Mattia Marconcini

A Novel Transductive SVM for Semisupervised Classification of Remote-Sensing Images

This paper introduces a semisupervised classification method that exploits both labeled and unlabeled samples for addressing ill-posed problems with support vector machines (SVMs). The method is based on recent developments in statistical learning theory concerning transductive inference and in particular transductive SVMs (TSVMs). TSVMs exploit specific iterative algorithms which gradually search a reliable separating hyperplane (in the kernel space) with a transductive process that incorporates both labeled and unlabeled samples in the training phase. Based on an analysis of the properties of the TSVMs presented in the literature, a novel modified TSVM classifier designed for addressing ill-posed remote-sensing problems is proposed. In particular, the proposed technique: 1) is based on a novel transductive procedure that exploits a weighting strategy for unlabeled patterns, based on a time-dependent criterion; 2) is able to mitigate the effects of suboptimal model selection (which is unavoidable in the presence of small-size training sets); and 3) can address multiclass cases. Experimental results confirm the effectiveness of the proposed method on a set of ill-posed remote-sensing classification problems representing different operative conditions

Domain Adaptation Problems: A DASVM Classification Technique and a Circular Validation Strategy

This paper addresses pattern classification in the framework of domain adaptation by considering methods that solve problems in which training data are assumed to be available only for a source domain different (even if related) from the target domain of (unlabeled) test data. Two main novel contributions are proposed: 1) a domain adaptation support vector machine (DASVM) technique which extends the formulation of support vector machines (SVMs) to the domain adaptation framework and 2) a circular indirect accuracy assessment strategy for validating the learning of domain adaptation classifiers when no true labels for the target--domain instances are available. Experimental results, obtained on a series of two-dimensional toy problems and on two real data sets related to brain computer interface and remote sensing applications, confirmed the effectiveness and the reliability of both the DASVM technique and the proposed circular validation strategy.

An advanced system for the automatic classification of multitemporal SAR images

A novel system for the classification of multitemporal synthetic aperture radar (SAR) images is presented. It has been developed by integrating an analysis of the multitemporal SAR signal physics with a pattern recognition approach. The system is made up of a feature-extraction module and a neural-network classifier, as well as a set of standard preprocessing procedures. The feature-extraction module derives a set of features from a series of multitemporal SAR images. These features are based on the concepts of long-term coherence and backscattering temporal variability and have been defined according to an analysis of the multitemporal SAR signal behavior in the presence of different land-cover classes. The neural-network classifier (which is based on a radial basis function neural architecture) properly exploits the multitemporal features for producing accurate land-cover maps. Thanks to the effectiveness of the extracted features, the number of measures that can be provided as input to the classifier is significantly smaller than the number of available multitemporal images. This reduces the complexity of the neural architecture (and consequently increases the generalization capabilities of the classifier) and relaxes the requirements relating to the number of training patterns to be used for classifier learning. Experimental results (obtained on a multitemporal series of European Remote Sensing 1 satellite SAR images) confirm the effectiveness of the proposed system, which exhibits both high classification accuracy and good stability versus parameter settings. These results also point out that properly integrating a pattern recognition procedure (based on machine learning) with an accurate feature extraction phase (based on the SAR sensor physics understanding) represents an effective approach to SAR data analysis.

A Novel Approach to Unsupervised Change Detection Based on a Semisupervised SVM and a Similarity Measure

This paper presents a novel approach to unsupervised change detection in multispectral remote-sensing images. The proposed approach aims at extracting the change information by jointly analyzing the spectral channels of multitemporal images in the original feature space without any training data. This is accomplished by using a selective Bayesian thresholding for deriving a pseudotraining set that is necessary for initializing an adequately defined binary semisupervised support vector machine classifier. Starting from these initial seeds, the performs change detection in the original multitemporal feature space by gradually considering unlabeled patterns in the definition of the decision boundary between changed and unchanged pixels according to a semisupervised learning algorithm. This algorithm models the full complexity of the change-detection problem, which is only partially represented from the seed pixels included in the pseudotraining set. The values of the classifier parameters are then defined according to a novel unsupervised model-selection technique based on a similarity measure between change-detection maps obtained with different settings. Experimental results obtained on different multispectral remote-sensing images confirm the effectiveness of the proposed approach.

Urban Footprint Processor—Fully Automated Processing Chain Generating Settlement Masks From Global Data of the TanDEM-X Mission

The German TerraSAR-X add-on for Digital Elevation Measurement (TanDEM-X) mission (TDM) collects two global data sets of very high resolution (VHR) synthetic aperture radar (SAR) images between 2011 and 2013. Such imagery provides a unique information source for the identification of built-up areas in a so far unique spatial detail. This letter presents the novel implementation of a fully automated processing system for the delineation of human settlements worldwide based on the SAR data acquired in the context of the TDM. The proposed Urban Footprint Processor (UFP) includes three main processing stages dedicated to: i) the extraction of texture information suitable for highlighting regions characterized by highly structured and heterogeneous built-up areas; ii) the generation of a binary settlement layer (built-up, non-built-up) based on an unsupervised classification scheme accounting for both the original backscattering amplitude and the extracted texture; and iii) a final post-editing and mosaicking phase aimed at providing the final Urban Footprint (UF) product for arbitrary geographical regions. Experimental results assess the high potential of the TDM data and the proposed UFP to provide highly accurate geo-data for an improved global mapping of human settlements.

Toward the Automatic Updating of Land-Cover Maps by a Domain-Adaptation SVM Classifier and a Circular Validation Strategy

In this paper, we address automatic updating of land-cover maps by using remote-sensing images periodically acquired over the same investigated area under the hypothesis that a reliable ground truth is not available for all the considered acquisitions. The problem is modeled in the domain-adaptation framework by introducing a novel method designed for land-cover map updating, which is based on a domain-adaptation support vector machine technique. In addition, a novel circular accuracy assessment strategy is proposed for the validation of the results obtained by domain-adaptation classifiers when no ground-truth labels for the considered image are available. Experimental results obtained on a multitemporal and multispectral data set confirmed the effectiveness and the reliability of the proposed system.

A Composite Semisupervised SVM for Classification of Hyperspectral Images

This letter presents a novel composite semisupervised support vector machine (SVM) for the spectral-spatial classification of hyperspectral images. In particular, the proposed technique exploits the following: 1) unlabeled data for increasing the reliability of the training phase when few training samples are available and 2) composite kernel functions for simultaneously taking into account spectral and spatial information included in the considered image. Experiments carried out on a hyperspectral image pointed out the effectiveness of the presented technique, which resulted in a significant increase of the classification accuracy with respect to both supervised SVMs and progressive semisupervised SVMs with single kernels, as well as supervised SVMs with composite kernels.

Combined use of multi-seasonal high and medium resolution satellite imagery for parcel-related mapping of cropland and grassland

Abstract A key factor in the implementation of productive and sustainable cultivation procedures is the frequent and area-wide monitoring of cropland and grassland. In particular, attention is focused on assessing the actual status, identifying basic trends and mitigating major threats with respect to land-use intensity and its changes in agricultural and semi-natural areas. Here, multi-seasonal analyses based on satellite Earth Observation (EO) data can provide area-wide, spatially detailed and up-to-date geo-information on the distribution and intensity of land use in agricultural and grassland areas. This study introduces an operational, application-oriented approach towards the categorization of agricultural cropland and grassland based on a novel scheme combining multi-resolution EO data with ancillary geo-information available from currently existing databases. In this context, multi-seasonal high (HR) and medium resolution (MR) satellite imagery is used for both a land parcel-based determination of crop types as well as a cropland and grassland differentiation, respectively. In our experimental analysis, two HR IRS-P6 LISS-3 images are first employed to delineate the field parcels in potential agricultural and grassland areas (determined according to the German Official Topographic Cartographic Information System – ATKIS). Next, a stack of seasonality indices is generated based on 5 image acquisitions (i.e., the two LISS scenes and three additional IRS-P6 AWiFS scenes). Finally, a C5.0 tree classifier is applied to identify main crop types and grassland based on the input imagery and the derived seasonality indices. The classifier is trained using sample points provided by the European Land Use/Cover Area Frame Survey (LUCAS). Experimental results for a test area in Germany assess the effectiveness of the proposed approach and demonstrate that a multi-scale and multi-temporal analysis of satellite data can provide spatially detailed and thematically accurate geo-information on crop types and the cropland-grassland distribution, respectively.

A Novel Context-Sensitive SVM for Classification of Remote Sensing Images

In this paper, a novel context-sensitive classification technique based on Support Vector Machines (CS-SVM) is proposed. This technique aims at exploiting the promising SVM method for classification of 2-D (or n-D) scenes by considering the spatial-context information of the pixel to be analyzed. In greater detail, the proposed architecture properly exploits the spatial-context information for: i) increasing the robustness of the learning procedure of SVMs to the noise present in the training set (mislabeled training samples); ii) regularizing the classification maps. The first property is achieved by introducing a context-sensitive term in the objective function to be minimized for defining the decision hyperplane in the SVM kernel space. The second property is obtained including in the classification procedure of a generic pattern the information of neighboring pixels. Experiments carried out on very high geometrical resolution images confirm the validity of the proposed technique.

Targeted Land-Cover Classification

This paper addresses a specific typology of land-cover classification problems, hereinafter referred to as “targeted land-cover classification,” where the objective is the identification of only one or few specific “targeted” land-cover classes of interest, disregarding all the other potential classes present in the area under analysis. Such a challenging problem, which is common to a variety of operational information services and applications (e.g., agriculture, forestry, spatial planning, ecosystem monitoring, disaster management, habitat mapping, etc.), can be effectively solved by traditional supervised classification techniques provided that an exhaustive ground truth is available for all the land-cover classes present in the region of interest. Such a requirement is seldom satisfied and presents several practical drawbacks and limitations, both in terms of time and economic cost that may render this task difficult to achieve in most real-life cases. However, the possibility to perform an effective targeted classification using only ground-truth samples for the class(es) of interest (hence avoiding the burden and cost associated with the collection of a full and exhaustive ground-truth information) would represent a significant advantage. In this paper, we present a novel technique capable of identifying specific land-cover classes of interest by exploiting the ground truth only available for these targeted classes, while providing accuracies comparable to those of traditional fully supervised methods. The proposed technique jointly exploits both the unlabeled samples of the image under investigation and the training samples only available for the targeted classes. In particular, the expectation-maximization algorithm and Markov random fields are employed to estimate the probability density function of both the class(es) of interest and the unknown class representing the merger of all the unknown land-cover classes characterizing the study area for which no ground-truth information is available. An extensive experimental analysis and cross-comparisons with both fully supervised support vector machines and ensembles of multiple one-class support-vector-data-description classifiers on different data sets confirmed the effectiveness and the reliability of the proposed technique.