Description: We propose a procedure that efficiently adapts a classifier trained on a source image to a target image with similar spectral properties. The adaptation is carried out by adding new relevant training samples with active queries in the target domain following a strategy specifically designed for the case where class distributions have shifted between the two acquisitions. In fact, the procedure consists of two nested algorithms. An active selection of the pixels to be labeled is performed on a set of candidates of the target image in order to select the most informative pixels. Along the inclusion of the pixels to the training set, the weights associated with these samples are iteratively updated using different criteria, depending on their origin (source or target image). We study this adaptation framework in combination with a SVM classifier accepting instance weights. Experiments on two VHR QuickBird images and on a hyperspectral AVIRIS image prove the validity of the proposed adaptive approach with respect to existing techniques not involving any adjustments to the target domain.

Description: The accuracy of a conventional supervised classification is in part a function of the training set used, notably impacted by the quantity and quality of the training cases. Since it can be costly to acquire a large number of high quality training cases, recent research has focused on methods that allow accurate classification from small training sets. Previous work has shown the potential of support vector machine (SVM) based classifiers. Here, the potential of the relevance vector machine (RVM) and sparse multinominal logistic regression (SMLR) approaches is evaluated relative to SVM classification. With both airborne and spaceborne multispectral data sets, the RVM and SMLR were able to derive classifications of similar accuracy to the SVM but required considerably fewer training cases. For example, from a training set comprising 600 cases acquired with a conventional stratified random sampling design from an airborne thematic mapper (ATM) data set, the RVM produced the most accurate classification, 93.75%, and needed only 7.33% of the available training cases. In comparison, the SVM yielded a classification that had an accuracy of 92.50% and needed 4.5 times more useful training cases. Similarly, with a Landsat ETM+ (Littleport, Cambridgeshire, UK) data set, the SVM required 4.0 times more useful training cases than the RVM. For each data set, however, the classifications derived by each classifier were of similar magnitude, differing by no more than 1.25%. Finally, for both the ATM and ETM+ (Littleport) data sets, the useful training cases by SVM and RVM had distinct and potentially predictable characteristics. Support vectors were generally atypical but lay in the boundary region between classes in feature space while the relevance vectors were atypical but anti-boundary in nature. The SMLR also tended to mostly, but not always, use extreme cases that lay away from class boundary. The results, therefore, suggest a potential to design classifier-specific intelli- ent training data acquisition activities for accurate classification from small training sets, especially with the SVM and RVM.

Description: Remote sensing plays a major role in supporting decision-making and surveying compliance of several multilateral environmental treaties. In this paper, we present an approach for supporting monitoring compliance of river networks in context of the European Water Framework Directive. Only a few approaches have been developed for extracting river networks from satellite data and usually they require manual input, which seems not feasible for automatic and operational application. We propose a method for the automatic extraction of river structures in TerraSAR-X data. The method is based on mathematical morphology and supervised image classification, using automatically selected training samples. The method is applied on TerraSAR-X images from two different study sites. In addition, the results are compared to an alternative method, which requires manual user interaction. The detailed accuracy assessment shows that the proposed method achieves accurate results (Kappa $ {sim}$ 0.7) and performs almost similar in terms of accuracy, when compared to the alternative approach. Moreover, the proposed method can be applied on various datasets (e.g., multitemporal, multisensoral and multipolarized) and does not require any additional user input. Thus, the highly flexible approach is interesting in terms of operational monitoring systems and large scale applications.

Description: In this paper, a road network grouping algorithm for Synthetic Aperture Radar (SAR) images is proposed by exploiting multiscale geometric analysis of detector responses. Before running the algorithm, a response map made up of responses, which is binarized, skeletonized, and vectorized to generate road candidates, is obtained by applying a local detector to a SAR image first. Then the proposed method identifies real road segments among the candidates and fills gaps between them. It works in three steps. 1) Guidance segments are extracted at different resolutions from the response map using multiscale techniques and merged to get a more appropriate approximation. 2) Segments are labeled “road” or “noise” using relaxation labeling techniques, among which “road” ones are grouped as they may lie on different roads. 3) Connection points between candidates are acquired by mapping candidates to grouped “road” guidance segments. Those connection points are linked with straight lines or curvilinear segments after a segmentation process. The experiments on TerraSAR images show the effectiveness of this new method.

Description: This paper presents a new approach to sea ice segmentation in synthetic aperture radar (SAR) intensity images by combining an edge-preserving region (EPR)-based representation with region-level MRF models. To construct the EPR-based representation of a SAR image, edge strength is measured using instantaneous coefficient of variation (ICOV) upon which the watershed algorithm is applied to partition the image into primitive regions. In addition, two new metrics for quantitative assessment of region characteristics (region accuracy and region redundancy) are defined and used for parameter estimation in the ICOV extraction process towards desired region characteristics. In combination with a region-level MRF, the EPR-based representation facilitates the segmentation process by largely reducing the search space of optimization process and improving parameter estimation of feature model, leading to considerable computational savings and less probability of false segmentation. The proposed segmentation method has been evaluated using a synthetic sea ice image corrupted with varying levels of speckle noise as well as real SAR sea ice images. Relative to the existing region-level MRF-based methods, testing results have demonstrated that our proposed method substantially improves the segmentation accuracy at high speckle noise and achieves on average 29% reduction of computational time.

Description: This paper presents a novel method for segmenting the oil spill regions in the SAR satellite images taken in broad daylight using illumination-reflectance based level set model. These images of oil spills taken in broad daylight appear as a blend of dark areas with scintillations of glitter due to the illumination and reflectance components present. Most of the dark areas in the SAR images are the areas indicating oil spills because the oil dampens the capillary waves on the sea surface. The presence of the glitter induces speckle in SAR images. This does not only reduces the interpreter's ability to resolve fine detail, but also makes automatic segmentation of such images difficult. Segmentation of such images using conventional level set methods makes the process cumbersome and may lead to improper results. The accuracy of segmentation greatly depends on the amount of the illumination and reflectance (IR) components present in the images. To perform segmentation of such images we propose an adaptive level set evolution process based on the IR components in them. This can be achieved by combining a new signed pressure function which is derived from the amount illumination and reflectance present in the image. The IR components present in image are extracted by the process of homomorphic decomposition with the help of filters with specific cut off frequencies. This method is the first application successfully implemented on SAR images and the results are found to be superior when compared with earlier techniques. Comparative analysis is made with the conventional region based level sets in terms of accuracy of segmentation for complex images.

Description: The mixed pixel problem may be reduced through the use of a soft image classification and super-resolution mapping analyses. Here, the positive attributes of two popular super-resolution mapping methods, based on contouring and the Hopfield neural network, are combined. For a binary classification scenario, the method is based on fitting a contour of equal class membership to a pre-final output of a standard Hopfield neural network. Analyses of simulated and real image data sets show that the proposed method is more accurate than the standard contouring and Hopfield neural network based methods, with error typically reduced by a factor of two or more. The sensitivity of the Hopfield neural network based approaches to the setting of a gain function is also explored.

Description: In this paper, the link-based cluster ensemble (LCE) method is utilized to improve cloud classification and satellite precipitation estimation. High resolution Satellite Precipitation Estimation (SPE) is based on the Precipitation Estimation from Remotely Sensed Imagery using an Artificial Neural Network Cloud Classification (PERSIANN-CCS) algorithm. This modified SPE with the incorporation of LCE involves the following four steps: 1) segmentation of infrared cloud images into patches; 2) cloud patch feature extraction; 3) clustering cloud patches using LCE; and 4) dynamic application of brightness temperature (Tb) and rain-rate relationships, derived using satellite observations. In order to cluster the cloud patches, the LCE method combines multiple data partitions from different clustering methods. The results show that using the cluster ensemble increases the performance of rainfall estimates compared to the SPE algorithm using a Self Organizing Map (SOM) neural network. The false alarm ratio (FAR), probabilities of detection (POD), equitable threat score (ETS), and bias are used as quantitative measures to assess the performance of the algorithm. It is shown that both the ETS and bias provide improvement in the summer and winter seasons. Almost 5% ETS improvement is obtained at some threshold values for the winter season using the cluster ensemble.

Description: Combining super-resolution techniques can increase the accuracy with which the shape of objects may be characterised from imagery. This is illustrated with two approaches to combining the contouring and pixel swapping methods of super-resolution mapping for binary classification applications. In both approaches, the output of the pixel swapping method is softened to allow a contour of equal class membership to be fitted to it to represent the inter-class boundary. The accuracy of super-resolution mapping with the individual and combined techniques is explored, including an assessment of the effect of variation in the number of neighbors and zoom factor on pixel swapping based analyses. When combined, the error with which objects of varying shape were represented was typically greatly reduced relative to that observed from the application of the methods individually. For example, the root mean square error in mapping the boundary of an aeroplane represented in relatively fine spatial resolution imagery decreased from 14.41 m with contouring and 4.35 m with pixel swapping to 3.07 m when the approaches were combined.