commercial, military combatant, fishing) and type (e.g. Tanker, Cargo, Cruiser, Frigate, Destroyer, etc.)
Partners: Instituto de Telecomunicações (IT) 2 and Critical Software, SA (CSW)
REFERENCES
[1] J. Dias and P. Marques, "Multiple moving target detection and trajectory estimation using a single SAR sensor. IEEE Trans. on Aerospace and Electronic Systems}, vol. 39, no. 2, pp. 604 - 624, 2003.
[2] P. Marques and J. Dias, "Velocity estimation of fast moving targets using a single SAR sensor. IEEE Transactions on Aerospace and Electronic Systems, vol. 41, no. 1, pp. 75--89, 2005.
[3] P. Marques and J. Dias, "Moving targets velocity estimation using the signature curvature information", in EUSAR'04 European Conference on Synthetic Aperture Radar, vol. 2, pp. 529-532, 2004.
[4] P. Marques and J. Dias, "Moving targets velocity estimation using aliased sar raw-data from a single sensor", in EUSAR'02 European Conference on Synthetic Aperture Radar, 2002.
[5] P. Marques and J. Dias, "Imaging of fast moving targets using undersampled SAR raw-data", in IEEE International Conference on Image Processing-ICIP'01, vol. 3, pp. 624--627, Thessalonoki, 2001.
[6] F. Meyer and S. Hinz, "The Feasibility of traffic monitoring with TerraSAR-X. Analyses and consequences". Proceedings of the Geoscience and Remote Sensing Symposium - IGARSS04, pp. 1502-1505, 2004.
[7] R. Raney, "Synthetic aperture imaging radar and moving targets," IEEE Transactions on Aerospace and Electronic Systems, vol. 7, no. 3, pp. 499-505, 1971.
[8] S. Barbarossa, "Detection and imaging of moving objects with synthetic aperture radar", IEE Proceedings-F, vol. 139, no. 1, pp. 79-88, 1992.
[9] M. Soumekh, "Reconnaissance with ultra wideband UHF synthetic aperture radar" IEEE Signal Processing Magazine, pp. 21-40, July 1995.
[10] J. Ender, "Detection and estimation of moving target signals by multi-channel SAR", in Proceedings of the EUSAR'96, pp. 411-417, 1996.
[11] R. Klemm, "Introduction to space-time adaptive processing" IEE Electronics & Communication Engineering Journal, vol. 911, no. 1, pp. 5-12, 1999.
[12] H. Breit, M. Eineder, J. Holzner, H. Runge, and R. Bamler, "Traffic monitoring using SRTM along-track interferometry", Proceedings of the Geoscience and Remote Sensing Symposium - IGARSS03, pp. 1187-1189, 2003.
[13] A. Roth, and M. Eineder, and B. Schaettler, "TerraSAR-X: A new perpective for applications requiring high resolution spaceborne SAR data", 2nd GRSS/ISPRS Joint Workshop on Remote Sensing and Data Fusion over Urban Area, pp. 22-23, 2003
Oil Slick Surveillance Using ASAR/MERIS Data
The objective of the project is to develop an oil slick detector/classifier based on the ASAR/MERIS imagery that may assist in targeting cleanup, control efforts, and law enforcement.The final product should be an oil monitoring system that should be able to run automatically and to take advantage of the enhanced capabilities of the ASAR sensor (compared to ERS), in terms of coverage, range of incident angles, and modes of operation (e.g., the ASAR Wide Swath Mode in VV polarisation).The system should be able to classify the candidate regions as either oil slicks or natural surfactants, and to provide information on the rate and direction of oil movement. A set of warnings should be provided calling for human intervention whenever appropriate (e.g., detection of a region with high likelihood of containing oil, unresolved surfactants, etc.) By adding MERIS information and by adopting modern statistical/learning techniques such as Bayesian segmentation based on Markov Random Fields (MRF) and Belief Networks (BN) to fuse ASAR and MERIS data and to build the classifier, we aim at outperforming the state-of-the-art oil monitoring systems.The following points are a roadmap for the methodology to adopt in the project:
1. Combine ASAR and MERIS data to improve the classifier performance. The value of visible and near infrared data in complementing SAR information have been pointed out by many authors. For example, it has recently been demonstrated the possibility of estimating oil concentrations based on fluorescence spectroscopy. The poor resolution of the MERIS images, even in full resolution mode, may limit severely its value in oil detection. We intend to overcome this limitation by applying modern spectral unmixing techniques, allowing to decompose the materials present at a given pixel according to their abundance fractions.
2. Adopt Bayesian region-based segmentation approaches using Markov Random Fields, such as the multilevel logistic (MLL), leading to effective segmentation of dark regions in SAR images. This is a critical step in obtaining informative features to feed the classifier.
3. Use wind information derived from ASAR data. Wind information plays a crucial role in SAR oil slick detection/classification, and is available on the "Wave Mode Ocean Wave Spectra" (ASA_WVP_2P product).
4. Incorporate recent findings published in [FIDR02], according to which “A stricking conclusion … is that high order moments (from the second on) of oil slick SAR images statistics are quite different compared to those pertinent to an equivalent wind speed decrease on the image area. This suggests a convenient way to define new appropriate oil slick signatures” .
5. Detect vessels and their moving target parameters using recent techniques proposed by the PI and its students (paper [1] of the PI Curriculum Vitae) .
6. Adopt Belief Networks (BN) to fuse ASAR and MERIS data and to build the classifier. BNs provide optimal ways to deal with uncertainty, vagueness, and error. Furthermore, BNs can deal with partial information.
References
[1] J. Dias and P. Marques, "Multiple Moving target detection and trajectory estimation using a single SAR sensor", IEEE Transactions on Aerospace and Electronic Systems, vol. 39, no. 2, pp. 604-624, 2003.
[FIDR02] G. Franceschetti, A. Iodice, D.Riccio, and G. Ruello. "SAR rawsignal simulation of oil slicks in ocean environments". IEEE Transactions on Geoscience and Remote Sensing, 40(9):1935-1949, 2002.
Signal and Image Processing in Synthetic Aperture Radar
Synthetic Aperture (SA) is an imaging technique that achieves high azimuth resolution by exploiting the relative motion between a platform-mounted antenna and the observation target field. Typical SA resolution are orders of magnitude higher than the correspondent real antenna resolution. SA is used, for example, in RADAR, SONAR, and LIDAR.This project addresses the following challenging research topics topics in SA:
Moving target detection and estimation; Moving target detection, estimation, and imaging is important for wide area surveillance systems with limited revisit times
Interferometry; The problem of interferometry is essentially the estimation of absolute phase from a pair (or more) SA images. Topographic maps are then obtained with basis on the estimated absolute phase images.
Image restoration; The pixel amplitude of SA images are, under mild conditions, a sample of a Rayleigh random variable. Therefore, the image amplitudes corresponding to an homogeneous region exhibit random fluctuations, termed "speckled noise".
The following goals are pursued, concerning each of the three referred topics:
Developing solutions to the so-called "blind angle ambiguity" problem. It means that using just an antenna it is impossible to infer the direction of the velocity vector. One way to remove the blind angle ambiguity is to make simultaneous bistatic SA measurements with two different antennas. Aiming at the resolution of the blind angle ambiguity problem with only one antenna, it is proposed the exploitation of recent results, derived by the proponents, on the relations between the received signal and the moving target velocity.
Developing Bayesian approaches to the absolute phase estimation in the presence of discontinuities. The inclusion of discontinuities is relevant to obtain meaningful estimates. On the other hand, most methods proposed in the open literature do not take discontinuities into account.
Developing Bayesian approaches to the amplitude restoration that, as in item (2), take the amplitude discontinuity into account.The Bayesian perspective, based on the data generation model and on existing "a priori" knowledge, is common to methodologies to follow in the items (1), (2), and (3). In the case of interferometry and amplitude restoration, "compound Markov random fields" are adopted as "a priori" probability density function. This model enforces smoothness on homogeneous regions and simultaneously preserves discontinuities between neighbouring regions.