Vous êtes ici : Accueil » Kiosque » Annonce


Mot de passe : 

Mot de passe oublié ?
Détails d'identification oubliés ?


24 novembre 2020

M2 internship: Anomaly detection schemes in SAR imaging

Catégorie : Stagiaire

Contact: Sébastien Angélliaume, Chengfang Ren, Jean-Philippe Ovarlez

sebastien.angelliaume@onera.fr, chengfang.ren@centralesupelec.fr, jean-philippe.ovarlez@onera.fr

M2 internship in the SONDRA laboratory at Centralesupelec and at ONERA
Starting date: between January and April 2021
Duration: 5 to 6 months
To apply, send a CV and a short description of your motivation to the supervisors
A doctoral thesis can be continued after this internship.
Keywords: Anomaly Detection, Robust detection, Synthetic Aperture Radar (SAR)
The topic description is also available here


Anomaly detection aims to discover abnormal patterns hidden in multidimensional radar signals and images.
This research field is essential in data mining for quickly isolating irregular or suspicious segments in large amounts
of the database. Some examples are given below: a) Oil Slick, b) Turbulent ship wake, c) Levee anomaly, d)
Archeology. Among them, the unsupervised methods are the most interesting since they are widely applicable and do not require to label the

In this framework, we are focused on statistical detection, distance-based and density-based methods that
constitute the main families of anomaly detection methods. The latter already has remarkable success for detecting
anomalies in biological data [3, 4], hyperspectral data [5, 6, 7, 8, 9], etc. However, none of them is well adapted
for SAR imaging applications since (i) the diversity of SAR images (i.e. the number of channels, frequency bands,
etc.) is reduced compared to hyperspectral images rendering the extraction process of features limited. (ii) The
noise distribution is generally heavy-tailed, notably for high-resolution SAR images. (iii) The signal to noise ratio
is weaker compared to optical images. Therefore, the discrimination between a real anomaly and the noise is more
challenging than for RGB images. Note that there is some marginal tentative in the literature [10, 11].
In this thesis project, we propose to design new schemes tackling these issues and limitations. More specifically,
the goal of this thesis is to develop efficient and robust methods to extract abnormal ROIs in SAR images. We
firstly propose to handle heavy-tailed noise with the family of elliptical distributions [12, 13], which better fit the
noise distribution. Secondly, statistical and pixel-wise distances will be improved by using a natural Riemannian
metric rather than the standard Euclidean metric for better detecting abnormal ROIs. Under the availability of
labeled data, metric learning approaches [14] are also under the scope. Additionally, whitening and denoising
preprocessing steps are going under investigation to mitigate noise power. Finally, the developed methods will
1be applied for detecting anomalies in SAR images obtained from SETHI, Sentinel-1, UAVSAR and TerraSAR-X
missions. Notably, these proposed methods could be used for monitoring intrusion in vegetation zones.
[1] Varun Chandola, Arindam Banerjee, and Vipin Kumar. Anomaly detection: A survey. ACM computing surveys
(CSUR), 41(3):1–58, 2009.
[2] Hongzhi Wang, Mohamed Jaward Bah, and Mohamed Hammad. Progress in outlier detection techniques: A
survey. IEEE Access, 7:107964–108000, 2019.
[3] James W Macdonald and Debashis Ghosh. Copa—cancer outlier profile analysis. Bioinformatics, 22(23):2950–
2951, 2006.
[4] Robert Tibshirani and Trevor Hastie. Outlier sums for differential gene expression analysis. Biostatistics,
8(1):2–8, 2007.
[5] Joana Frontera-Pons, Miguel Angel Veganzones, Frédéric Pascal, and Jean-Philippe Ovarlez. Hyperspectral
anomaly detectors using robust estimators. IEEE Journal of Selected Topics in Applied Earth Observations
and Remote Sensing, 9(2):720–731, 2015.
[6] Joana Frontera-Pons, Miguel Angel Veganzones, Santiago Velasco-Forero, Frédéric Pascal, Jean Philippe Ovar-
lez, and Jocelyn Chanussot. Robust anomaly detection in hyperspectral imaging. In 2014 IEEE Geoscience
and Remote Sensing Symposium, pages 4604–4607. IEEE, 2014.
[7] Miguel Angel Veganzones, Joana Frontera-Pons, Frédéric Pascal, J-P Ovarlez, and Jocelyn Chanussot. Bi-
nary partition trees-based robust adaptive hyperspectral rx anomaly detection. In 2014 IEEE International
Conference on Image Processing (ICIP), pages 5077–5081. IEEE, 2014.
[8] Eugénie Terreaux, Jean-Philippe Ovarlez, and Frédéric Pascal. Anomaly detection and estimation in hyperspec-
tral imaging using random matrix theory tools. In 2015 IEEE 6th International Workshop on Computational
Advances in Multi-Sensor Adaptive Processing (CAMSAP), pages 169–172. IEEE, 2015.
[9] Ahmad W Bitar, Jean-Philippe Ovarlez, and Loong-Fah Cheong. Sparsity-based cholesky factorization and its
application to hyperspectral anomaly detection. In 2017 IEEE 7th International Workshop on Computational
Advances in Multi-Sensor Adaptive Processing (CAMSAP), pages 1–5. IEEE, 2017.
[10] Yuval Haitman, Itay Berkovich, Shiran Havivi, Shimrit Maman, Dan G Blumberg, and Stanley R Rotman.
Machine learning for detecting anomalies in sar data. In 2019 IEEE International Conference on Microwaves,
Antennas, Communications and Electronic Systems (COMCAS), pages 1–5. IEEE, 2019.
[11] Florian Mouret, Mohanad Albughdadi, Sylvie Duthoit, Denis Kouamé, Guillaume Rieu, and Jean-Yves
Tourneret. Detecting anomalous crop development with multispectral and sar time series using unsupervised
outlier detection at the parcel-level: application to wheat and rapeseed crops. 2020.
[12] Esa Ollila, David E Tyler, Visa Koivunen, and H Vincent Poor. Complex elliptically symmetric distributions:
Survey, new results and applications. IEEE Transactions on signal processing, 60(11):5597–5625, 2012.
[13] Bruno Meriaux, Chengfang Ren, Mohammed Nabil El Korso, Arnaud Breloy, and Philippe Forster. Asymptotic
performance of complex m-estimators for multivariate location and scatter estimation. IEEE Signal Processing
Letters, 26(2):367–371, 2019.
[14] Eric Xing, Michael Jordan, Stuart J Russell, and Andrew Ng. Distance metric learning with application to
clustering with side-information. Advances in neural information processing systems, 15:521–528, 2002.

Dans cette rubrique

(c) GdR 720 ISIS - CNRS - 2011-2020.