This PDF file contains the front matter associated with SPIE Proceedings Volume 8892 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

In this paper, pansharpening methods suitable for the spatial enhancement of hyperspectral images are critically discussed and assessed on simulated data from the upcoming PRISMA mission, featuring a panchromatic camera mounted on the hyperspectral payload. Since most fusion techniques are limited to the fusion of multispectral (MS) images with panchromatic (Pan) images, the focus shall be on the extension of such methods towards hyperspectral (HS) images. In particular, the impact of the bandwidth of Pan on fusion performances will be analyzed and discussed.

While developing high resolution payloads, it is also necessary to make full use of the present spaceborne/airborne payload resources by super resolution (SR). SR is a technique of restoring a high spatial resolution image from a series of low resolution images of the same scene captured at different times in a short period. Common SR methods, however, may fail to overcome the irregular local warps and transformation in low resolution remote sensing images caused by platform vibration and air turbulence. It is also difficult to choose a generalized prior for remote sensing images for Maximum a Posteriori based SR methods. In this paper, irregular local warps and transformation within low resolution remote sensing images will be corrected by incorporating an elastic registration method. Moreover, combined sparse representation will be proposed for remote sensing SR problem. Experimental results show that the new method constructs a much better high resolution image than other common methods. This method is promising for real applications of restoring high resolution images from current low resolution on-orbit payloads.

This paper presents a new fusion approach for pan-sharpening multispectral remote sensing images. This approach, related to Linear Spectral Unmixing (LSU) techniques, includes Extended Nonnegative Matrix Factorization (ExNMF) for combining low spatial resolution multispectral and high spatial resolution panchromatic data. ExNMF is applied to different real multispectral and panchromatic data sets with different spatial resolutions and different number of spectral bands. The quality of pan-sharpened multispectral images is evaluated by the jointly spectral and spatial Quality with No Reference (QNR) index. Obtained results show that our proposed method outperforms the Principal Component Analysis (PCA) and Gram-Schmidt (GS)-based standard literature methods.

The METEOSAT Third Generation (MTG) Programme will provide the geostationary platforms for operational meteorological data acquisitions over Europe in 2018-2030. The Flexible Combined Imager (FCI) instrument is one of the MTG imager instruments and has a heritage from SEVIRI flown on the current METEOSAT Second Generation (MSG) satellites. It is a radiometer providing measurements in 16 spectral bands with a full Earth coverage every 10 minutes. For the Level 2 processing of FCI datasets the measurements have to be re-sampled on a constant reference grid in a geostationary projection – this process is referred to as rectification. The use of a three-axis stabilised platform and the scanning scheme applied to the FCI make rectification in MTG more challenging than in the MSG/SEVIRI case. Classical interpolation formulas assume a uniform sampling spacing of the measurements. However, non-uniform sampling may occur in the FCI sampling acquisition due to platform dynamics, micro-vibrations, thermo-elastic focal plane and optical distortions. In such a case, classical methods can cause significant rectification errors and interpolation algorithms, which can cope with non-uniform sampling, are required. This paper analyses the effect of non-uniform sampling in the FCI rectification process and aims to select and assess suitable resampling algorithms for the FCI L1 processing chain. Several techniques tailored to non-uniform resampling have been implemented. Performances of both uniform and non-uniform interpolation algorithms have been evaluated and compared using simulated FCI-like data samples. The analysis has been done for a nominal and a worst-case sample acquisition scenario. The presentation will show the results of our simulations with respect to the MTG requirements.

In this paper a new hyperspectral image based on wavelets and sparse regularization is proposed. This new method is called Wavelet Based Sparse Restoration (WBSR). The hyperspectral signal is restored by utilizing penalized least squares and the `1 penalty. Iterative Soft Thresholding (IST) algorithm is used to solve the convex optimization problem. It is shown that not only WBSR improves the denioising results both visually and based on Signal to Noise Ratio (SNR) but also increases the classification accuracies.

The present work has to be seen in the context of real-time on-board image evaluation of optical satellite data. With on board image evaluation more useful data can be acquired, the time to get requested information can be decreased and new real-time applications are possible. Because of the relative high processing power in comparison to the low power consumption, Field Programmable Gate Array (FPGA) technology has been chosen as an adequate hardware platform for image processing tasks. One fundamental part for image evaluation is image segmentation. It is a basic tool to extract spatial image information which is very important for many applications such as object detection. Therefore a special segmentation algorithm using the advantages of FPGA technology has been developed. The aim of this work is the evaluation of this algorithm. Segmentation evaluation is a difficult task. The most common way for evaluating the performance of a segmentation method is still subjective evaluation, in which human experts determine the quality of a segmentation. This way is not in compliance with our needs. The evaluation process has to provide a reasonable quality assessment, should be objective, easy to interpret and simple to execute. To reach these requirements a so called Segmentation Accuracy Equality norm (SA EQ) was created, which compares the difference of two segmentation results. It can be shown that this norm is capable as a first quality measurement. Due to its objectivity and simplicity the algorithm has been tested on a specially chosen synthetic test model. In this work the most important results of the quality assessment will be presented.

For urban terrain data with many outliers, we compare an ℓ^1-regression-based and a RANSAC-based planar segmentation procedure. The procedure consists of 1) calculating the normal at each of the points using ℓ¹ regression or RANSAC, 2) clustering the normals thus generated using DBSCAN or fuzzy c-means, 3) within each cluster, identifying segments (roofs, walls, ground) by DBSCAN-based-subclustering of the 3D points that correspond to each cluster of normals and 4) fitting the subclusters by the same method as that used in Step 1 (ℓ¹ regression or RANSAC). Domain decomposition is used to handle data sets that are too large for processing as a whole. Computational results for a point cloud of a building complex in Bonnland, Germany obtained from a depth map of seven UAV-images are presented. The ℓ^1-regression-based procedure is slightly over 25% faster than the RANSAC-based procedure and produces better dominant roof segments. However, the roof polygonalizations and cutlines based on these dominant segments are roughly equal in accuracy for the two procedures. For a set of artificial data, ℓ¹ regression is much more accurate and much faster than RANSAC. We outline the complete building reconstruction procedure into which the ℓ¹-regression-based and RANSAC-based segmentation procedures will be integrated in the future.

In recent years, an automatic urban road extraction, as part of Intelligent Transportation research, has attracted the researchers due to the important role for the next modern transportation where urban area plays the main role within the transportation system. In this work, we propose a new combination of fuzzy ART clustering, Region growing, Morphological Operations and Radon transform (ARMOR) for automatic extraction of urban road networks from the digital surface model (DSM). The DSM data, which is based-on the elevation of surface, overcome a serious building's shadow problem as in the aerial photo image. Due to the different elevation between the road and the buildings, the thresholding technique yields a fast initial road extraction. The threshold values are obtained from Fuzzy ART clustering of the geometrical points in the histogram. The initial road is then expanded using region growing. Though most of the road regions are extracted, it contains a lot of non-road areas and the edge is still rough. A fast way to smoothing the region is by employing the morphology closing operation. Furthermore, we perform the road line filter by opening operation with a line shape structuring element, where the line orientation is obtained from the Radon Transform. Finally, the road network is constructed based-on B-Spline from the extracted road skeleton. The experimental result shows that the proposed method running faster and increases the quality and the accuracy about 10% higher than the highest result of the compared method.

Many scientists use a global characterization of bare soil surface random roughness. Surface roughness is often characterized by statistical parameters deduced from its autocorrelation function. Assuming an autocorrelation model and a Gaussian height distribution, some authors have developed algorithms for numerical generation of soil surfaces that have the same statistical properties. This approach is widespread and does not take into account morphological aspects of the soil surface micro-topography. Now a detail surface roughness analysis reveals that the micro-topography is structured by holes, aggregates and clods. In the present study, we clearly show that when describing surface roughness as a whole, some information related to morphological aspects is lost. Two Digital Elevation Model (DEM) of a same natural seedbed surface were recorded by stereo photogrammetry. After estimating global parameters of these natural surfaces, we generated numerical surfaces of the same average characteristics by linear filtering. Big aggregates and clods were then captured by a contour-based approach. We show that the two-dimensional autocorrelation functions of generated surfaces and of the two agricultural surfaces are close together. Nevertheless, the number and shape of segmented object contours change from generated surfaces to the natural surfaces. Generated surfaces show fewer and bigger segmented objects than in the natural case. Moreover, the shape of some segmented objects is unrealistic in comparison to real clods, which have to be convex and of low circularity.

We investigate the potential of multidimensional mutual information for the registration of multi-spectral remote sensing images. We devise a gradient flow algorithm which iteratively maximizes the multidimensional mutual information with respect to a differentiable displacement map, accounting for partial derivatives of the multivariate joint distribution and the multivariate marginal of the float image with respect to each variable of the mutual information derivative. The resulting terms are shown to weight the band specific gradients of the warp image, and we propose in addition to compute them with a method based on the k-nearest neighbours. We apply our method to the registration of Ikonos and CHRIS-Proba images over the region of Baabdat, Lebanon, for purposes of cedar pines detection. A comparison between (crossed) single band and multi-band registration results obtained shows that using the multidimensional mutual information brings a significant gain in positional accuracy and is suitable for multispectral remote sensing image registration.

Image registration is a fundamental and crucial step in remote sensing image analysis. However, it is known that image registration method is application-based. The type and content of remote sensing images affect the choice of image registration methods. Previous image registration task took experts to manually choose the image registration elements. This paper presents a self-adaptive image registration method which could automatically choose registration elements which are more appropriate for remote sensing images under processing. The proposed method first chooses several local regions for the representation of the whole image, and then different registration elements are tested on these local regions. The local registration results are evaluated and the registration of the whole image is done with learned registration elements from local registrations. The registration chain is automatic; therefore it is a self-adaptive registration method. The proposed method is demonstrated on several real remote sensing image pairs, and its feasibility and superiority are proved by the results.

We introduce an approach for the detection of approximately rectangular structures in gray scale images. Our research is motivated by the Silvretta Historica project that aims at automated detection of remains of livestock enclosures in remotely sensed images of alpine regions. The approach allows detection of enclosures with linear sides of various sizes and proportions. It is robust to incomplete or fragmented rectangles and tolerates deviations from a perfect rectangular shape. Morphological operators are used to extract linear features. They are grouped into parameterized linear segments by means of a local Hough transform. To identify appropriate configurations of linear segments we define convexity and angle constraints. Configurations meeting these constraints are rated by a proposed rectangularity measure that discards overly fragmented configurations and configurations with more than one side completely missing. The search for appropriate configurations is efficiently performed on a graph. Its nodes represent linear segments and edges encode the above constraints. We tested our approach on a set of aerial and GeoEye-1 satellite images of 0.5m resolution that contain ruined livestock enclosures of approximately rectangular shape. The approach showed encouraging results in finding configurations of linear segments originating from the objects of our interest.

Recently, earthquake damage assessment using satellite images has been a very popular ongoing research direction. Especially with the availability of very high resolution (VHR) satellite images, a quite detailed damage map based on building scale has been produced, and various studies have also been conducted in the literature. As the spatial resolution of satellite images increases, distinguishability of damage patterns becomes more cruel especially in case of using only the spectral information during classification. In order to overcome this difficulty, textural information needs to be involved to the classification to improve the visual quality and reliability of damage map. There are many kinds of textural information which can be derived from VHR satellite images depending on the algorithm used. However, extraction of textural information and evaluation of them have been generally a time consuming process especially for the large areas affected from the earthquake due to the size of VHR image. Therefore, in order to provide a quick damage map, the most useful features describing damage patterns needs to be known in advance as well as the redundant features. In this study, a very high resolution satellite image after Iran, Bam earthquake was used to identify the earthquake damage. Not only the spectral information, textural information was also used during the classification. For textural information, second order Haralick features were extracted from the panchromatic image for the area of interest using gray level co-occurrence matrix with different size of windows and directions. In addition to using spatial features in classification, the most useful features representing the damage characteristic were selected with a novel feature selection method based on high dimensional model representation (HDMR) giving sensitivity of each feature during classification. The method called HDMR was recently proposed as an efficient tool to capture the input-output relationships in high-dimensional systems for many problems in science and engineering. The HDMR method is developed to improve the efficiency of the deducing high dimensional behaviors. The method is formed by a particular organization of low dimensional component functions, in which each function is the contribution of one or more input variables to the output variables.

In this study, data from the EO-1 Hyperion instrument were used. Apart from atmospheric influences or topographic effects, the data represent a good choice in order to show different steps of the preprocessing process targeting sensorinternal sources of errors. These include the diffuse sensor noise, the striping effect, the smile effect, the keystone effect and the spatial misalignments between the detector arrays. For this research paper, the authors focus on the striping effect by comparing and evaluating different algorithms, methods and configurations to correct striping errors. The correction of striping effects becomes necessary due to the imprecise calibration of the detector array. This inaccuracy affects especially the first 12 visual and near infrared bands (VNIR) and also a large number of the bands in the short wave infrared array (SWIR). Altogether six destriping techniques were tested on the basis of a Hyperion dataset covering a test site in Central Europe. For the final evaluation, various analyses across all Hyperion channels were performed. The results show that some correction methods have almost no effect on the striping in the images. Other methods may eliminate the striping, but analyses show that these algorithms also alter pixel values in adjacent areas which originally had not been disturbed by the striping effect.

In this paper a penalized least squares cost function with a new spatial-spectral penalty is proposed for hyper- spectral image restoration. The new penalty is a combination of a Group LASSO (GLASSO) and First Order Roughness Penalty (FORP) in the wavelet domain. The restoration criterion is solved using the Alternative Direction Method of Multipliers (ADMM). The results are compared with other restoration methods where the proposed method outperforms them for the simulated noisy data set based on Signal to Noise Ratio (SNR) and visually outperforms them on a real degraded data set.

In this paper a new unsupervised nonparametric cooperative and adaptive hyperspectral image segmentation approach is presented. The hyperspectral images are partitioned band by band in parallel and intermediate classification results are evaluated and fused, to get the final segmentation result. Two unsupervised nonparametric segmentation methods are used in parallel cooperation, namely the Fuzzy C-means (FCM) method, and the Linde-Buzo-Gray (LBG) algorithm, to segment each band of the image. The originality of the approach relies firstly on its local adaptation to the type of regions in an image (textured, non-textured), and secondly on the introduction of several levels of evaluation and validation of intermediate segmentation results before obtaining the final partitioning of the image. For the management of similar or conflicting results issued from the two classification methods, we gradually introduced various assessment steps that exploit the information of each spectral band and its adjacent bands, and finally the information of all the spectral bands. In our approach, the detected textured and non-textured regions are treated separately from feature extraction step, up to the final classification results. This approach was first evaluated on a large number of monocomponent images constructed from the Brodatz album. Then it was evaluated on two real applications using a respectively multispectral image for Cedar trees detection in the region of Baabdat (Lebanon) and a hyperspectral image for identification of invasive and non invasive vegetation in the region of Cieza (Spain). A correct classification rate (CCR) for the first application is over 97% and for the second application the average correct classification rate (ACCR) is over 99%.

This paper addresses lossy compression of hyperspectral images acquired by sensors of new generation for which signaldependent component of the noise is prevailing compared to the noise-independent component. First, for sub-band (component-wise) compression, it is shown that there can exist an optimal operation point (OOP) for which MSE between compressed and noise-free image is minimal, i.e., maximal noise filtering effect is observed. This OOP can be observed for two approaches to lossy compression where the first one presumes direct application of a coder to original data and the second approach deals with applying direct and inverse variance stabilizing transform (VST). Second, it is demonstrated that the second approach is preferable since it usually provides slightly smaller MSE and slightly larger compression ratio (CR) in OOP. One more advantage of the second approach is that the coder parameter that controls CR can be set fixed for all sub-band images. Moreover, CR can be considerably (approximately twice) increased if sub-band images after VST are grouped and lossy compression is applied to a first sub-band image in a group and to “difference” images obtained for this group. The proposed approach is tested for Hyperion hyperspectral images and shown to provide CR about 15 for data compression in the neighborhood of OOP.

A classical spectral un-mixing of hyperspectral image involves identifying the unique signatures of the endmembers (i.e. pure materials) and estimating the proportions of endmembers for each pixel by inversion. The key to successful spectral un-mixing is indicating the number of endmembers and their corresponding spectral signatures. Currently, eigenvaluebased estimation of the number of endmembers in hyperspectral image is widely used. However, the eigenvalue-based methods are difficult to separate signal sources such as anomalies. In this paper, a two-stage process is proposed to estimate the endmember numbers. At the preprocessing stage, the spectral dimensions are reduced using principal component analysis and the spatial dimensions are reduced using convex hull computation based on reduced-spectral bands. At the hierarchical agglomerate clustering stage, a pixel vector is found by applying orthogonal subspace projection (OSP) and cluster pixel vectors using the spectral angle mapper (SAM), hierarchically. If the number of pixel vectors in a cluster is greater than the predefined number, the found pixel vector is set as the endmember. Otherwise, anomalous vectors are found. The proposed method was carried with both synthetic and real images for estimating the number of endmembers. The results demonstrate that the proposed method can be used to estimate more reasonable and precise number of endmembers than the eigenvalue-based methods.

Singular value decomposition (SVD) and principal component analysis enjoy a broad range of applications, including, rank estimation, noise reduction, classification and compression. The resulting singular vectors form orthogonal basis sets for subspace projection techniques. The procedures are applicable to general data matrices. Spectral matrices belong to a special class known as non-negative matrices. A key property of non-negative matrices is that their columns/rows form non-negative cones, with any non-negative linear combination of the columns/rows belonging to the cone. This special property has been implicitly used in popular rank estimation techniques know as virtual dimension (VD) and hyperspectral signal identification by minimum error (HySime). Data sets of spectra reside in non-negative orthants. The subspace spanned by a SVD of a set of spectra includes all orthants. However SVD projections can be constrained to the non-negative orthants. In this paper two types of singular vector projection constraints are identified, one that confines the projection to lie within the cone formed by the spectral data set, and a second that only restricts projections to the non-negative orthant. The former is referred to here as the inner constraint set, the latter the outer constraint set. The outer constraint set forms a broader cone since it includes projections outside the cone formed by the data array. The two cones form boundaries for the cones formed by non-negative matrix factorizations (NNF). Ambiguities in the NNF lead to a variety of possible sets of left and right non-negative vectors and their cones. The paper presents the constraint set approach and illustrates it with applications to spectral classification.

The new generation of hyperspectral sensors can provide images with a high spectral and spatial resolution. Recent improvements in mathematical morphology have developed new techniques such as the Attribute Profiles (APs) and the Extended Attribute Profiles (EAPs) that can effectively model the spatial information in remote sensing images. The main drawbacks of these techniques is the selection of the optimal range of values related to the family of criteria adopted to each filter step, and the high dimensionality of the profiles, which results in a very large number of features and therefore provoking the Hughes phenomenon. In this work, we focus on addressing the dimensionality issue, which leads to an highly intrinsic information redundancy, proposing a novel strategy for extracting spatial information from hyperspectral images based on the analysis of the Differential Attribute Profiles (DAPs). A DAP is generated by computing the derivative of the AP; it shows at each level the residual between two adjacent levels of the AP. By analyzing the multilevel behavior of the DAP, it is possible to extract geometrical features corresponding to the structures within the scene at different scales. Our proposed approach consists of two steps: 1) a homogeneity measurement is used to identify the level L in which a given pixel belongs to a region with a physical meaning; 2) the geometrical information of the extracted regions is fused into a single map considering their level L previously identified. The process is repeated for different attributes building a reduced EAP, whose dimensionality is much lower with respect to the original EAP ones. Experiments carried out on the hyperspectral data set of Pavia University area show the effectiveness of the proposed method in extracting spatial features related to the physical structures presented in the scene, achieving higher classification accuracy with respect to the ones reported in the state-of-the-art literature

The affinity propagation (AP)¹ is now among the most used methods of unsupervised classification. However, it has two major disadvantages. On the one hand, the algorithm implicitly controls the number of classes from a preference parameter, usually initialized as the median value of the similarity matrix, which often gives over-clustering. On the other hand, when partitioning large size hyperspectral images, its computational complexity is quadratic and seriously hampers its application. To solve these two problems, we propose a method which consists of reducing the number of individuals to be classified before the application of the AP, and to concisely estimate the number of classes. For the reduction of the number of pixels, a pre-classification step that automatically aggregates highly similar pixels is introduced. The hyperspectral image is divided into blocks, and then the reduction step is applied independently within each block. This step requires less memory storage since the calculation of the full similarity matrix is not required. The AP is then applied on the new set of pixels which are then set up from the representatives of each previously formed cluster and non-aggregated individuals. To estimate the number of classes, we introduced a dichotomic method to assess classification results using a criterion based on inter-class variance. The application of this method on various test images has shown that AP results are stable and independent to the choice of the block size. The proposed approach was successfully used to partition large size real datasets (multispectral and hyperspectral images).

Spectral Mixture Analysis is a widely used image analysis tool with many applications. Yet, one of the major issues with this technique remains the lack of ability to properly account for the spectral variability of endmembers or ground cover components that occur throughout an image scene. Endmember variability is most often addressed using iterative mixture cycles (e.g. MESMA) in which different endmember combination models are compared for each pixel. The model with the best fit is assigned to the pixel. The drawback of MESMA is the computational burden which often hampers the operational use. In an attempt to address this issue we proposed a new geometric based methodology to more efficiently evaluate different endmember combinations in MESMA. This geometric unmixing methodology has a two-fold benefit. First of all, geometric unmixing allows a fast and fully constrained unmixing, which was previously unfeasible in MESMA due to the long processing times of the available fully constrained unmixing methods. Secondly, whereas the traditional MESMA explores all different endmember combinations separately, and selects the most appropriate combination as a final step, our approach selects the best endmember combination prior to unmixing, as such increasing the computational efficiency of MESMA. To do so, we built upon the equivalence between the reconstruction error in least-squares unmixing and spectral angle minimization in geometric unmixing. With the inclusion of the proposed endmember combination selection technique, the computation time decreased by a factor between 5 and 8.5, depending on the size and organization of the libraries. The spectral angle can as such be used as a proxy for model fit, enabling the selection of the proper endmember combination from large spectral libraries prior to unmixing.

We present a sparse coding based spectral-spatial classification model for hyperspectral image (HSI) datasets. The proposed method consists of an efficient sparse coding method in which the l₁/l_q regularized multi-class logistic regression technique was utilized to achieve a compact representation of hyperspectral image pixels for land cover classification. We applied the proposed algorithm to a HSI dataset collected at the Kennedy Space Center and compared our algorithm to a recently proposed method, Gaussian process maximum likelihood (GP-ML) classifier. Experimental results show that the proposed method can achieve significantly better performances than the GP-ML classifier when training data is limited with a compact pixel representation, leading to more efficient HSI classification systems.

A new binary optimization method inspired on the Fractional-Order Darwinian Particle Swarm Optimization is proposed and applied to a novel spectral-spatial classification framework.Afterwards, the new optimization algorithm is used in a novel spectral-spatial classification frameworkfor the selection of the most effective group of bands. In the proposed approach, first, the raw data set (only spectral data) along with the morphological profiles of the first effective principal components are integrated into a stacked vector. Then, the output of this step is considered as the input of the new optimization method. The Random Forest classifier is used as a fitness function for the cross-validation samples and the overall classification accuracy for the evaluation of the group of bands.Finally, the selected bands are classified by a classifier and the output provides the final classification map. Experimental results successfully confirm that the new approach works better than whenconsideringall the raw bands, the whole morphological profile and the combination of the raw bands and morphological profile.

Markov random field (MRF) is currently the most common method to find the optimal solution for the classification of image data incorporating contextual visual information. The labeling for a site in MRF is dependent on smoothing parameters. Therefore, this paper deals with the development of a new robust two-step method to determine the smoothing parameter which balances spatial and spectral energies for the purpose of maximizing the classification accuracy. Multispectral images obtained by WorldView-2 satellite were employed in this research. In the first step, a support vector machine (SVM) was used to provide a vector of multi-class probability and a class label for each pixel. Then, the summation of the maximum probability of each pixel and its 8 neighbors is calculated for a dynamic block and this value is assigned to the central pixels of each block. The blocks of each class are sorted and an equal proportion of blocks of each class with the highest probability are selected. Then, the class codes and spectral information of the selected blocks are extracted from the classified map and multispectral image, respectively. This information is used to calculate class label co-occurrence matrices of the blocks (CLCMB), class label co-occurrence matrix (CLCM) and class separability indices. Finally, different smoothing parameters are calculated and the results show that estimated smoothing parameter can produce a more accurate map.

A fully automatized method which can extract road networks by using the spectral and structural features of the roads is proposed. First, Anti-parallel Centerline Extraction (ACE) is used to obtain road seed points. Then, the road seeds are improved with perceptual grouping method and the road class is determined with Maximum Likelihood Estimation (MLE) by modeling the seed points with Gaussian Mixture. The morphological operations (opening, closing and thinning) are performed for improving classification results and determining the road topology roughly. Finally, perceptual grouping is performed for removing non-road line segments and filling the gaps on the topology. The proposed algorithm is tested on 1 meter resolution IKONOS images and results better than previous algorithms are obtained.

In this paper, we present an approach to generate a 3D model of an urban scene out of sensor data. The first milestone on that way is to classify the sensor data into the main parts of a scene, such as ground, vegetation, buildings and their outlines. This has already been accomplished within our previous work. Now, we propose a four-step algorithm to model the building structure, which is assumed to consist of several dominant planes. First, we extract small elevated objects, like chimneys, using a hot-spot detector and handle the detected regions separately. In order to model the variety of roof structures precisely, we split up complex building blocks into parts. Two different approaches are used: To act on the assumption of underlying 2D ground polygons, we use geometric methods to divide them into sub-polygons. Without polygons, we use morphological operations and segmentation methods. In the third step, extraction of dominant planes takes place, by using either RANSAC or J-linkage algorithm. They operate on point clouds of sufficient confidence within the previously separated building parts and give robust results even with noisy, outlier-rich data. Last, we refine the previously determined plane parameters using geometric relations of the building faces. Due to noise, these expected properties of roofs and walls are not fulfilled. Hence, we enforce them as hard constraints and use the previously extracted plane parameters as initial values for an optimization method. To test the proposed workflow, we use both several data sets, including noisy data from depth maps and data computed by laser scanning.

In this paper we discuss how multisource data (wind, ocean-current, optical, bathymetric, automatic identification systems (AIS)) may be used to improve oil spill detection in SAR images, with emphasis on the use of automatic oil spill detection algorithms. We focus particularly on AIS, optical, and bathymetric data. For the AIS data we propose an algorithm for integrating AIS ship tracks into automatic oil spill detection in order to improve the confidence estimate of a potential oil spill. We demonstrate the use of ancillary data on a set of SAR images. Regarding the use of optical data, we did not observe a clear correspondence between high chlorophyll values (estimated from products derived from optical data) and observed slicks in the SAR image. Bathymetric data was shown to be a good data source for removing false detections caused by e.g. sand banks on low tide. For the AIS data we observed that a polluter could be identified for some dark slicks, however, a precise oil drift model is needed in order to identify the polluter with high certainty.

In the present work we applied a recently developed procedure for multidimensional data clustering to multispectral satellite images. The core of our approach lays in projection of the multidimensional image to a two dimensional space. For this purpose we used extensively investigated family of recurrent artificial neural networks (RNN) called “Echo state network” (ESN). ESN incorporates a randomly generated recurrent reservoir with sigmoid nonlinearities of neurons outputs. The procedure called Intrinsic Plasticity (IP) that is aimed at reservoir output entropy maximization was applied for adapting of reservoir steady states to the multidimensional input data. Next we consider all possible combinations between steady states of each two neurons in the reservoir as two-dimensional projections of the original multidimensional data. These low dimensional projections were subjected to subtractive clustering in order to determine number and position of data clusters. Two approaches to choose a proper projection among the all possible combinations between neurons were investigated. The first one is based on the calculation of two-dimensional density distributions of each projection, determination of number of their local maxima and choice of the projections with biggest number of these maxima. The second one applies clustering to all projections and chooses those with maximum number of clusters. Multispectral data from Landsat 7 Enhanced Thematic Mapper Plus (ETM+) instrument are used in this work. The obtained number and position of clusters of a multi-spectral image of a mountain region in Bulgaria is compared with the regional landscape classification.

Optical remote sensing data is commonly used for estimating biophysical characteristics of forest like tree biodiversity and species richness. Recent advances in radar remote sensing technology raise significant interest to take advantage of the complementary nature of optical and radar data. This paper proposes an approach combining Landsat ETM+ and Advanced Land-Observing Satellite- Phased Array L-band Synthetic Aperture Radar (ALOS/PALSAR) data for forest biodiversity and species richness monitoring. Inventory data from one part of the Hyrcanian forest in north of Iran is used as field data. Visible and infrared ETM+ bands, indices and textures information as well as HH, HV backscattering, polarimetric features (alpha angle, entropy and anisotropy) and SAR texture are extracted. We use the multiple linear regression model to find the best components to describe the biodiversity indices in the study area. We show how tree biodiversity is related to information derived from ETM+ and ALOS/PALSAR data at 95% confidence level (R!=0.63, root mean square error (RMSE) =1.70). Also, the effects of each component on the variation of biodiversity are shown. ETM+ reflectance, polarimetric features and texture from ALOS/PALSAR can describe approximately 63% of biodiversity. The results of multi source monitoring of tree biodiversity and species richness are promising and worth further investigation.

Forest fragmentation studies have increased since the last 3 decades. Land use and land cover maps (LULC) are important tools for this analysis, as well as other remote sensing techniques. The object oriented analysis classifies the image according to patterns as texture, color, shape, and context. However, there are many attributes to be analyzed, and data mining tools helped us to learn about them and to choose the best ones. In this way, the aim of this paper is to describe data mining techniques and results of a heterogeneous area, as the municipality of Silva Jardim, Rio de Janeiro, Brazil. The municipality has forest, urban areas, pastures, water bodies, agriculture and also some shadows as objects to be represented. Worldview 2 satellite image from 2010 was used and LULC classification was processed using the values that data mining software has provided according to the J48 method. Afterwards, this classification was analyzed, and the verification was made by the confusion matrix, being possible to evaluate the accuracy (58,89%). The best results were in classes “water” and “forest” which have more homogenous reflectance. Because of that, the model has been adapted, in order to create a model for the most homogeneous classes. As result, 2 new classes were created, some values and some attributes changed, and others added. In the end, the accuracy was 89,33%. It is important to highlight this is not a conclusive paper; there are still many steps to develop in highly heterogeneous surfaces.

Monitoring of earth surface changes from space by using multi-date satellite imagery was always a main concern to researchers in the field of remotely sensed image processing. Thus, several techniques have been proposed to saving technicians from interpreting and digitizing hundreds of areas by hand. The exploiting of simple, easy to memorize and often comprehensible mathematical models such band-ratios and indices are one of the widely used techniques in remote sensing for the extraction of particular land-cover/land-use like urban and vegetation areas. The results of these models generally only need the definition of adequate threshold or using simple unsupervised classification algorithms to discriminate between the class of interest and the background. In our work a genetic programming based approach has been adopted to evolve simple mathematical expression to extract urban areas from image series. The model is built from a single image by using a basic set of operators between spectral bands and maximizing a fitness function, which is based on the using of the M-statistic criterion. The model was constructed from the Landsat 5 TM image acquired in 2006 by using training samples extracted with the help of a Quick-bird high spatial resolution satellite image acquired the same day as the Landsat image over the city of Oran, Algeria. The model has been tested to extract urban areas from multi-date series of Landsat TM imagery

Building damage such as to side-walls or mid-story collapse is often overlooked in vertical optical images. Hence, in order to observe such building damage modes, high-resolution SAR images are introduced considering the side-looking nature of SAR. In the 2011 Tohoku, Japan, earthquake, a large number of buildings were collapsed or severely damaged due to repeated tsunamis. One of the important tsunami effects on buildings is that the damage is concentrated to their side-walls and lower stories. Thus this paper proposes the method to detect this kind damage from the change in layover areas in SAR intensity images. Multi-temporal TerraSAR-X images covering the Sendai-Shiogama Port were employed to detect building damage due to the tsunamis caused by the earthquake. The backscattering coefficients in layover areas of individual buildings were extracted and then, the average value in each layover area was calculated. The average value was seen to decrease in the post-event image due to the reduced backscatter from building side-walls. This example demonstrated the usefulness of high-resolution SAR intensity images to detect severe damage to building side-walls based on the changes of the backscattering coefficient in the layover areas.

The temporal evolution of pixel values in Satellite Image Time Series (SITS) is considered as criterion for the characterization, discrimination and identification of terrestrial objects and phenomena. Due to the exponential behavior of sequences number with specialization, Sequential Data Mining (SDM) techniques need to be applied. The huge search and solution spaces imply the use of constraints according to the user’s knowledge, interest and expectation. The spatial aspect of the data was taken into account by the introduction of connectivity measures that characterize the pixels tendency to form objects. These measures can highlight stratifications in data structure, can be useful for shape recognition and offer a base for post-processing operations similar to those from mathematical morphology (dilation, erosion etc.). The conjunction of corresponding Connectivity Constraints (CC) with the Support Constraint (SC) leads to the extraction of Grouped Frequent Sequential Patterns (GFSP), a concept with proved capability for preliminary description and localization of terrestrial events. This work is focused on efficient SITS extraction of evolutions that fulfill SC and CC. Different types of extractions using anti-monotone constraints are analyzed. Experiments performed on two interferometric SITS are used to illustrate the potential of the approach to find interesting evolution patterns.

In this paper we propose a framework for fusion of very high resolution (VHR) optical aerial images, satellite images (optical or SAR) and other ancillary data (e.g. a digital elevation model) for identification and modeling of nature types typically present in mountain vegetation in Arctic alpine areas. The data fusion methodology consists of three steps. (i) Segmentation of VHR aerial photo into spectrally homogeneous regions (polygons). (ii) Estimation of complementary information for each polygon using geo-referenced data from other sources. (iii) Analysis of the constructed feature vectors. We also demonstrated the strength of satellite data by qualitatively evaluating the potential for creating high resolution snow cover maps. These maps may be used to describe important environmental variables. Using a set of data consisting of an aerial photo, two SPOT 5 images and a Radarsat-2 quad-pol image, we demonstrated the potential of the data fusion methodology by an example where the polygon-derived features were analysed using PCA.

In this paper, we propose a change detection approach based on nonlinear scale space analysis of change images for robust detection of various changes incurred by natural phenomena and/or human activities in Synthetic Aperture Radar (SAR) images using Maximally Stable Extremal Regions (MSERs). To achieve this, a variant of the log-ratio image of multitemporal images is calculated which is followed by Feature Preserving Despeckling (FPD) to generate nonlinear scale space images exhibiting different trade-offs in terms of speckle reduction and shape detail preservation. MSERs of each scale space image are found and then combined through a decision level fusion strategy, namely “selective scale fusion” (SSF), where contrast and boundary curvature of each MSER are considered. The performance of the proposed method is evaluated using real multitemporal high resolution TerraSAR-X images and synthetically generated multitemporal images composed of shapes with several orientations, sizes, and backscatter amplitude levels representing a variety of possible signatures of change. One of the main outcomes of this approach is that different objects having different sizes and levels of contrast with their surroundings appear as stable regions at different scale space images thus the fusion of results from scale space images yields a good overall performance.

The aim of this paper is to analyse spectral reflectance data from Landsat TM of vegetation that has been exposed to hydrocarbon contamination from oil spills from pipelines. The study is undertaken in an area of mangrove and swamp vegetation where the detection of an oil spill is traditionally difficult to make. We used a database of oil spill records to help identify candidate sites for spectral analysis. Extracted vegetation spectra were compared between polluted and nonpolluted sites and supervised (neural network) classification was carried out to map hydrocarbon (HC) contaminated sites from the sample areas. Initial results show that polluted sites are characterised by high reflectance in the visible (VIS) 0.4μm - 0.7μm, and a lower reflectance in the near-infrared (NIR) 0.7μm - 1.1μm. This suggests that the vegetation is in a stressed state. Samples taken from pixels surrounding polluted sites show similar spectral reflectance values to that of polluted sites suggesting possible migration of HC to the wider environment. Further work will focus on increasing the sample size and investigating the impact of an oil spill on a wider buffer zone around the spill site.

This paper outlines the development of a robust, automated and real-time horizon detection and tracking system for the purpose of distance estimation of the sea surface. The distance information facilitates visual monitoring of the sea surface in remote sensing or monitoring applications. The specific application of the system in marine mammal mitigation is used to trial and demonstrate the effectiveness of the system. Results from the trials are presented and shows the effectiveness of the system.

Solar Orbiter is an ESA space mission devoted to improve the knowledge of those effects nowadays still not fully understood on the physical mechanisms underlying the behaviour of our star. The mission has a peculiar trajectory that will bring the S/C close to the Sun up to 0.28 AU, exploiting the opportunity to follow up our star as close as never before. METIS, one of the instruments selected to be part of the Solar Orbiter payload, is a coronagraph that will investigate the inner part of the heliosphere performing imaging in the visible band and in the hydrogen Lyman α line @ 121.6 nm. METIS will be able to simultaneously operate the two detectors: an Intensified APS for the UV channel and an APS for the visible light, and a Liquid Crystal Variable Retarder (LCVR) plate, for broadband visible polarimetry. They will be operated by means of the centralised management unit of the instrument, the METIS Processing and Power Unit. This payload subsystem hosts a microprocessor that implements, thanks to the application software, all the needed functionalities to fully control the instrument subsystems and its own processing capabilities. Both sensors will be readout at high rate and the acquired data shall undergo through a preliminary on-board processing to maximize the scientific return and to provide the necessary information to validate the results on ground. Being Solar Orbiter a deepspace mission, some METIS procedures have been designed to provide to the instrument an efficient autonomous behaviour in case of an immediate reaction is required as for arising transient events or occurrence of safety hazards condition. METIS will implement an on-board algorithm for the automatic detection of this kind of events in order to promptly react and autonomously adapt the observing procedure.

Land-water segmentation is an important preprocessing step in a number of remote sensing applications such as target detection, environmental monitoring, and map updating. A Normalized Optical Water Index (NOWI) is proposed to accurately discriminate between land and water regions in multi-spectral satellite imagery data from DubaiSat-1. NOWI exploits the spectral characteristics of water content (using visible bands) and uses a non-linear normalization procedure that renders strong emphasize on small changes in lower brightness values whilst guaranteeing that the segmentation process remains image-independent. The NOWI representation is validated through systematic experiments, evaluated using robust metrics, and compared against various supervised classification algorithms. Analysis has indicated that NOWI has the advantages that it: a) is a pixel-based method that requires no global knowledge of the scene under investigation, b) can be easily implemented in parallel processing, c) is image-independent and requires no training, d) works in different environmental conditions, e) provides high accuracy and efficiency, and f) works directly on the input image without any form of pre-processing.

The characterization of urban environments in synthetic aperture radar (SAR) images is becoming increasingly challenging with the increased spatial ground resolutions. In SAR images having a geometrical resolution of few meters (e.g. 3 m), urban scenes are roughly speaking characterized by three main types of backscattering: low intensity, medium intensity, and high intensity, which correspond to different land-cover types. Based on the observations of the behavior of the backscattering, in this paper we propose the labeled co-occurrence matrix (LCM) technique to detect and extract built-up areas. Two textural features, autocorrelation and entropy, are derived from LCM. The image classification is based on a similarity classifier defined in the general Lukasiewicz structure. Experiments have been carried out on TerraSAR-X images acquired on Nanjing (China) and Barcelona (Spain), respectively. The obtained classification accuracies point out the effectiveness of the proposed technique in identifying and detecting built-up areas compared with the traditional grey level co-occurrence matrix (GLCM) texture features.

Image coregistration is an important step in SAR interferometry which is a well known method for DEM generation and surface displacement monitoring. A practical and widely used automatic coregistration algorithm is based on selecting a number of tie points in the master image and looking for the correspondence of each point in the slave image using correlation technique. The characteristics of these points, their number and their distribution have a great impact on the reliability of the estimated transformation. In this work, we present a method for automatic selection of suitable tie points that are well distributed over the common area without decreasing the desired tie points’ number. First we select candidate points using Harris operator. Then from these points we select tie points depending on their cornerness measure (the highest first). Once a tie point is selected, its correspondence is searched for in the slave image, if the similarity measure maximum is less than a given threshold or it is at the border of the search window, this point is discarded and we proceed to the next Harris point, else, the cornerness of the remaining candidates Harris points are multiplied by a spatially radially increasing function centered at the selected point to disadvantage the points in a neighborhood of a radius determined from the size of the common area and the desired number of points. This is repeated until the desired number of points is selected. Results of an ERS1/2 tandem pair are presented and discussed.

The noise like quality characteristic of SAR images known as speckle become most critical impediment for automatic segmentation and classification of targets. For last few decades many adaptive speckle filters were proposed. Most of these classical filters are single stage i.e. repeated use of them causes blurring of image features (e.g. edges or textures) and generates artefacts. But iterative uses of filters are required for desired amount of smoothing. To retain structure in adaptive filtering, key component is precise estimation of scene heterogeneity. Failure of traditional filters to retain features upon iteration is due to failure to measure scene heterogeneity optimally. Perona and Malik proposed an Anisotropic Diffusion (AD) equation which iteratively diffuses gray values by preserving edges. Black et al developed a robust statistical interpretation of AD, and opened a broader context to choose between alternative diffusion functions. Depending on Black’s work earlier we proposed a robust speckle reducing anisotropic diffusion (ROSRAD) filter and also a heterogeneity scale based on Otsu’s thresholding algorithm. This scale is optimal but not robust. This paper is an extension to our work, here we propose a different heterogeneity scale which is robust and performs better for speckle noise distribution.

The dynamic responses of the polar ice sheets in Greenland and Antarctica can have substantial impacts on sea level rise. Understanding the mass balance requires accurate assessments of the bedrock and surface layers, but identifying each layer in ground-penetrating radar imagery must typically be performed by time-consuming hand selection. We have developed an approach for semi-automatically estimating bedrock and surface layers from radar depth sounder imagery acquired from Antarctica. Our solution utilizes an active contours method (level sets"), which identifies surface and bedrock boundaries by evolving initial estimates of a layer's position and depth until a gradient-based cost function is minimized. We evaluated the semi-automatic proposed method on 20 images with respect to hand labeled ground-truth. Compared to an existing automatic technique, our approach reduced labeling error by factors of 5 and 3.5 for tracing bedrock and surface layers, respectively.

This paper reports an experiment conducted in order to compare lossy JPEG compression and lossy JPEG 2000 compression on the accuracy of image matching. The experiment has been conducted by using 54 color images of diverse textures and diverse tones of color on the assumption that image matching utilizes a pair of images reconstructed from image data which are lossily compressed in an ordinary digital camera. Lossy JPEG compression has been executed with a set of compression parameters utilized in a digital camera Canon EOS 20D, while lossy JPEG 2000 compression has been executed in the way as the file size of a piece of JPEG 2000 compressed image data is the same file size of the corresponding JPEG compressed image data. Moreover, we have prepared another set of JPEG and JPEG 2000 compressed image data which are lossily compressed with the compression ratio expected for an ordinary color image when it is compressed in the EOS 20D. When the file size of a piece of JPEG 2000 compressed image data is the same as that of a piece of JPEG compressed image data, the experiment results clearly show that the compression performance of JPEG 2000 compression would be superior to that of JPEG compression on image quality as to pixel value in RGB color space. On the contrary, the results do not necessarily indicate that JPEG 2000 compression would be able to provide more accurate matching results than JPEG compression.

In remote sensing image applications, the image matching is a very key technology, its quality directly related to the quality of the subsequent results. This paper studied an improved SIFT features matching method for muili-source remote-sensing image registration based on GPU computing, epipolar line and least squares, its main purpose is to take both accuracy and efficiency into consideration. This method is firstly based on tonal balanced methods matching, and then exracts SIFT features based on the GPU computing technology, and then matchs feature points based on epipolar line and least squares matching method with RANSAC method, finally analies error sources of SIFT mismatch, researchs an improved SIFT mismatch reduce strategy.The experimental results prove that the method can effectively improve the efficiency and precision of SIFT feature matching.

This research is conducted in order to develop a spectral device for instant remote sensing. The spectral device should process radiation from sources, direct contact with radiation of which is either impossible or undesirable. In proposed spectral device optical radiation is guided out of unfavorable environment via a piece of optical fiber with high negative dispersion. Dispersion properties of optical fiber allow using it as a key element of proposed spectral device. In this paper stationary in time optical radiation energy spectrum assessment is proposed using a modified Bartlett algorithm. Original Bartlett algorithm is based on power sample spectrum calculation on the series of following samples with duty cycle Q=1. Bartlett algorithm modification is a significant increase in duty cycle (Q>>1) of stationary process samples. Proposed algorithm is to be realized basing on complex sample spectra calculation using a piece of optical fiber as a dispersive system. Dispersion properties application to acquire sample optical spectra causes that time function of a single sample spectrum has much longer duration than the original sample. This is the reason to modify the Bartlett algorithm. Significant increase of duty cycle Q means that separate analyzed radiation samples are uncorrelated in time. This allows considering set of separate samples as an ensemble of N stochastic process realizations. Temporal stationarity condition allows acquiring a big number of sample spectra from a single realization. In this paper we perform mathematical analysis of optical radiation processing using modified Bartlett algorithm. A complex sample spectrum is described as convolution of window spectrum function and complex spectral function of optical radiation. Power spectrum is considered as product of complex spectrum and its complex. Spectral slot of modified Bartlett algorithm matches the spectral slot of original Bartlett algorithm.

Coherent multiple-input multiple-output (MIMO) radar systems with co-located antennas, form monostatic vir- tual arrays by discrete convolution of a bistatic setup of transmitters and receivers. Thereby, a trade-off between maximum array dimension, element spacing and hardware efforts exists. In terms of estimating the direction of arrival, the covariance matrix of the array element signals plays an important role. Here, minimum redundancy arrays aim at a hardware reduction with signal reconstruction by exploiting the Toeplitz characteristics of the covariance matrix. However, the discrete spatial convolution complicates the finding of an optimal antenna setup with minimum redundancy. Combinatorial effort is the consequence. This paper presents a possible simplified algorithm in order to find MIMO array setups of maximum dimension with minimum redundancy.

With the rapid development of remote sensing technology and variety of earth observation satellites have been successfully launched, the volume of image datasets is growing exponentially in many application areas. The Contentbased image retrieval (CBRSIR), as an efficient means for management and utilization of the information in image database from the viewpoint of comprehension of image content, is applied on the remote sensing images retrieval. However, one kind of features always can’t express the image content exactly. So, a multi-feature retrieval model based on three color features and four texture features is proposed in this paper. The experiment results show that the multifeatures model can improve the retrieval results than other model just by each singular feature.

Wind turbine towers affect the visual appearance of the landscape, as an example in the touristic woodland of Dalecarlia, and the fear is that the visual impact will be too negative to the important tourist trade. The landscape analysis, developed by municipalities around Lake Siljan, limited expansion of wind power, due to the strong visual impression of wind turbine towers. In order to facilitate the assessment of the visual impact of towers a view, from Tällberg, over the ring of height on the other side of Lake Siljan, has been photographed every ten minutes for a year (34,727 images, about 65% of the possible number during a year). Four towers are possible to see in the photos, three of them have been used in the assessment of visual impression. This contribution presents a method to assess visibility of wind turbine towers from photographs, describing the measuring situation (location and equipment) as well as the analytical method and results of the analysis. The towers are possible to see in about 48% of analyzed images taken during daytime with the used equipment. During the summer (winter) months the towers were apparent in 49% (46%) of the images. At least one red warning light was possible to see on towers in about 66% of the night images. One conclusion of this work is that the method to assess the visibility within digital photographs and translate it into the equivalent of a normal eye can only provide an upper limit for visibility of an object.

A non-uniformity correction algorithm is proposed and implemented on a Field-Programmable Gate Array (FPGA) hardware platform to solve a pixel response non-uniformity(PRNU) problem of multi Time Delay and Integration Charge Couple Device(TDICCD) camera. The non-uniformity are introduced and the synthetical correction algorithm is presented, in which the two-point correction method is used in a single channel, gain averaging correction method among multi-channel and the sceneadaptive correction method among multi-TDICCD. Then, the correction algorithm is designed. Finally, analyzing the FPGA ability for fix-point processing, the correction algorithm is optimized, and implemented on FPGA. Testing results indicate that the non-uniformity can be decreased from 8.27% to 0.51% for three TDICCDs camera's images with the proposed correction algorithm, proving that this correction algorithm is with high real-time performance, great engineering realization and satisfaction for the system requirements.