Search

In the last few years active learning has been gaining growing interest in the remote sensing community in optimizing the process of training sample collection for supervised image classification. Current strategies formulate the active learning problem in the spectral domain only. However remote sensing images are intrinsically defined both in the spectral and spatial domains. In this paper we explore this fact by proposing a new active learning approach for support vector machine classification. In particular we suggest combining spectral and spatial information directly in the iterative process of sample selection. For this purpose three criteria are proposed to favor the selection of samples distant from the samples already composing the current training set. In the first strategy the Euclidean distances in the spatial domain from the training samples are explicitly computed whereas the second one is based on the Parzen window method in the spatial domain. Finally the last criterion involves the concept of spatial entropy. Experiments on two very high resolution images show the effectiveness of regularization in spatial domain for active learning purposes. © 2013 IEEE.

The high-performance camera control systems carried aboard the DigitalGlobe WorldView satellites, WorldView-1 and WorldView-2, are capable of rapid retargeting and high off-nadir imagery collection. This provides the capability to collect dozens of multiangle very high spatial resolution images over a large target area during a single overflight. In addition, WorldView-2 collects eight bands of multispectral data. This paper discusses the improvements in urban classification accuracy available through utilization of the spatial and spectral information from a WorldView-2 multiangle image sequence collected over Atlanta, GA, in December 2009. Specifically, the implications of adding height data and multiangle multispectral reflectance, both derived from the multiangle sequence, to the textural, morphological, and spectral information of a single WorldView-2 image are investigated. The results show an improvement in classification accuracy of 27% and 14% for the spatial and spectral experiments, respectively. Additionally, the multiangle data set allows the differentiation of classes not typically well identified by a single image, such as skyscrapers and bridges as well as flat and pitched roofs.

We investigate the relevance of morphological operators for the classification of land use in urban scenes using sub-metric panchromatic imagery. A support vector machine is used for the classification. Six types of filters have been employed: opening and closing, opening and closing by reconstruction, and opening and closing top hat. The type and scale of the filters are discussed, and a feature selection algorithm called recursive feature elimination is applied to decrease the dimensionality of the input data. The analysis performed on two QuickBird panchromatic images showed that simple opening and closing operators are the most relevant for classification at such a high spatial resolution. Moreover, mixed sets combining simple and reconstruction filters provided the best performance. Tests performed on both images, having areas characterized by different architectural styles, yielded similar results for both feature selection and classification accuracy, suggesting the generalization of the feature sets highlighted.

In this paper, we propose two active learning algorithms for semiautomatic definition of training samples in remote sensing image classification. Based on predefined heuristics, the classifier ranks the unlabeled pixels and automatically chooses those that are considered the most valuable for its improvement. Once the pixels have been selected, the analyst labels them manually and the process is iterated. Starting with a small and nonoptimal training set, the model itself builds the optimal set of samples which minimizes the classification error. We have applied the proposed algorithms to a variety of remote sensing data, including very high resolution and hyperspectral images, using support vector machines. Experimental results confirm the consistency of the methods. The required number of training samples can be reduced to 10% using the methods proposed, reaching the same level of accuracy as larger data sets. A comparison with a state-of-the-art active learning method, margin sampling, is provided, highlighting advantages of the methods proposed. The effect of spatial resolution and separability of the classes on the quality of the selection of pixels is also discussed.

International audience; The 2007 Data Fusion Contest that was organized by the IEEE Geoscience and Remote Sensing Data Fusion Technical Committee was dealing with the extraction of a land use/land cover maps in and around an urban area, exploiting multitemporal and multisource coarse-resolution data sets. In particular, synthetic aperture radar and optical data from satellite sensors were considered. Excellent indicators for mapping accuracy were obtained by the top teams. The best algorithm is based on a neural classification enhanced by preprocessing and postprocessing steps.

The advent of new high spatial resolution optical satellite imagery has greatly increased our ability to monitor land cover changes from space. Satellite observations are carried out regularly and continuously, and provide a great deal of insight into the temporal changes of land cover use. High spatial resolution imagery better resolves the details of these changes and makes it possible to overcome the "mixed-pixel" problem that is inherent with more moderate resolution satellite sensors. At the same time, high-resolution imagery presents a new challenge over other satellite systems, in that a relatively large amount of data must be analyzed and corrected for registration and classification errors to identify the land cover changes. To obtain the accuracies that are required by many applications to large areas, very extensive manual work is commonly required to remove the classification errors that are introduced by most methods. To improve on this situation, we have developed a new method for land surface change detection that greatly reduces the human effort that is needed to remove the errors that occur with many classification methods that are applied to high-resolution imagery. This change detection algorithm is based on neural networks, and it is able to exploit in parallel both the multiband and the multitemporal data to discriminate between real changes and false alarms. In general, the classification errors are reduced by a factor of 2-3 using our new method over a simple postclassification comparison based on a neural-network classification of the same images.

In this study we propose an evaluation of the angular effects altering the spectral response of the land-cover over multi-angle remote sensing image acquisitions. The shift in the statistical distribution of the pixels observed in an in-track sequence of WorldView-2 images is analyzed by means of a kernel-based measure of distance between probability distributions. Afterwards, the portability of supervised classifiers across the sequence is investigated by looking at the evolution of the classification accuracy with respect to the changing observation angle. In this context, the efficiency of various physically and statistically based preprocessing methods in obtaining angle-invariant data spaces is compared and possible synergies are discussed.

In this paper, we present a method to analyze the impact of data space normalization on spectral classification model portability using multi-angle very-high spatial resolution imagery. In-track multi-angle data provide images of a single scene, from different observation angles, during a very short period of time. This creates a sequence of images with relatively static atmospheric and illumination conditions. With this data, the only changes in the scene are due to observation angle and surface reflectance properties. Using this information, we present an analysis of both the impact of surface anisotropy and data space normalization on spectral classification accuracy and model portability.

Active learning process represents an interesting solution to the problem of training sample collection for the classification of remote sensing images. In this work, we propose a criterion based on the spatial information that can be used in combination with a spectral criterion in order to improve the selection of training samples. Experimental results obtained on a very high resolution image show the effectiveness of regularization in spatial domain and open challenging perspectives for terrain campaigns planning. © 2011 IEEE.

We apply a comparative data mining framework to the multispectral classification of WorldView-2 (WV2) imagery. Our goal is two-fold. First, we want to identify land covers for which the combination of extended spectral coverage and high spatial resolution provide a distinctive advantage in classification accuracy. Second, we perform predictor analyses to determine which combinations of bands are more effective in resolving individual targets. This experimental approach provides a basis for building a spectral atlas that can offer guidance on the optimal combination of WV2 spectral bands for different application areas.

International audience; The 2008 Data Fusion Contest organized by the IEEE Geoscience and Remote Sensing Data Fusion Technical Committee deals with the classification of high-resolution hyperspectral data from an urban area. Unlike in the previous issues of the contest, the goal was not only to identify the best algorithm but also to provide a collaborative effort: The decision fusion of the best individual algorithms was aiming at further improving the classification performances, and the best algorithms were ranked according to their relative contribution to the decision fusion. This paper presents the five awarded algorithms and the conclusions of the contest, stressing the importance of decision fusion, dimension reduction, and supervised classification methods, such as neural networks and support vector machines.

An extensive analysis based on the use of different morphological filters for the classification of very-high resolution panchromatic images is presented. Feature selection on high-dimensional input space is performed using recursive feature elimination, a support vector machines specific method performing backward elimination based on margin-estimation criterion. Experimental results on an eight-classes image of Las Vegas (USA) confirmed the effectiveness of the analysis pointing out the relevancy of the most contributing morphological features which resulted in high classification accuracy using panchromatic imagery.

An active learning method is proposed for the semi-automatic selection of training sets in remote sensing image classification. The method adds iteratively to the current training set the unlabeled pixels for which the prediction of an ensemble of classifiers based on bagged training sets show maximum entropy. This way, the algorithm selects the pixels that are the most uncertain and that will improve the model if added in the training set. The user is asked to label such pixels at each iteration. Experiments using support vector machines (SVM) on an 8 classes QuickBird image show the excellent performances of the methods, that equals accuracies of both a model trained with ten times more pixels and a model whose training set has been built using a state-of-the-art SVM specific active learning method.

Parametric and nonparametric approaches to evaluate land-cover change detection using very high resolution (VHR) satellite imagery are applied to the analysis of the demolition of the Rocky Flats nuclear weapons facility located near Denver, CO. Both maximum-likelihood and neural network classifiers are used to validate a new parallel architecture which improves the accuracy when applied to VHR satellite imagery for the study of land-cover change between sequential satellite acquisitions. An enhancement of about 14% was found between the single-step classification and the new parallel architecture, confirming the advantage and the robust improvement obtained with this architecture regardless of the classification algorithm used. In this paper, we demonstrate and document the demolition and removal of hundreds of buildings taken down to bare soil between 2003 and 2005 at the Rocky Flats site.

This paper studies the impact of several learning issues in an image classification task with SVMs, such as rich feature-based representations, optimization and sensitivity to novelty in the test data sets. The employed imagery refers to the city of Rome, Italy and is acquired in different years and seasons by the European Remote Sensing Satellites ERS-1 and ERS-1/2 tandem mission. A comprehensive evaluation according to varying training conditions is reported, showing that SVMs provide robust and largely applicable tools.

This paper reports on monitoring land cover in the urban and sub-urban area of Rome, Italy, by multi-temporal ERS 1-2 SLC SAR images. The identification of the SAR image parameters, including backscattering, degree of interferometric coherence and textural pixel-based features to be exploited in classification, is discussed. The information extracted from the SAR images is fused and processed by a Multi-Layer Perceptron (MLP) neural network (NN) to produce land cover maps. The network topology has been carefully designed, paying special attention to the number of hidden units. Once trained, the net's performance has been validated over a statistically significant ensemble of patterns independent of the learning set. The net has been used for the automatic classification of a large area in two different years, 1994 and 1999, thus obtaining a map of hot spots corresponding to areas where changes had occurred.

A comparative study on the complexity of the urban environments in SAR imagery at different spatial resolutions is presented. Two datasets have been considered, including the city of Rome, Italy imaged in decametric resolution and the Frascati area (Rome, Italy) acquired in very high spatial resolution (~2 m) at L-band in a fully polarimetric mode. The different characteristics of the radar sensors require careful managing of the corresponding product capabilities to maximize the various pieces of information contained in the variety of scattering mechanisms.

We present two different change detection techniques to monitor surface changes that occurred at the Rocky Flats nuclear weapons facility located immediately to the North West of the city of Denver, Colorado, USA. The site started being cleaned up and dismantled in 1998 and was completed in 2005. The first Change Detection method is based on a Maximum Likelihood classifier, while the other is an approach based on a Neural Network architecture called NAHIRI (Neural Architecture for High-Resolution Imagery) to produce change detection maps from very high-resolution satellite imagery. NAHIRI simultaneously exploits spectral and temporal information by adding a filter, directly stemming from the multi- temporal information, to the classification changes derived from the multi-spectral data. In fact, the distinctive feature of this method is that the NNs exploit both the multi-spectral and the multi-temporal information in parallel that are associated with the changed values of the pixel spectral reflectances. The quantitative results are analyzed in order to single out advantages and shortcomings of the two different approaches.

The advent of very high spatial resolution optical satellite imagery has greatly increased our ability to monitor land cover changes in urban environments where the spatial resolution plays a key role related to the detection of fine-scale objects such as a single house or small structures. At the same time, very high spatial resolution imagery presents a new challenge over other satellite systems, in that a relatively large amount of data must be analyzed and corrected for registration and classification errors to identify the land cover changes, commonly resulting in a very extensive manual work. To improve on this situation we have developed a new method for land surface change detection that greatly reduces the human effort needed to remove the errors that occur with many methods applied to very high spatial resolution imagery. This change detection algorithm is based on Neural Networks and it is able to exploit in parallel both the multi-band and the multi-temporal data to discriminate between real changes and false alarms. In general the classification errors are reduced by a factor of 2-3 using this new method over a simple Post Classification Comparison based on a neural network classification of the same images.

The effectiveness of multilayer perceptron (MLP) networks as a tool for the classification of remotely sensed images has been already proven in past years. However, most of the studies consider images characterized by high spatial resolution (around 15-30 m) while a detailed analysis of the performance of this type of classifier on very high resolution images (around 1-2 m) such as those provided by the Quickbird satellite is still lacking. Moreover, the classification problem is normally understood as the classification of a single image while the capabilities of a single network of performing automatic classification and feature extraction over a collection of archived images has not been explored so far. In this paper, besides assessing the performance of MLP for the classification of very high resolution images, we investigate on the generalization capabilities of this type of algorithms with the purpose of using them as a tool for fully automatic classification of collections of satellite images, either at very high or at high-resolution. In particular, applications to urban area monitoring have been addressed

In the above titled paper (ibid., vol. 47, no. 7, pp. 2218-2232, Jul. 09), three lines from Algorithm 2 were inadvertently omitted during the paper's typesetting. The corrected algorithm is presented here.