Introduction to the Issue on Advances in Remote ... - IEEE Xplore

IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 5, NO. 3, JUNE 2011

365

Introduction to the Issue on Advances in Remote Sensing Image Processing “Satellite’s gone up to the skies Things like that drive me out of my mind” —Lou Reed

O

UR planet is continuously observed by airborne and satellite sensors that acquire tons of data to be processed and analyzed daily. The statistical characterization of remote sensing images turns out to be more difficult than in grayscale natural images because of the pixel’s higher dimensionality, particular noise and uncertainty sources, the high spatial and spectral redundancy, and their inherently nonlinear nature. It is worth noting that all these problems can be addressed in different ways depending on the sensor and the acquisition process. Consequently, the methods for the analysis and processing of remote sensing images need to be carefully designed attending to these needs. Different problems are posed from a signal processing and machine learning point of view: the acquired signals have to be processed in a timely manner, transmitted, further corrected from different distortions, eventually compressed, and ultimately analyzed to extract valuable information from them with, for instance, advanced classification or regression methods. Recently, new learning paradigms have been introduced and the latest advances in signal and image processing tools have been incorporated to the current toolbox of the remote sensing data users. It is a great pleasure for us to introduce this special issue on remote sensing image and signal processing. The goal is to summarize the recent advances in the field in a comprehensive manner, but also we would like it to promote the cross-fertilization between the remote sensing and the signal processing communities, which we foresee it as increasingly necessary. The response to the call for papers for this special issue was extraordinary, and finally 22 papers have been accepted for publication. They cover the most relevant steps of the remote sensing data processing chain: image coding, feature extraction and selection, advances in optical and radar signal processing, sparse signal analysis, image denoising, signal unmixing, image fusion, target detection, and data classification. The issue contains contributions in all steps of the chain; see Fig. 1. A. Data Transmission and Storage Along with the increasing demand of hyperspectral data, the sensor technology used to acquire the images has been significantly developed, improving, among others, the spatial and spectral resolution. Such improvement on quality leads to an increasing demand on storage and bandwidth transmission Digital Object Identifier 10.1109/JSTSP.2011.2142490

Fig. 1. Remote sensing image analysis chain.

capabilities. Both lossy and lossless image coding have been investigated extensively. The paper “Successive Approximation Wavelet Coding of AVIRIS Hyperspectral Images” by Dutra et al. presents a couple of compression algorithms which build on a state-of-the-art codec, the Set Partitioned Embedded Block Coder (SPECK), by incorporating a lattice vector quantizer code-book. This allows processing multiple samples at one time. The authors also tackle the relevant problem of reducing the number of codewords in a very efficient way. B. Image and Signal Processing Analyzing the particular characteristics of the remote sensing data is a very active research field. While optical image processing exploits most of the tools of grayscale and color image processing methods generalized to the mulispectral case, in the case of active radar imagery, the acquired signals must be processed with methods that exploit the specific statistical properties of the data. In the paper “On the Empirical-Statistical Modeling of SAR Images with Generalized Gamma Distribution” by Li et al., an efficient statistical model based on the generalized Gamma distribution is proposed to characterize synthetic aperture radar (SAR) images. This model of the probability density functions (pdfs) is computationally efficient, and thus the proposed method is very convenient for online SAR image processing. In the work “Displacement Estimation by Maximum Likelihood Texture Tracking” by Harant et al., a novel method is presented to estimate displacement by maximum-likelihood (ML) texture tracking. The work is motivated by the fact that with the new generation of launched PolSAR sensors, the Earth’s surface is imaged with meter resolution. This property enables identifying very small spatial features from the space so texture analysis and statistical modeling is becoming more and more important to develop tracking methods from these data. Identifying distinct image features in radar data and hyperspectral images is crucial for many applications. The field of sparse signal processing has emerged in the last years, and two relevant works within this context are presented in this issue. In the last years, we observed an increasing interest in 3-D reconstruction of scenes from radar measurements. Traditional 3-D SAR image formation requires data collection over a densely sampled azimuth-elevation sector, which is very difficult or

1932-4553/$26.00 © 2011 IEEE

366


even impossible in practice. The work “Sparse Signal Methods for 3-D Radar Imaging” by Austin et al. proposes models for 3-D reconstruction that exploit reconstruction sparsity to tackle the limitations of sparse measurements. On the other hand, the paper “Sparse matrix transform for hyperspectral image processing” by Theiler et al. studies the important issue of estimating the data covariance matrix in poorly-sampled spaces. The authors analyze a particular estimator, named sparse matrix transform, in several problems: detection, dimension reduction, anomaly detection, and anomalous change detection. Following an alternative approach, the paper “Statistical inference in PCA for hyperspectral images” by Bajorski studies the sampling properties of the covariance matrix of hyperspectral images in terms of sampling, noise, and variability. Confidence intervals for the estimation of the eigenvalue problem are also given.

estimating hyperspectral sensors random noise components, both dependent and independent from the signal. The main advantage of the proposed method is its ability to accurately estimate band noise variances locally by using spatial and spectral texture correlations. Locality of the method tries to accommodate nonlinear signal–noise relations. Alternative approaches consider the use of nonlinear methods, as in the paper “Noise Reduction of Hyperspectral Images Using Kernel Nonnegative Tucker Decomposition” by Karami et al., where a new noise reduction algorithm based on kernels and genetic algorithms is proposed. Finally, the paper “Suppressing Moving Target Artifacts in Multi-Channel Stripmap SAR Images by Space-Doppler Filtering” by Schulz proposes a joint spatial-spectral processing to suppress echoes from moving targets in multi-channel strip-map SAR data while preserving the echoes of the fixed scene.

C. Fusion and Pansharpening Spatial resolution of sensors is often limited with respect to their spectral resolution. Multi- or hyperspectral sensors give a unique amount of spectral information, but they often lack the spatial detail necessary for the application. On the contrary, panchromatic sensors provide information with higher level of spatial detail, but lack spectral information. Since the design of a high-resolution sensor in both spectral and spatial domains would be extremely costly and challenging in terms of engineering, image fusion methods are often employed to create an image taking advantage of both panchromatic and multi- or hyperspectral sensors. The work “A Theoretical Analysis of the Effects of Aliasing and Misregistration on Pansharpened Imagery” by Baronti et al. pays attention to the issues of aliasing and misregistration errors in multiresolution fusion algorithms. Several methods are analyzed and interesting conclusions drawn: under mild assumptions, aliasing and/or misregistration do not greatly affect fusion products, especially in terms of spatial quality. D. Feature Selection and Extraction When dealing with high-dimensional datasets, such as hyperspectral images, the computational time of the data analysis is increased and the high collinearity and presence of noisy bands can degrade the quality of the model. Feature selection and extraction are central issues in these situations because of the curse of dimensionality. In the paper “Optimal Feature Set for Automatic Detection and Classification of Underwater Objects in SAS Images” by Fandos and Zoubir, the problem of automatic detection and classification for mine hunting applications is addressed with special focus on the extraction of relevant discriminative features: from shape and shadow descriptors to accurate image statistics. E. Image Restoration and Enhancement Image restoration is an important step in the image processing chain. Several problems are encountered in this application: different noise sources and amounts are present in the data and scattered either in the spatial or specific spectral bands. This makes necessary appropriate spatial–spectral image restoration and enhancement steps. The paper “Local Signal-Dependent Noise Variance Estimation From Hyperspectral Textural Images” by Uss et al. presents a maximum-likelihood method for

F. Spectral Unmixing The issue of finding the remote sensing image basis is a central research topic. Hence, the development of automatic extraction of spectral endmembers directly from the input hyperspectral data set has captured the attention of researchers. When pure pixels are identified in the image, all pixels can be synthesized as a linear (or nonlinear) combination of them, and this, in turn, allows for example subpixel detection or mapping. Some classic techniques for this purpose include vertex component algorithms and orthogonal subspace projections among others. The paper “Maximum Orthogonal Subspace Projection Approach to Estimating the Number of Spectral Signal Sources in Hyperspectral Imagery” by Chang et al. tackles the relevant problem of estimating the number of sources present in the data. A new method is evaluated and theoretically related to previous works. Lately, the problem of including spatial information in the unmixing process, and the study of nonlinear unmixing have attracted the attention of researchers. These shortcomings are also addressed in this issue. In the work “Spectral Unmixing for the Classification of Hyperspectral Images at a Finer Spatial Resolution” by Villa et al., the problem of classification of hyperspectral images containing mixed pixels is addressed. The method exploits the advantages of both soft classification techniques and spectral unmixing algorithms in order to determine the fractional abundances of the classes at a sub-pixel scale. A spatial regularization by Simulated Annealing is finally performed to spatially locate the obtained classes. On the other hand, the nonlinear issue is addressed in “Nonlinear spectral unmixing by geodesic simplex volume maximization” by Heylen et al.. The paper presents an unmixing algorithm that can determine endmembers and their abundances in hyperspectral imagery under nonlinear mixing assumptions. The algorithm accounts for the nontrivial geometry of the data manifold in an efficient way. G. Clustering, Classification and Target Detection Classification maps are probably the main product of remote sensing image processing. Important applications include urban monitoring, land-cover mapping, damage assessment, change, and target detection. Three main groups of approaches can be found: unsupervised, supervised, and one-class classification.


On the one hand, unsupervised methods aim at clustering the image pixels into a predefined number of homogeneous and consistent groups. In “A New Weighted Fuzzy C-Means Clustering Algorithm for Image Classification in Remote Sensing” by Hung et al., the problem is approached with a fuzzy -means combined with nonparametric feature extraction. The method incorporates soft information for improved stability and performs consistently better than previous approaches. When labeled information is available, supervised methods may be used to recognize predefined classes. At present, this field is probably the most active in remote sensing image processing, and the issue contains several contributions in this area. In the paper “Supervised Very High Resolution Dual Polarization SAR Image Classification by Finite Mixtures, Copulas, and Markov Random Fields” by Krylov et al., a novel supervised classification approach is proposed for very high resolution single-channel and dual polarization amplitude satellite synthetic aperture radar (SAR) images. The proposed technique combines the Markov random field approach for Bayesian image classification with finite mixture modeling for probability density function estimation and copulas for multivariate distribution modeling. In “Statistical Classification for Heterogeneous Polarimetric SAR Images” by Formont et al., a general approach for high-resolution polarimetric SAR data classification in heterogeneous clutter, based on a statistical test of equality of covariance matrices. Comparison to standard approaches show very promising results. The paper “Classifying Vehicles in Wide-Angle Radar using Pyramid Match Hashing” by Dungan and Potter presents a fast and scalable method to simultaneously register and classify vehicles in circular synthetic aperture radar imagery. The method is robust to occlusions and partial matches, and achieves very high classification rates. The important problem of classification with illumination changes is addressed in “An Optimum Land Cover Mapping Algorithm In the Presence of Shadows” by Kasetkasem and Varshney. Results show that a large number of misclassified pixels can be corrected and shadowy regions can also be successfully reconstructed. When designing a supervised classifier, the performance of the model depends strongly on the quality of the labeled information. This constraint makes the generation of an appropriate training set a difficult and expensive task requiring extensive manual (and often subjective) human-image interaction. This phase is highly redundant, as well as time consuming. Active learning aims at responding to this need: active learning methods select the most relevant samples for training. The work “A Survey of Active Learning Algorithms for Supervised Remote Sensing Image Classification” by Tuia et al. presents a complete review of active learning methods in the field of remote sensing image classification. A theoretical comparison and experimental evaluation of the main families is included. Following this, the work “Active Learning via Multi-View and Local Proximity Co-Regularization for Hyperspectral Image Classification” by Di and Crawford introduces a novel co-regularization framework for active learning for hyperspectral image classification. The method exploits manifold learning through combined regularizers that deal with spatial and spectral information efficiently.

367

Finally, target and anomaly detection are important problems typically faced in remote sensing image processing. In these cases, there is only one class of interest to be identified in the scene. The work “Sparse Representation for Target Detection in Hyperspectral Imagery” by Chen et al. proposes a new sparsity-based algorithm for automatic target detection in hyperspectral images. The method imposes sparseness in the solution via the convenient solution of an -norm problem. Simulation results show that the algorithm outperforms the classical hyperspectral target detection algorithms, such as the popular spectral matched filters, matched subspace detectors, adaptive subspace detectors, as well as binary classifiers such as support vector machines. The papers in this special issue span a wide range of problems arising in modern remote sensing data analysis and provide a snapshot in the state-of-the-art of remote sensing image processing. More advances are expected in the near future, mainly due to the increasing user demands in terms of spatial, spectral, and temporal resolutions of data, and of products generated from these data by automatic processing techniques. We hope that the collection of papers in this special issue will inspire new approaches to address the important technical challenges in current and future remote sensing data processing. We would like to gratefully acknowledge the help of all people involved in the collation and review process of this issue, without whose support the project could not have been satisfactorily completed. We wish to thank all of the authors for their response letters, insights, and excellent contributions. Thanks go to all reviewers who provided with constructive and comprehensive reviews. A further special note of thanks goes also to the Editor-in-Chief, Prof. Vikram Krishnamurthy, for his support and guidance as well as the Managing Editor, Ms. Rebecca Wollman, who continuously prodded via e-mail trying to keep the project on schedule.

GUSTAVO CAMPS-VALLS, Lead Guest Editor Department of Electronics Engineering University of Valencia Valencia 46010, Spain JÓN ATLI BENEDIKTSSON, Guest Editor Department of Electrical and Computer Engineering University of Iceland 101 Reykjavik, Iceland LORENZO BRUZZONE, Guest Editor Department of Information Engineering and Computer Science University of Trento I-38122 Trento , Italy JOCELYN CHANUSSOT, Guest Editor GIPSA-Lab Grenoble Institute of Technology 38402 Grenoble, France

368


Gustavo Camps-Valls (M’04–SM’07) was born in València, Spain, in 1972. He received the B.Sc. degree in physics, the B.Sc. degree in electronics engineering, and the Ph.D. degree in physics from the Universitat de València in 1996, 1998, and 2002, respectively. He is currently an Associate Professor in the Department of Electronics Engineering, University of Valencia, where teaches electronics, advanced time series processing, and machine learning for remote sensing. He is also a leading researcher at the Image Processing Laboratory (IPL), and has been a visiting researcher at the Remote Sensing Laboratory (University of Trento, Italy) and at the Max Planck Institute for Biological Cybernetics (Tübingen, Germany). His research interests are tied to the development of machine learning algorithms for signal and image processing with a special focus on remote sensing data analysis. He conducts and supervises research within the frameworks of several national and international projects, and he is an Evaluator of project proposals and scientific organizations. He is the author (or coauthor) of 70 international peer-reviewed journal papers, more than 100 international conference papers, 20 international book chapters, and editor of the books Kernel Methods in Bioengineering, Signal and Image Processing (IGI, 2007) and Kernel Methods for Remote Sensing Data Analysis (Wiley, 2009). Prof. Camps-Valls is a referee of many international journals and conferences, and currently serves on the Program Committees of the International Society for Optical Engineers (SPIE) Europe, International Geoscience and Remote Sensing Symposium (IGARSS), International Workshop on Artificial Neural Networks (IWANN), Machine Learning for Signal Processing (MLSP), and International Conference on Image Processing (ICIP). Since 2007, he has been a member of the Data Fusion Technical Committee of the IEEE Geoscience and Remote Sensing Society, and since 2009 he has been member of the Machine Learning for Signal Processing Technical Committee of the IEEE Signal Processing Society. He is involved in the MTG-IRS Science Team (MIST) of the European Organization for the Exploitation of Meteorological Satellites (EUMETSAT). He is an Associate Editor of the ISRN Signal Processing Journal and the IEEE GEOSCIENCE AND REMOTE SENSING LETTERS. Visit http://www.uv.es/gcamps for more information.

Jón Atli Benediktsson (S’84–M’90–SM’99–F’04) received the Cand.Sci. degree in electrical engineering from the University of Iceland, Reykjavik, in 1984 and the M.S.E.E. and Ph.D. degrees from Purdue University, West Lafayette, IN, in 1987 and 1990, respectively. He joined the Department of Electrical and Computer Engineering, University of Iceland, in 1991 and is currently Pro Rector for Academic Affairs and Professor of Electrical and Computer Engineering at the University of Iceland. He has held visiting positions at the Department of Information and Communication Technology, University of Trento, Trento, Italy (2002–present), School of Computing and Information Systems, Kingston University, Kingston upon Thames, U.K. (1999–2004), the Joint Research Centre of the European Commission, Ispra, Italy (1998), Denmark’s Technical University (DTU), Lyngby (1998), and the School of Electrical and Computer Engineering, Purdue University (1995). He was a Fellow at the Australian Defence Force Academy, Canberra, A.C.T., Australia, in August 1997. From 1999 to 2004, he was a Chairman of the energy company Metan Ltd. His research interests are in remote sensing, pattern recognition, neural networks, image processing, and signal processing, and he has published extensively in those fields. Dr. Benediktsson was Editor (2003–2008) of the IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING (TGRS) and is currently an Associate Editor of TGRS and the IEEE GEOSCIENCE AND REMOTE SENSING LETTERS. From 1996 to 1999, he was the Chairman of the GRSS Technical Committee on Data Fusion and was elected to the Administrative Committee of the GRSS in 2000. He is the GRSS President for 2011. He was GRSS Vice President of Technical Activities in 2002, GRSS Vice President of Professional Activities in 2008–2009, and GRSS Executive Vice President for 2010. He was the founding Chairman of the IEEE Iceland Section (2000–2003). Currently, he is the Chairman of the University of Iceland’s Quality Assurance Committee (2006–present). He was the Chairman of the University of Iceland’s Science and Research Committee (1999–2005), a member of Iceland’s Science and Technology Council (2003–2006), and a member of the Nordic Research Policy Council (2004). He was a member of a NATO Advisory Panel of the Physical and Engineering Science and Technology SubProgramme (2002–2003). He received the Stevan J. Kristof Award from Purdue University in 1991 as an outstanding graduate student in remote sensing. In 1997, he was the recipient of the Icelandic Research Council’s Outstanding Young Researcher Award; in 2000, he was granted the IEEE Third Millennium Medal; in 2004, he was a co-recipient of the University of Iceland’s Technology Innovation Award; in 2006, he received the yearly research award from the Engineering Research Institute of the University of Iceland; and in 2007, he received the Outstanding Service Award from the IEEE Geoscience and Remote Sensing Society. He is a co-founder of the start up company Oxymap and member of Societas Scinetiarum Islandica and Tau Beta Pi.


369

Lorenzo Bruzzone (S’95–M’98–SM’03–F’10) received the laurea (M.S.) degree in electronic engineering (summa cum laude) and the Ph.D. degree in telecommunications from the University of Genoa, Genoa, Italy, in 1993 and 1998, respectively. He is currently a Full Professor of telecommunications at the University of Trento, Trento, Italy, where he teaches remote sensing, pattern recognition, radar, and electrical communications. He is the Head of the Remote Sensing Laboratory in the Department of Information Engineering and Computer Science, University of Trento. His current research interests are in the areas of remote sensing, radar and SAR, signal processing, and pattern recognition. He conducts and supervises research on these topics within the frameworks of several national and international projects. He is the author (or coauthor) of 101 scientific publications in referred international journals (68 in IEEE journals), more than 150 papers in conference proceedings, and 15 book chapters. He is editor/coeditor of many books and conference proceedings. He is a referee for many international journals and has served on the scientific committees of several international conferences. He is a member of the Managing Committee of the Italian Inter-University Consortium on Telecommunications and a member of the Scientific Committee of the India–Italy Center for Advanced Research. Since 2009, he has been a member of the Administrative Committee of the IEEE Geoscience and Remote Sensing Society. Dr. Bruzzone ranked first place in the Student Prize Paper Competition of the 1998 IEEE International Geoscience and Remote Sensing Symposium (Seattle, July 1998). He was a recipient of the Recognition of the IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING (TGRS) Best Reviewers in 1999 and was a Guest Co-Editor of different Special Issues of the IEEE TGRS. In the past years, joint papers presented by his students at international symposia and master theses that he supervised have received international and national awards. He was the General Chair and Co-chair of the First and Second IEEE International Workshop on the Analysis of Multi-temporal Remote-Sensing Images (MultiTemp), and is currently a member of the Permanent Steering Committee of this series of workshops. Since 2003, he has been the Chair of the SPIE Conference on Image and Signal Processing for Remote Sensing. From 2004 to 2006, he served as an Associated Editor of the IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, and currently is an Associate Editor for the IEEE TGRS and the Canadian Journal of Remote Sensing. Since April 2010, he has been the Editor of the IEEE Geoscience and Remote Sensing Newsletter. In 2008, he has been appointed as a member of the joint NASA/ESA Science Definition Team for the radar instruments for Outer Planet Flagship Missions. He is a member of the Italian Association for Remote Sensing (AIT).

Jocelyn Chanussot (M’04–SM’04) received the M.Sc. degree in electrical engineering from the Grenoble Institute of Technology (Grenoble INP), Grenoble, France, in 1995, and the Ph.D. degree from Savoie University, Annecy, France, in 1998. In 1999, he was with the Geography Imagery Perception Laboratory for the Delegation Generale de l’Armement (DGA—French National Defense Department). Since 1999, he has been with Grenoble INP, where he was an Assistant Professor from 1999 to 2005, an Associate Professor from 2005 to 2007, and is currently a Professor of signal and image processing. He is currently conducting his research at the Grenoble Images Speech Signals and Automatics Laboratory (GIPSALab). His research interests include image analysis, multicomponent image processing, nonlinear filtering, and data fusion in remote sensing. Dr. Chanussot is the founding President of IEEE Geoscience and Remote Sensing French chapter (2007–2010) which received the 2010 IEEE GRS-S Chapter Excellence Award “for excellence as a Geoscience and Remote Sensing Society chapter demonstrated by exemplary activities during 2009.” He was a member of the IEEE Geoscience and Remote Sensing AdCom (2009–2010), in charge of membership development. He was the General Chair of the first IEEE GRSS Workshop on Hyperspectral Image and Signal Processing, Evolution in Remote Sensing (WHISPERS). He is the Chair (2009–2011) and was the Cochair of the GRS Data Fusion Technical Committee (2005–2008). He was a member of the Machine Learning for Signal Processing Technical Committee of the IEEE Signal Processing Society (2006–2008) and the Program Chair of the IEEE International Workshop on Machine Learning for Signal Processing, (2009). He was an Associate Editor for the IEEE GEOSCIENCE AND REMOTE SENSING LETTERS (2005–2007) and for Pattern Recognition (2006–2008). Since 2007, he has been an Associate Editor for the IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. Since 2011, he has been the Editor-in-Chief of the IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING. He is a Senior Member of the IEEE (2004).