This paper highlights a series of recent developments in morphological image processing that are of interest for the analysis of remote sensing data. Each selected development is briefly presented while emphasising its usefulness in the context of real applications. The following topics are covered: grey level hit-or-miss transforms applied to the detection of specific man-made structures in panchromatic images, iterative area filtering for multichannel image simplification, constrained connectivity applied to hierarchical image partitioning of very high resolution images, mosaicing of an arbitrary number of images by automatically finding seam lines following salient image structures, and morphological segmentation of binary patterns applied to the characterisation of land cover classes. The paper concludes with challenges ahead and an open question.

In this paper we propose to model the structural information in very high geometrical resolution optical images with morphological attribute filters. In particular we propose to perform a multilevel analysis based on different features of the image in contraposition to the use of conventional morphological profiles. We show how morphological attribute filters are conceptually and experimentally more capable to describe the characteristics of buildings with respect to morphological filters by reconstruction. Furthermore, we address the issue of selecting the most suitable parameters of the filters by proposing an architecture which embeds in the filtering procedure an optimization step based on genetic algorithms. The effectiveness of the proposed technique is stated by the experiments which were carried out on a panchromatic image acquired by the Quickbird satellite.

Data-driven unsupervised segmentation of high resolution remotely sensed images is a primary step in understanding remotely sensed images. A new fully automatic method to delineate the segments corresponding to objects in high resolution remotely sensed images is introduced. There are extensive methods proposed in the literature which are mainly concentrated on pixel level information. The proposed method combines the structural information extracted by morphological processing with feature space analysis based on mean shift algorithm. The spectral and spatial bandwidth parameters of mean shift are adaptively determined by exploiting differential morphological profile (DMP). Spectral bandwidth is determined in relation to the first maximum value of DMP at each pixel and spatial bandwidth is determined by the corresponding index in DMP. In this method there is also no need to specify initially the maximum size of the structuring element for the morphological processes. By the use of mean shift filtering, the feature space points are grouped together which are close to each other both in the range of spatial and spectral bandwidths. The proposed method is applied on panchromatic high resolution QuickBird satellite images taken from urban areas. The results we obtained appear to be effective in terms of segmentation and combining the spectral and spatial information to extract more precise and more meaningful objects compared to fixed bandwidth mean shift segmentation.

We present a fast pan-sharpening method, namely FWLS, which is based on unsupervised segmentation of the original multispectral (MS) data for improved parameter estimation in a weighted least square fusion scheme. The use of simple thresholding of the normalized difference vegetation index (NDVI) dramatically reduces the computation time with respect to the recently proposed WLS method which is based on accurate supervised classification through kernel support vector machines. The fusion performances of the FWLS algorithm are the same that those obtained by the WLS algorithm, and even higher in some cases, since accurate extraction of vegetated/non-vegetated areas is only needed and high-performance supervised classification is generally not required for fusion parameter estimation. Experimental results and comparisons to state-of-the-art fusion methods are reported on Ikonos and QuickBird data. Both visual and objective quality assessment of the fusion results confirm the validity of the proposed FWLS algorithm.

Advances in hardware and pattern recognition techniques, along with the widespread utilization of remote sensing satellites, have urged the development of automatic target detection systems in satellite images. Automatic detection of airports is particularly essential, due to the strategic importance of these targets. In this paper, a runway detection method using a segmentation process based on textural properties is proposed for the detection of airport runways, which is the most distinguishing element of an airport. Several local textural features are extracted including not only low level features such as mean, standard deviation of image intensity and gradient, but also Zernike Moments, Circular-Mellin Features, Haralick Features, as well as features involving Gabor Filters, Wavelets and Fourier Power Spectrum Analysis. Since the subset of the mentioned features, which have a role in the discrimination of airport runways from other structures and landforms, cannot be predicted trivially, Adaboost learning algorithm is employed for both classification and determining the feature subset, due to its feature selector nature. By means of the features chosen in this way, a coarse representation of possible runway locations is obtained. Promising experimental results are achieved and given.

While most remote sensing data are in images, and thus image processing techniques are most often considered, the roles of signal processing, especially of those emerging or advanced signal processing techniques received much less attention. In this tutorial paper, we will examine four advanced signal processing techniques and assess their roles in remote sensing: vector time series analysis, compressed sensing, the new component analysis using non-negative matrix and tensor factorization and the Hilbert-Huang transform. The strength of each approach and its role in remote sensing are presented.

The interpretation of complex scenes requires a large amount of prior knowledge and experience. To utilize prior knowledge in a computer vision or a decision support system for image interpretation, a probabilistic scene model for complex scenes is developed. In conjunction with a model of the observer's characteristics (a human interpreter or a computer vision system), it is possible to support bottom-up inference from observations to interpretation as well as to focus the attention of the observer on the most promising classes of objects. The presented Bayesian approach allows rigorous formulation of uncertainty in the models and permits manifold inferences, such as the reasoning on unobserved object occurrences in the scene. Monte-Carlo methods for approximation of expectations from the posterior distribution are presented, permitting the efficient application even for high-dimensional models. The approach is illustrated on the interpretation of airfield scenes.

Most modern methods of image processing exploit a priori knowledge or estimates of noise type and its characteristics obtained in blind or interactive manner. However, the results of noise type blind determination can be false with some hopefully rather small probability. Similarly, the obtained estimates of noise parameters are characterized by certain accuracy. Clearly, false decisions and errors of estimates influence performance of image processing techniques that exploit the information on noise properties obtained in a blind manner. In this paper, we consider some aspects of such influence for several typical applications.

Image registration is a major issue in the field of Remote Sensing because it provides a support for integrating information from two or more images into a model that represents our knowledge on a given application. It may be used for comparing the content of two segmented images captured by the same sensor at different times; but it also may be used for extracting and assembling information from images captured by various sensors corresponding to different modalities (optical, radar,). The registration of images from different modalities is a very difficult problem because data representations are different (e.g. vectors for multispectral images and scalar values for radar ones) but also, and especially, because an important part of the information is different from an image to another (e.g. hyperspectral signature and radar response). And precisely, any registration process is based, explicitly or not, on matching the common information in the two images. The problem we are interested in is to develop a generic approach that enables the registration of two images from different modalities when their spatial representations are related by a rigid transformation. This situation often occurs, and it requires a very robust and accurate registration process to provide the spatial correspondence. First, we show that this registration problem between images from different modalities can be reduced to a matching problem between binary images. There are many approaches to tackle this problem, and we give an overview of these approaches. But we have to take into account the specificity of the context in which we have to solve this problem: we must select those points of both images that are associated with the same information, and not the other ones, in order to process the pairing that will lead to the registration parameters. The approach we propose is a Hough-like method that induces a separation between relevant and non-relevant pairings, the Hough space being a representation of the rigid transformation parameters. In order to characterize the relevant items in each image, we propose a new primitive that provides a local representation of patterns in binary images. We give a complete description of this approach and results concerning various types of images to register.

High Performance Computing (HPC) hardware solutions such as grid computing and General Processing on a Graphics Processing Unit (GPGPU) are now accessible to users with general computing needs. Grid computing infrastructures in the form of computing clusters or blades are becoming common place and GPGPU solutions that leverage the processing power of the video card are quickly being integrated into personal workstations. Our interest in these HPC technologies stems from the need to produce near real-time maps from a combination of pre- and post-event satellite imagery in support of post-disaster management. Faster processing provides a twofold gain in this situation: 1. critical information can be provided faster and 2. more elaborate automated processing can be performed prior to providing the critical information. In our particular case, we test the use of the PANTEX index which is based on analysis of image textural measures extracted using anisotropic, rotation-invariant GLCM statistics. The use of this index, applied in a moving window, has been shown to successfully identify built-up areas in remotely sensed imagery. Built-up index image masks are important input to the structuring of damage assessment interpretation because they help optimise the workload. The performance of computing the PANTEX workflow is compared on two different HPC hardware architectures: (1) a blade server with 4 blades, each having dual quad-core CPUs and (2) a CUDA enabled GPU workstation. The reference platform is a dual CPU-quad core workstation and the PANTEX workflow total computing time is measured. Furthermore, as part of a qualitative evaluation, the differences in setting up and configuring various hardware solutions and the related software coding effort is presented.

In this paper, we cover a decade of research in the field of spectral-spatial classification in hyperspectral remote sensing. While the very rich spectral information is usually used through pixel-wise classification in order to recognize the physical properties of the sensed material, the spatial information, with a constantly increasing resolution, provides insightful features to analyze the geometrical structures present in the picture. This is especially important for the analysis of urban areas, while this helps reducing the classification noise in other cases. The very high dimension of hyperspectral data is a very challenging issue when it comes to classification. Support Vector Machines are nowadays widely aknowledged as a first choice solution. In parallel, catching the spatial information is also very challenging. Mathematical morphology provides adequate tools: granulometries (the morphological profile) for feature extraction, advanced filters for the definition of adaptive neighborhoods, the following natural step being an actual segmentation of the data. In order to merge spectral and spatial information, different strategies can be designed: data fusion at the feature level or decision fusion combining the results of a segmentation on the one hand and the result of a pixel wise classification on the other hand.

This paper introduces a new semi-supervised Bayesian approach to hyperspectral image segmentation. The algorithm mainly consists of two steps: (a) semi-supervised learning, by using the LORSAL algorithm to infer the class distributions, followed by (b) segmentation, by inferring the labels from a posterior density built on the learned class distributions and on a Markov random field. Active label selection is performed. Encouraging results are presented on real AVIRIS Indiana Pines data set. Comparisons with state-of-the-art algorithms are also included.

Independent component analysis (ICA) has been widely used for hyperspectral image classification in an unsupervised fashion. It is assumed that classes are statistically mutual independent. In practice, this assumption may not be true. In this paper, we apply dependent component analysis (DCA) to unsupervised classification, which does not require the class independency. The basic idea of our DCA approaches is to find a transform that can improve the class independency but leave the basis mixing matrix unchanged; thus, an original ICA method can be employed to the transformed data where classes are less statistically dependent. Linear transforms that possess such a required invariance property and generate less dependent sources include: high-pass filtering, innovation, and wavelet transforms. These three transforms correspond to three different DCA algorithms, which will be investigated in this paper. Preliminary results show that the DCA algorithms can slightly improve the classification accuracy.

Information extraction from hyperspectral imagery is highly affected by difficulties in accounting for flux density variation and Bidirectional reflectance effects. However, its full implementation requires extremely detailed information regarding the spatial structures or mini-structure of each material. This information is frequently not available at the accuracy needed (if it even exists). Thus, reflectance estimations for hyperspectral images will not fully account for flux density effects and consequently, the reflectance of the same surface material would vary, resulting in increased spectral confusion. Utilization of normalization, band selection, ratioing, spectral angle (SAM), and derivative techniques for this purpose provide only partial solutions under unknown illumination conditions. In this work we introduce a novel signal processing approach, based on wavelet analysis, aimed at reducing the effects of flux density variations on imagery objects' identification. Wavelet analysis is a space localized periodic analysis tool, which enables analysis of a signal in both spectral and frequency domains. This new technique is based on the observation that detailed wavelet coefficients, which result from wavelet decomposition, vary linearly with increasing scaling level. Since both the coefficient of variation of these linear relationships (a) and reflectance (R) at each wavelength position are affected by flux density, their ratio (R2a) was hypothesized to be invariant to flux density effects in particular and multiplicative effects in general. Advantage of this method was supported by higher accuracies and reliabilities gained for classifying with R2a when compared to classification of the real spectral images of Mediterranean and domestic plants and lithological formations.

This paper addresses the problem of unmixing hyperspectral images, when the light suffers multiple interactions among distinct endmembers. In these scenarios, linear unmixing has poor accuracy since the multiple light scattering effects are not accounted for by the linear mixture model. Herein, a nonlinear scenario composed by a single layer of vegetation above the soil is considered. For this class of scene, the adopted mixing model, takes into account the second-order scattering interactions. Higher order interactions are assumed negligible. A semi-supervised unmixing method is proposed and evaluated with simulated and real hyperspectral data sets.

Performance of the matched filter and anomaly detection algorithms relies on the quality of the inverse sample covariance matrix, which depends on sample size (number of vectors). The "RMB rule" provides the number of vectors required to achieve a specific average performance loss of the matched filter. In this paper we extend the RMB rule to provide the number of vectors needed to ensure a minimum performance loss (within a certain confidence). We also review a general metric for covariance estimation accuracy based on the Wishart distribution and discuss anomaly detector performance loss.

The benchmark anomaly detection algorithm for hyperspectral images is the Reed-Xiaoli (RX) Detector, which is based on the Local Multivariate Normality of background. RX algorithm, along with its many modified versions, has been widely explored, and the main concerns identified are related to local background covariance matrix estimation. Besides the well-known small-sample size problem, other limitations have been found affecting covariance matrix estimation, e.g. local background non-homogeneity and contamination from adjacent targets. These critical aspects are deeply different in nature, like the situations from which they arise, and hence they have been typically discussed within different frameworks, disregarding possible existing links while developing different approaches to solution. Nevertheless, these critical aspects may occur together in reality, and all of them have to be taken into consideration when approaching anomaly detection, since they may strongly affect detection performance. Therefore, an analysis of the possible existing connections seems crucial in order to asses if existing algorithms, maybe designed ad-hoc to solve a specific problem, can handle more complex situations. In this work, the aforementioned limitations have been investigated from an anomaly detection perspective, and the corresponding approaches to improved covariance matrix estimation have been analyzed by using real hyperspectral data.

Spectral unmixing is an important tool for interpreting remotely sensed hyperspectral scenes with sub-pixel precision. It relies on the identification of a set of spectrally pure components (called endmembers) and the estimation of the fractional abundance of each endmember in each pixel of the scene. Fractional abundance estimation is generally subject to two constraints: non-negativity of estimated fractions and sum-to-one for all abundance fractions of endmembers in each single pixel. Over the last decade, several algorithms have been proposed for simultaneous and sequential extraction of image endmembers from hyperspectral scenes. In this paper, we develop a new sequential algorithm that automatically extracts endmembers by using an unconstrained linear mixture model. Our assumption is that fractional abundance estimation using a set of properly selected image endmembers should naturally incorporate the constraints mentioned above, while imposing the constraints for an inadequate set of spectral endmembers may introduce errors in the model. Our proposed approach first applies an unconstrained linear mixture model and then uses a new metric for measuring the deviation of the unconstrained model with regards to the ideal, fully constrained model. This metric is used to derive a set of spectral endmembers which are then used to unmix the original scene. The proposed algorithm is experimentally compared to other algorithms using both synthetic and real hyperspectral scenes collected by NASA/JPL's Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS).

Juventae Chasma is a depression north of Valles Marineris on Mars, approximately 185 km wide and 270 km long. It contains several mounds of light-toned, layered deposits several tens of kilometers of maximum extension and up to 3300 m in elevation. Near infrared spectral data from the Observatoire pour l'Eau, des Glaces et l'Activité onboard ESA's Mars Express indicated mono- and polyhydrated sulfates as main constituents of these deposits, including gypsum in one of the mounds (Gendrin et al., 2005, Science). We analyze the light-toned outcrops based on data from NASA's Compact Reconnaissance Imaging Spectrometer for Mars (CRISM), featuring an increased spatial resolution of up to 18m/pixel and increased spectral resolution of 7 nm. We perform Spectral Mixture Analysis (SMA) in order to introduce physical modeling and to enhance some surface units. We use one type of SMA, the Multiple-Endmember Linear Unmixing Model MELSUM (Combe et al., 2008, PSS), which guarantees positive mixing coefficients and allows us to limit the number of spectral components used at a time. We use linear unmixing both as a similarity measure using spectra from the image itself as endmembers to assess the internal variability of the data, and to detect mineral spectra within the observations. We successfully confirm the presence of the monohydrated sulfate szomolnokite (previously detected by Kuzmin et al., 2008, and Rossi et al., 2008) in all of the four light-toned deposits observed. Based on our analysis, we reject the presence of gypsum on mound B (previously detected by Gendrin et al., 2005). A possible match for the polyhydrated sulfate present here could be rozenite, but other sulfate minerals also have to be considered. The implications of the possible presence of iron bearing polyhydrated sulfates such as rozenite and the absence of calcium - bearing gypsum for the geological history of the outcrops are not yet fully understood. Our next step is the geochemical modeling of the weathering of Martian basaltic rocks, dominated by iron and magnesium silicates, to iron bearing sulfates, under acidic conditions.

This paper examines several factors and attempts to understand the phenomenology and driving factors associated with exploitation of oblique hyperspectral imagery. The study takes advantage of established physical models such as the radiative transfer code MODTRAN, and simulation codes such as DIRSIG and FASSP (i.e., Forecasting and Analysis of Spectroradiometric System Performance). This paper studies the impact of a variety of parameters related to the oblique imaging problem. Topics include impact of path scatter and variation in upwelled radiance with azimuth view angle (i.e., amount of backscatter) in additional trade studies and material identification. Target detection is performed on DIRSIG 3D scenes where results are in the form of ROC curves. Lastly, required pixel fill factor is analyzed as it relates to sensor elevation angle, slant range, calibration error and object shade.

The Consultative Committee for Space Data System (CCSDS) is developing new international standards for satellite multispectral and hyperspectral data compression. The Canadian Space Agency's Successive Approximation Multi- Stage Vector Quantization (SAMVQ) has been selected as a candidate. The preliminary evaluation results show that the SAMVQ produces competitive rate-distortion performance on the CCSDS test images acquired by the hyperspectral sensors and hyperspectral sounders. There is a constraint to achieve lower bit rates on the multispectral images when the SAMVQ is applied to them due to the small number of bands. This is because the SAMVQ was designed for compression of hyperspectral imageries, which contain much more spectral bands than the multispectral images. This paper briefly reports the compression results of the SAMVQ on the CCSDS hyperspectral and hyperspectral sounders test images and studies on how to enhance the capability of the SAMVQ for compressing multispectral images while maintaining its unique properties for hyperspectral images.

State of the art and coming hyperspectral optical sensors generate large amounts of data and automatic analysis is necessary. One example is Automatic Target Recognition (ATR), frequently used in military applications and a coming technique for civilian surveillance applications. When sensors communicate in networks, the capacity of the communication channel defines the limit of data transferred without compression. Automated analysis may have different demands on data quality than a human observer, and thus standard compression methods may not be optimal. This paper presents results from testing how the performance of detection methods are affected by compressing input data with COTS coders. A standard video coder has been used to compress hyperspectral data. A video is a sequence of still images, a hybrid video coder use the correlation in time by doing block based motion compensated prediction between images. In principle only the differences are transmitted. This method of coding can be used on hyperspectral data if we consider one of the three dimensions as the time axis. Spectral anomaly detection is used as detection method on mine data. This method finds every pixel in the image that is abnormal, an anomaly compared to the surroundings. The purpose of anomaly detection is to identify objects (samples, pixels) that differ significantly from the background, without any a priori explicit knowledge about the signature of the sought-after targets. Thus the role of the anomaly detector is to identify "hot spots" on which subsequent analysis can be performed. We have used data from Imspec, a hyperspectral sensor. The hyperspectral image, or the spectral cube, consists of consecutive frames of spatial-spectral images. Each pixel contains a spectrum with 240 measure points. Hyperspectral sensor data was coded with hybrid coding using a variant of MPEG2. Only I- and P- frames was used. Every 10th frame was coded as I frame. 14 hyperspectral images was coded in 3 different directions using x, y, or z direction as time. 4 different quantization steps were used. Coding was done with and without initial quantization of data to 8 bbp. Results are presented from applying spectral anomaly detection on the coded data set.

A wide variety of automatic image registration methods have been proposed in the last years. However, under the scope of remote sensing applications, geometric correction is still mostly a manual work. A methodology for automatic image registration is proposed, which consists in three major steps: pre-processing, segmentation, and registration. The considered pre-processing is a new method, which is an iterative process based on a joint histogram analysis. Regarding the segmentation stage, global thresholding and a new method were used. The later comprises global thresholding and distance transforms in a single method. For both methods the following object properties were extracted: area, major and minor axis lengths of the adjusted ellipse and perimeter. The registration phase incorporates the matching of corresponding objects, a template matching technique to compute the distance between each pair of matched objects, and the computation of the transformation function parameters. The used dataset consisted in the pairs ETM+/ASTER, ETM+/SPOT and Orthophoto/IKONOS. The proposed methodology allows for the registration of a pair of images with translation and rotation effects, and to some extent with different spectral content, leading to a subpixel accuracy. Furthermore, it has been shown that the proposed pre-processing method allowed for the achievement of suitable segmented objects for later matching, even using global thresholding.

The paper presents some recent developments on object-based change detection and classification. In detail, the following algorithms were implemented either as Matlab or IDL programmes or as plug-ins for Definiens Developer: i) object-based change detection: segmentation of bitemporal datasets, change detection using the Multivariate Alteration Detection1 based on object features; ii) object features and object feature extraction: moment invariants, automated extraction of object features using Bayesian statistics; iii) object-based classification by neural networks: FFN and Class- dependent FFN using five different learning algorithms. The paper introduces the methodologies, describes the implementation and gives some examples results on the application.

Kernel versions of the principal components (PCA) and maximum autocorrelation factor (MAF) transformations are used to postprocess change images obtained with the iteratively re-weighted multivariate alteration detection (MAD) algorithm. It is found that substantial improvements in the ratio of signal (change) to background noise (no change) can be obtained especially with kernel MAF.

The most successful one-class classification methods are discriminative approaches aimed at separating the class of interest from the outliers in a proper feature space. For instance, the support vector domain description (SVDD) has been successfully introduced for solving one-class remote sensing classification problems when scarce and uncertain labeled data is available. The success of this kernel method is due to that maximum margin nonlinear separation boundaries are implicitly defined, thus avoiding the hard and ill-conditioned problem of estimating probability density functions (PDFs). Certainly, PDF estimation is not an easy task, particularly in the case of high-dimensional PDFs such as is the case of remote sensing data. In high-dimensional PDF estimation, linear models assumed by widely used transforms are often quite restrictive to describe the PDF. As a result, additional non-linear processing is typically needed to overcome the limitations of the models. In this work we focus on the multivariate Gaussianization method for PDF estimation. The method is based on the Projection Pursuit Density Estimation (PPDE) technique.1 The original PPDE procedure consists in iteratively project the data in the most non-Gaussian directions (like in ICA algorithms) and Gaussianizing them marginally. However, the extremely high computational cost associated to multiple ICA evaluations has prevented its practical use in high-dimensional problems such as those encountered in image processing. Here, we propose a fast alternative to iterative Gaussianization that makes it suitable for remote sensing applications while ensuring its theoretical convergence. Method's performance is successfully illustrated in the challenging problem of urban monitoring.

The new spaceborne very high resolution (VHR) synthetic aperture radar (SAR) sensors, such as TerraSAR-X and Cosmo-SkyMed permit to extract information from VHR SAR data in urban areas at the level of individual buildings. To support the widespread usage of VHR SAR for different application scenarios (e.g. damage assessment after natural disasters), and hence to increase the value of the data, automatic methods for building detection and reconstruction should be developed. In this paper we present a novel method for automatic building detection from single detected VHR SAR imagery. In a first step we extract basic features (e.g. lines) and identify their semantic meaning modeled as the degree of membership of the feature to a certain scattering class (e.g. double bounce). Then, we analyze the features extracted in a neighborhood system to identify building candidates using a production system. Finally, a score is calculated for each building candidate based on the semantic meaning of its features. This score is used to select or reject a candidate for the final set of extracted buildings. The preliminary results obtained on single detected VHR SAR images of an urban area show that the method effectively detects flat and gable roof buildings with only few missed and almost no false alarms.

It is often useful to fuse remotely sensed data taken from different sensors. However, before this multi-sensor data fusion can be performed the data must first be registered. In this paper we investigate the use of a new information-theoretic similarity measure known as Cross-Cumulative Residual Entropy (CCRE) for multi-sensor registration of remote sensing imagery. The results of our experiments show that the CCRE registration algorithm was able to automatically register images captured with SAR and optical sensors with 100% success rate for initial maximum registration errors of up to 30 pixels and required at most 80 iterations in the successful cases. These results demonstrate a significant improvement over a recent mutual-information based technique.

The early detection and identification of oil spills are critical prerequisites for performing cost-effective maritime salvage operations. This paper presents a new approach for distinguishing oil spills that are produced by stationary offshore sources, during their early phase of occurrence. The results were reached after analyzing over 100 images of satellite remote sensing data that were produced by either active microwave sensors like the Synthetic Aperture Radar (SAR) sensors or passive optical sensors like the Moderate Resolution Imaging Spectroradiometer (MODIS). The laws of conservation of mass and momentum that describe the dynamics of an oil spill over the water surface, were used for the development of a new detection algorithm that encompasses a parallel concept of shape conservation. The validity of this new empirical algorithm depends upon a number of assumptions that were made about the oil viscosity, temperature, water currents, wind speeds and the spills' spatial extent and duration. It can also be shown that unique texture differences can be revealed between an oil spill and other look-alikes' features like, for example, wind patterns, rain cells and algal mats by applying edge filtering operations on the patches that are under investigation, and therefore the reduction of false positives. The work presented here may have profound implications on future studies that examine the use of automatic recognitions methods, that are based on pattern and texture analysis. The results may also lead to new methodologies by which the dispersion and trajectory models of oil spills can be studied in new detail and ultimately used in environmental impact assessment operations.

This paper presents a novel segmentation algorithm based on optimizing histogram multi-level thresholding of images by employing a variation of particle swarm optimization (PSO) Algorithm which improves the accuracy and the speed of segmentation based on the conventional PSO algorithm. Entropy has been chosen as the criteria for segmentation based on the multi-level thresholding. Entropy is input parameter of a fitness function for finding the best segmentation level. We have to find the optimum thresholding level based on the entropy of different image segments. A new optimization algorithm that called Hybrid cooperative- comprehensive learning PSO (HCOCLPSO), is used for optimization in this paper. This algorithm overcomes on common problems of basic variants of PSO, which are curse of dimensionality and tendency of premature convergence or in other word, getting stuck in local optima. This segmentation technique has been compared with conventional segmentation based on PSO and genetic algorithm (GA). We presented our segmentation results to experts. Our subjective measurements by experts show that we can achieve about 80 percents accuracy which is a better result when compared with conventional PSO and genetic algorithm. In terms of seed we can achieve much higher performance than two other schemes.

We have developed a house detection method based on machine learning for classification of houses and non-houses. In order to achieve precise classification, it is important to select features and to determine a dimensionality reduction method and a learning method. We first applied Gabor wavelet filters to generate the feature vectors and then developed a new method using the Adaboost algorithm to reduce the dimensionality of feature space. If a linear classifier made by one element of a feature vector is considered as a weak classifier in Adaboost, higher contribution dimensions can be selected. We used support vector machines (SVM) for the learning method. We evaluated our method by using QuickBird panchromatic images. Despite the significant variations in house shape and rooftop color, and in background clutter, our algorithm achieved high accuracy in house detection.

This paper presented a method for precisely computing the ground 2D size of any pixel in a satellite image received by a push-broom linear array sensor, and further calculating the planar distance between any two pixels in the image. The algorithm is deduced from the imaging principle of linear array sensors, with consideration of the arc of the earth's surface, the height of ground, and the light refraction caused by the air. The author used the method to measure the ground planar distance between two pixels in a Worldview-1 image, compared with the site measurement, the errors were less than the size of pixel. As the computation is based on the instant position and orientation of the sensor, the method is useful for local small area measuring and real time measuring.

The procedure of radiometric normalization is necessary as change detection (CD) is highly dependent on accurate geometric and radiometric correction. The quality of radiometric normalization highly affects the results of change detection. During the last years, many methods on radiometric normalization have been proposed. However, they are generally proposed not for CD. In this paper, we will discuss these methods for CD. The ultimate objective of CD is identifying changes in the state of an object, or other earth-surface features, between different data. If you can't properly deal with radiometric normalization in CD, or the relationship between radiometric normalization and CD, you will fall into trouble. With respect to the accuracy, efficiency and operation of radiometric normalization for CD in VHR(very high resolution images such as spot5,QuickBird,Iknos), we design the processing procedure of radiometric normalization for CD in multitemporal images. Moreover, an improved matching algorithm based on Harris corner detection is described in this paper, which makes full use of graphic point feature, gray-level pixel and location information. In experiment, the proposed procedure has been proved effective and can be recommended for use in CD projects.

The Compact Airborne Imaging Spectrometer System (CAISS) was designed and developed as the airborne hyperspectral imaging system. The mission of the CAISS is to provide full contiguous spectral information with high spatial resolution for advanced applications in the field of remote sensing. The CAISS has an ability to control the spectral and spatial configuration of the imaging instruments. In order to understand the mechanism of imaging spectrometer system and its characteristics, the several verification tests with the CAISS were conducted in the laboratory. Especially, the verification of camera system was performed with the integrating sphere and spectral lamps. In order to verify the spectral characteristics, four spectral binning (x1, x2, x4, and x8) were measured using each of the spectral lamps and the position of the peaks was compared to the reference data sheet of each spectral lamps. For all measurements, it was found that the spectral deviation was lower than the Full Width Half Maximum (FWHM) of the system for each of the spectral binning. Also several interface verification tests between the CAISS and the airplane were conducted on the ground. This paper presents the preliminary results of verification test in the camera system level and interface test with airplane on the ground.

A new change detection method for remotely sensed images is proposed. This method can be applied to two images which have different number of spectral bands and/or have different spectral ranges. The proposed method converts two multi-spectral-multi-temporal images into two sets of canonical variate images which have limited correlation called the canonical correlation. Then, one or more canonical variate images which are the most suitable for change detection are selected and change detection regions in the original images are extracted by using statistical modeling and statistical test. In this paper, the detail of the proposed method is described. Some experiments using simulated multi-spectral-multi-temporal images based on spectral profiles in ASTER Spectral Library are conducted to confirm change detection accuracy. The experimental results show reasonable changed regions and their change quantities.

In this paper, we present a model to define an analogy between text and image data using ICA basis functions and we apply it to index Very High Resolution (VHR) satellite images. We introduce our text-image analogy by defining visual documents, visual words, visual vocabulary and labeling each word of document using vocabulary words. Further, we propose a classification using a simple Bayesian method to evaluate our model for VHR satellite image characterization. The results for two types of vocabularies, one based on visual words clustering and other based on obtaining ICA components for each class, is compared for a variety of natural and man-made scenes.

A vacuum infrared standard radiation thermometer (VIRST) was jointly developed by the Physikalisch-Technische Bundesanstalt (PTB) and the Raytek GmbH for temperature measurements from -150 °C to 170 °C under vacuum. The radiation thermometer is a purpose built instrument to be operated with the PTB reduced-background infrared calibration facility (RBCF). The instrument is a stand-alone system with an air-tight housing which allows operation inside a vacuum chamber, attached to a vacuum chamber and on air. The radiation thermometer serves to calibrate thermal radiation sources, i.e. blackbody radiators, for their radiation temperature by comparison with the vacuum variable-temperature reference blackbodies inside the RBCF. Furthermore, since it can be operated under vacuum and on air, the instrument also allows to compare the water- and ammonia-heat-pipe reference blackbodies of the PTB low-temperature calibration facility on air with the vacuum variable-temperature blackbodies of the RBCF. Finally the instrument shall be used as a transfer radiation thermometer to carry and compare the temperature scale of PTB by means of radiation thermometry to remote sensing calibration facilities outside PTB. The mechanical, optical and electrical design of the instrument are shortly reported. Results on the temperature resolution, size-of-source effect, long-term stability, and the reference function are given. A comparison of the heat-pipe blackbodies on air with the vacuum variable-low-temperature blackbody (VLTBB) and a comparison of the VLTBB with the vacuum variable-medium- temperature blackbody (VMTBB) via VIRST are presented. The long term stability of the infrared precision radiation thermometer after 17 months is shown.

Soil moisture estimation is one of the most challenging problems in the context of biophysical parameter estimation from remotely sensed data. Different approaches have been proposed in the literature, but in the last years there is a growing interest in the use of non-linear machine learning estimation techniques. This paper presents an experimental analysis in which two non-linear machine learning techniques, the well known and commonly adopted MultiLayer Perceptron neural network and the more recent Support Vector Regression, are applied to solve the problem of soil moisture retrieval from active and passive microwave data. Thank to the use of both simulated and real in situ data, it was possible to investigate the effectiveness of both techniques in different operative scenarios, including the situation of limited availability of training samples which is typical in real estimation problems. Moreover, for each scenario, different configurations of the input channels (polarization, acquisition frequency and angle) have been considered. The comparison between the two methods has been carried out in terms of different figure of merits, including error measurements and correlation coefficients between estimated and true values of the desired biophysical parameter. The results achieved indicate the Support Vector Regression as an effective alternative to the neural network approach, due to a general better estimation accuracy and a higher robustness to outliers, especially in case of limited availability of samples.

Recent developments, based on lattice auto-associative memories, have been proposed as novel and alternative techniques for endmember determination in hyperspectral imagery. The present paper discusses and compares three such methods using, as a case study, the generation of vegetation abundance maps by constrained linear unmixing. The first method uses the canonical min and max autoassociative memories as detectors for lattice independence between pixel spectra; the second technique scans the image by blocks and selects candidate spectra that satisfies the strong lattice independence criteria within each block. Both methods give endmembers which correspond to pixel spectra, are computationally intensive, and the number of final endmembers are parameter dependent. The third method, based on the columns of the matrices that define the scaled min and max autoassociative memories, gives an approximation to endmembers that do not always correspond to pixel spectra; however, these endmembers form a high-dimensional simplex that encloses all pixel spectra. It requires less computations and always gives a fixed number of endmembers, from which final endmembers can be selected. Besides a quantification of computational performance, each method is applied to discriminate vegetation in the Jasper Ridge Biological Preserve geographical area.

Simulated SAR images can be used for a wide variety of purposes such as classification, object recognition or the education of SAR image analysts. In this paper, a SAR simulation tool based on an extended raytracing approach is introduced. The approach allows for a coherent simulation of the most important SAR image features. Another aspect is the modeling of material properties necessary for the creation of synthetic SAR images that look realistic. Also, the image formation process is described in detail since it has immense impact on the overall appearance of the final image. The different steps of the simulation process and their influence on the appearance of the final simulated image are demonstrated with simulated images of a simple building. Then, examples for the simulation of large scale scenes at high resolution are given.

A Compact Real-Time Optical SAR Processor has been successfully developed and tested. SAR, or Synthetic Aperture Radar, is a powerful tool providing enhanced day and night imaging capabilities. SAR systems typically generate large amounts of information generally in the form of complex data that are difficult to compress. Specifically, for planetary missions and unmanned aerial vehicle (UAV) systems with limited communication data rates this is a clear disadvantage. SAR images are typically processed electronically applying dedicated Fourier transformations. This, however, can also be performed optically in real-time. Indeed, the first SAR images have been optically processed. The optical processor architecture provides inherent parallel computing capabilities that can be used advantageously for the SAR data processing. Onboard SAR image generation would provide local access to processed information paving the way for real-time decision-making. This could eventually benefit navigation strategy and instrument orientation decisions. Moreover, for interplanetary missions, onboard analysis of images could provide important feature identification clues and could help select the appropriate images to be transmitted to Earth, consequently helping bandwidth management. This could ultimately reduce the data throughput requirements and related transmission bandwidth. This paper reviews the design of a compact optical SAR processor prototype that would reduce power, weight, and size requirements and reviews the analysis of SAR image generation using the table-top optical processor. Various SAR processor parameters such as processing capabilities, image quality (point target analysis), weight and size are reviewed. Results of image generation from simulated point targets as well as real satellite-acquired raw data are presented.

The TerraSAR-X (copyright) mission, launched in 2007, carries a new X-band Synthetic Aperture Radar (SAR) sensor optimally suited for SAR interferometry (InSAR), thus allowing very promising application of InSAR techniques for the risk assessment on areas with hydrogeological instability and especially for multi-temporal analysis, such as Persistent Scatterer Interferometry (PSI) techniques, originally developed at Politecnico di Milano. The SPINUA (Stable Point INterferometry over Unurbanised Areas) technique is a PSI processing methodology which has originally been developed with the aim of detection and monitoring of coherent PS targets in non or scarcely-urbanized areas. The main goal of the present work is to describe successful applications of the SPINUA PSI technique in processing X-band data. Venice has been selected as test site since it is in favorable settings for PSI investigations (urban area containing many potential coherent targets such as buildings) and in view of the availability of a long temporal series of TerraSAR-X stripmap acquisitions (27 scenes in all). The Venice Lagoon is affected by land sinking phenomena, whose origins are both natural and man-induced. The subsidence of Venice has been intensively studied for decades by determining land displacements through traditional monitoring techniques (leveling and GPS) and, recently, by processing stacks of ERS/ENVISAT SAR data. The present work is focused on an independent assessment of application of PSI techniques to TerraSAR-X stripmap data for monitoring the stability of the Venice area. Thanks to its orbital repeat cycle of only 11 days, less than a third of ERS/ENVISAT C-band missions, the maximum displacement rate that can be unambiguously detected along the Line-of-Sight (LOS) with TerraSAR-X SAR data through PSI techniques is expected to be about twice the corresponding value of ESA C-band missions, being directly proportional to the sensor wavelength and inversely proportional to the revisit time. When monitoring displacement phenomena which are known to be within the C-band rate limits, the increased repeat cycle of TerraSAR-X offers the opportunity to decimate the stack of TerraSAR-X data, e.g. by doubling the temporal baseline between subsequent acquisitions. This strategy can be adopted for reducing both economic and computational processing costs. In the present work, the displacement rate maps obtained through SPINUA with and without decimation of the number of Single Look Complex (SLC) acquisitions are compared. In particular, it is shown that with high spatial resolution SAR data, reliable displacement maps could be estimated through PSI techniques with a number of SLCs much lower than in C-band.

Many works and European projects have proven the ability of Permanent Scatterers Synthetic Aperture Radar Interferometry (PS-InSAR) to measure the slow deformation of the persistent ground objects with a millimetric precision measurement. Compared to the classical differential SAR Interferometry (DInSAR), PS-InSAR is an approach that estimates several contributions: atmospheric disturbances, orbital errors, deformation signal as well as topographical errors. In this paper, we propose to apply PS-InSAR for the analysis of the ground deformation phenomena in a urban context. For that purpose, the Stanford Method for Permanent Scatterers (StaMPS) is applied using an ERS data archive. StaMPS was developed in Stanford University by Andy Hooper. The advantages to use StaMPS were that it is free and many scripts are already available to process the dataset. In first steps of the processing, differential interferograms are produced using the Delft Object-oriented Radar Interferometric Software (DORIS), developed in Delft University of Technology . Doris is also a free tool. StaMPS method is briefly explained. A first experiment on the city of Paris, France, is presented, especially because PSINSAR and DInSAR results have already been published by several researchers. Therefore, the processing of Nantes (French city) is carried out. Some important results are shown.

Persistent Scatterers Interferometric SAR (PSInSAR) is a powerful method for measuring surface deformations with millimeter-accuracy. Currently available algorithms are designed to detect large-scale earth movement or apply a predefined movement model. The analysis of the displacement of corner reflectors on our test area required the capability to detect localized deformation. To enable that, we modified StaMPS (Stanford Method for PS). We oppose the modified algorithm to the original algorithm, using time series of TerraSAR-X data from our test area. The finding is a distinct enhancement in the detection of localized deformation.

In 2007 and 2008 ASI launched three out of four X-band SAR satellites of the COSMO-SkyMed Mission, making available to the users a unique SAR constellation dedicated to the Earth Observation. The constellation will be completed with the launch of the fourth satellite in 2010. The complete deployed system consists of a constellation of four Low Earth Orbit mid-sized satellites, each equipped with a multi-mode high-resolution Synthetic Aperture Radar (SAR) operating at X-band. The results coming from the utilisation of the first three satellites reveal an outstanding performance of the X-band SAR and the importance of fast response times in several applications such as risk and emergency management and ice monitoring. During the tragic event of the L'Aquila earthquake CSK monitored the areas affected by the quake detecting a surface displacement up to 19 cm. Also analyses were conducted to identify models of the fault causing the earthquake. Wilkins Ice Shelf relevant disintegration started in 2007-2008. On April 2009 another consistent collapse occurred, causing the brake of the ice bridge located between the Charcot Island and the Antarctic Peninsula. The phenomenon is going ahead and main cracks appeared also in the shelf placed between Latady Island and the Antarctic Peninsula.

The new constellation of COSMO -SkyMed (CSK) satellites, which carry on board a SAR-X instrument, offers a unique opportunity for the study of sea ice distribution in polar regions and for the detection of drifting icebergs. To exploit this new class of SAR images, a specific processing scheme, consisting of 4 sequential blocks, was developed. The 4 blocks perform: i) ingestion of the raw image into a commercial software for precise geo-location; ii) extraction from the full scene of image subsets containing iceberg infested areas; iii) speckle filtering; iv) segmentation by different algorithms taken from a "free and open source software" (FOSS) library. The results of the application of this processing scheme to a CSK image of Antarctica containing icebergs are presented and discussed.

In many regions of the globe an update knowledge of the snow cover extent and wetness is of primary importance for both scientific and application aspects. Indeed, snow plays a major role in the global water cycle and the climate system, as well as in natural disasters such as floods, landslides. An operational algorithm to produce snow cover maps from remote sensing data in the Italian Alps has been implemented in the framework of the Italian national project PROSA to contribute timely information to civil protection from floods and landslides. The algorithm can generate maps in presence of cloud cover by combining optical data from MODIS and SAR data from ENVISAT. It has been validated on a wide area in North Italy by comparing the algorithm output with ground measurements.

The COSMO-SkyMed mission offers a unique opportunity to obtain radar images useful for flood mapping, being characterized by high revisit time, thanks to the four satellites that form its constellation. To study the potentiality of Cosmo-SkyMed radar data for this purpose, two inundation events are analyzed in this paper, namely the flood occurred in Myanmar in May 2008 and the event that took place in the city of Alessandria (Northern Italy) in April 2009. For the first event, two radar images were considered, one temporally close to the peak of the event, and the other one that was acquired one week later. As for the Alessandria overflow, a time series of images was available. While most of the literature algorithms are based on fixed thresholds applied on an image temporarily close to the event, our method accounts both for specular reflection, typical bare flooded soils, and for double bounce backscattering often occurring on forested and urban inundated areas. Such a model-based approach is expected to improve the accuracy of flood mapping.

Monitoring and mapping of active or dormant landslides as well as vulnerable slopes can greatly contribute to both the mitigation of landslide hazards and reduction of their impact. Classical landslide surveying techniques have recently been complemented by satellite data analysis, namely by SAR interferometry. In particular the Permanent Scatterer Interferometry (PSI) technique can be considered complementary to conventional geological and geomorphological studies due to its capacity to detect small displacements over long time periods and large areas. The integration of both the 'classical' and the SAR based methods for monitoring surface displacements at regional scale have been developed and tested within the framework of different commercial and public founded projects, including SLAM, TERRAFIRMA and PREVIEW. Based on these achievements, landslide inventories of past events carried out over large areas and landslide monitoring of specific built-up areas have been selected to become pan-European GMES services to be implemented within the Emergency Response SAFER (Services and Applications for Emergency Response). One of the SAFER's test regions is the Autonomous Province of Bolzano (Nothern Italy). Landslide Inventory Mapping (LIM) based on SAR interferometry, landslide archives and other ancillary data will be carried out in cooperation with the provincial geological service as main end-user. The results will support hazard zone mapping, the analysis of mass movements in construction zones of large objects (tunnels, bridges, dams etc.) and the choice of trigonometric points.

In the present work we present first results of ground deformation measurements inferred through repeat-pass Synthetic Aperture Radar (SAR) Interferometry (InSAR) in C- and X-band over an Italian Alpine area, in Lombardia region. The activity was carried out in the framework of the MORFEO (MOnitoraggio e Rischio da Frana mediante dati EO) project founded by the Italian Spatial Agency (ASI) and dedicated to landslide risk assessment. A number of areas affected by hydrogeological instabilities have been selected and studied in detail by processing both C- and X-band SAR data through SPINUA, a Persistent Scatterer like algorithm. In particular, two stacks of 30 Ascending ENVISAT SAR images (October 2004 - January 2009) and 32 Descending ENVISAT SAR images (December 2004 - January 2009) have been independently processed to ensures the detection of movements occurring along both west and east facing slopes. Moreover, further deformation measurements have been obtained by processing a set of 12 COSMO-SkyMED ascending HIMAGE interferometric acquisitions (Satellite CSKS1; Beam HI-03; POL: HH; Incidence Angle: 29°) provided by ASI. After a proper tuning of the interferometric algorithmic solutions, even with very high normal baselines, we are able to appreciate the potentials of X-band interferometry. Although the number of COSMO-SkyMED acquisitions is quite limited, spanning a period of only 10 months (August 2008 - June 2009), SPINUA was capable to retrieve preliminary ground displacement patterns that are in good agreement with those previously estimated in C-band. Quite impressive is to realize that, because of the tenfold improved resolution of X-band images, multi-temporal InSAR techniques may be successfully applied for the estimation of the displacement maps with a number of acquisitions much lower than in C-band. InSAR-derived displacements provided on areas of hydrogeological interest are going to be validated in the framework of MORFEO project by the geological partnership thanks to the availability of ground truths. In the present work, we present the results obtained on the towns of Garzeno, Catasco and Germasino which are affected by landslide phenomena. We provide a first validation by comparing the deformation maps derived from ENVISAT C-band data, from COSMO-SkyMED X-band data as well as from ERS and RADARSAT C-band data freely available on the GeoIFFI web-catalogue.

The problem of soil moisture retrieval by means of a bistatic radar system, for the purpose of investigating a candidate bistatic mission, is addressed in this paper. The optimal geometric configuration of measurement in terms of incidence and observation (scattering) directions is identified. Such a configuration should ensure good sensitivity to soil moisture and good spatial coverage. We use a theoretical scattering model to simulate the bistatic scattering coefficient of bare soil for analyzing the sensitivity to soil moisture. The error variance of a linear regression estimator is employed for this purpose. This index is computed as function of elevation and azimuth scattering angles. Results shows that the best performances in terms of soil moisture retrieval can be achieved by complementing the bistatic measurements with the monostatic ones, which are supposed to be available through already operating spaceborne radars. The problem of the feasibility of the bistatic configuration identified through the sensitivity study is also addressed focusing on the duty cycle and the spatial coverage.

The main aim of the analysis presented in this paper is to cross-compare two retrieval methodologies, one based on Neural Network and the other on Bayesian approach in different types of test areas and verify if they are able to retrieve the same spatial and temporal soil moisture features. The test areas are located in three regions in Italy in order to take into account different soil and meteorological conditions. The comparison of the backscattering coefficients as a function of soil moisture values indicate the same sensitivity to soil moisture variations but with a different bias which may depend on soil characteristics, vegetation presence and roughness effect. The results of the two retrieval methodologies indicate an overall good agreement. Only in one single date, the discrepancy between the results is around 8%. The algorithms are also compared in terms of processing times.

In this work, scattering models and a Bayesian inversion algorithm are applied to Cassini SAR and radiometric data in order to characterize lake and land surfaces. Radar backscattering from lakes is described in terms of a double layer model, Bragg or facets scattering for the upper liquid layer and I.E.M model for the lower solid surface. This electromagnetic analysis is the starting point for the statistical inversion algorithm, to determine limits on the parameters values. Radiometer data are described with a forward radiative transfer model thus accounting for the presence of multiple layer emission and volume effects. A combined sensitivity study is performed on backscattering and brightness temperature models to define the best approach for the synergic use of active and passive data in the Bayesian algorithm. The use of e.m. scattering models allows evaluating the compatibility of the observed RCS with the expected scenarios in terms of dielectric constant of the surface constituents. A good correlation is found between the radar and the radiometric data. Brightness temperature modeling combined with SAR Bayesian inversion can improve parameter retrieval.

This paper aims at identifying an operational methodology to derive soil moisture status from optical images by using soil moisture values derived from SAR images as a calibration tool . In the first part of the paper, an algorithm based on Bayesian techniques for the retrieval of soil moisture from C-band SAR images is presented. The algorithm is composed of two modules, one for bare soil and the other for vegetated soil which includes also the use of optical images in order to take into account the vegetation contribution. soil moisture values retrieved from images are then used as a calibration tool for a soil moisture index derived from MODIS images. In this case, the method to estimate soil moisture index from optical and thermal images is based on the calculation of the Apparent Thermal Inertia (ATI). ATI is considered as an approximate (apparent) value of the thermal inertia and is obtained from spectral measurements of the albedo and the diurnal temperature range. soil moisture estimated from SAR images and the ATI are compared in order to find a calibration curve which should cover the entire soil moisture values from saturation to residual moisture values. For the calibration experiment, three main sites were chosen which exhibit different landscape and climatic characteristics. The Basento basin is located in Southern Italy and is characterized by long period of droughts. The Scrivia valley is flat alluvial plain measuring situated close to the confluence of the Scrivia and Po rivers in Northern Italy. The Cordevole watershed, located at the foothill of Mount Sella in Northern Italy is mainly covered by grassland and it was selected because of its relatively smooth topography. The first results indicate a good correlation between ATI and the soil moisture values derived both from measurements and estimated from SAR images.

Most of the current SAR systems aquire fully polarimetric data where the obtained scattering information can be represented by various coherent and incoherent parameters. In previous contributions we reviewed these parameters in terms of their "utility" for landcover classification, here, we investigate their impact on several classification algoritms. Three classifiers: the minimum-distance classifier, a multi-layer perceptron (MLP) and one based on logistic regression (LR) were applied on an L-Band scene acquired by the E-SAR sensor. MLP and LR were chosen because they are robust w.r.t. the data statistics. An interesting result is that MLP gives better results on the coherent parameters while LR gives better results on the incoherent parameters.

For many years, high resolution SAR (Synthetic Aperture Radar) imaging was limited to airborne instruments. Nowadays, the analysis of spaceborne high resolution SAR images with up to 1 meter spatial resolution has become possible with the advent of German, Italian, and Canadian missions and their subsequent data distribution. For instance, compared to previous missions with much lower resolution, the German TerraSAR-X data allow us to analyze SAR images containing an increased amount of details and information content. As a consequence, a robust detection and recognition of small scale man-made structures representing buildings, roads, harbors, bridges, etc has become a new challenging task. An important property of SAR data is the presence of speckle phenomena which, in most cases, precludes an automated interpretation of SAR images. Therefore, we use a Bayesian approach relying on models and their parameters to fit the data. We suggest an automated method being able to extract and interpret the genuine information contained in high resolution SAR images. Our solutions are provided for optimal processing both for visual and automated data interpretation. The image information content is extracted using model-based methods based on Gibbs Random Fields combined with a Bayesian inference approach. The approach enhances the local adaptation by using a prior model, which learns the image structure; it enables despeckling with minimum loss of resolution and simultaneously estimates the local description of the structures. Form these we may obtain detection, classification, and recognition of the image content. In the following, we present typical texture description and classification examples of 1 meter resolution TerraSAR-X images taken in spotlight mode. In particular, we describe how well speckle can be removed, how well local texture parameters of the data can be estimated using dedicated model-based methods, and what can be expected from automated classification. For our work, we use the Knowledge-based Information Mining system called KIM, which includes a graphical user interface for data handling, image inspection, and semantic image annotation.

For several years, image classification and pattern recognition algorithms have been developed for the land coverage mapping using radar and multispectral imagery with medium to large pixel size. As several satellites now distribute submetric-pixel and metric-pixel images (for example QUICKBIRD,TERRASAR-X), the research turns to the study of the structure of cities: building structuring, grassy areas, road networks, etc, and the physical description of the urban surfaces. In that context, we propose to underline new potentialities of submetric-pixel polarimetric SAR images. We deal with the characterization of roofs and the mapping of trees. For that purpose, a first analysis based on photo-interpretation and the assessement of several polarimetric descriptors is carried out. Then, an image classification scheme is built using the polarimetric H/alpha-Wishart algorithm, followed by a decision tree. This one is based on the most pertinent polarimetric descriptors and aims at reducing the classification errors. The result proves the potential of such data. Our work relies on an image of a suburban area, acquired by the airborne RAMSES SAR sensor of ONERA.

Performance analysis of Bistatic Synthetic Aperture Radar (SAR) characterized by arbitrary geometric configurations is usually complex and time-consuming since system impulse response has to be evaluated by bistatic SAR processing. This approach does not allow derivation of general equations regulating the behaviour of image resolutions with varying the observation geometry. It is well known that for an arbitrary configuration of bistatic SAR there are not perpendicular range and azimuth directions, but the capability to produce an image is not prevented as it depends only on the possibility to generate image pixels from time delay and Doppler measurements. However, even if separately range and Doppler resolutions are good, bistatic SAR geometries can exist in which imaging capabilities are very poor when range and Doppler directions become locally parallel. The present paper aims to derive analytical tools for calculating the geometric resolutions of arbitrary configuration of bistatic SAR. The method has been applied to a hybrid bistatic Synthetic Aperture Radar formed by a spaceborne illuminator and a receiving-only airborne forward-looking Synthetic Aperture Radar (F-SAR). It can take advantage of the spaceborne illuminator to dodge the limitations of monostatic FSAR. Basic modeling and best illumination conditions have been detailed in the paper.

The development of the polarimetric synthetic aperture radar (PolSAR) applications has been accelerated by coming of new generation of SAR polarimetric satellites (TerraSAR-X, COSMO-SkyMed, RADARSAT-2, ALOS, etc.). The aim of this article is to extract the information content of the polarimetric SAR data. Cross products of four channels "HH, HV, VH, and VV" could be at least nine features in vector space and by applying the different class separability criterion, the impacts of each feature, for extracting different patterns, could be tested. We have chosen the large distance between classes and small distance within-class variances as our criterion to rank the features. Due to high mutual correlation between some of the features, it is preferable to combine the features which result in the lower number of features. Also the computational complexity will be decreased when we have lower number of features. Due to these advantages, our goal would be to decrease the number of features in vector space. To achieve that, a subset of ranked features consists of two to nine ranked features will be classified and the classification accuracy of different subsets will be evaluated. It is possible that some of the new features that have been added to the old subsets change the classification accuracy. Finally different feature subsets which were selected based on the various class-separability approaches will be compared. The subset that gives the highest overall accuracy would be the best representative of the nine originally features.

Envisat ASAR images are used to observe atmospheric gravity waves over the East China Sea. Several case studies are presented in detail. It is shown that the atmospheric gravity waves are well organized in the form of wave packets. The wavelengths are ranging from 0.5 to 10 km. The atmospheric gravity waves locate near the stationary meteorological fronts. The generation mechanism is discussed.

As ENVISAT ASAR WS mode imagery is suitable for ship monitoring, a new ship detection model is presented in this paper. In the model, a new grid method and an improved CFAR algorithm were introduced. The grid method divides the SAR image into regular frames and chooses the highest average intensity as the stand average value in the round frames. Then we build the model using the deviation of the center frame. In order to examine the performance of the model, some ENVISAT ASAR WS images were used and the result shows that the performance is excellent in despite of the simple model.