Outcome of the “Best Paper Challenge” of the 2013 IEEE GRSS Data Fusion Contest

Approximately 30 teams worldwide entered the “Best Paper Challenge” of the 2013 IEEE GRSS Data Fusion Contest with innovative approaches to analyze hyperspectral and LiDAR data.

The top-ten results, ranked based on the cumulative scores from a double blind peer-review process are summarized below along with a brief description of the algorithms and the contact information of the authors.


1’st Place

Title: GRAPH-BASED FEATURE FUSION OF HYPERSPECTRAL AND LIDAR REMOTE SENSING DATA USING MORPHOLOGICAL FEATURES

Authors: Wenzhi Liao, Rik Bellens, Aleksandra Pizurica, Sidharta Gautama, Wilfried Philips

Affiliations: Ghent University, Belgium

Contact Email: wliao@telin.ugent.be

Abstract: Automatic interpretation of remote sensed images remains challenging. Nowadays, we have diverse sensor technologies and image processing algorithms that allow to measure different aspects of objects on the earth (spectral characteristics in hyperspectral images, height in LiDAR data, geometry in image processing technologies like Morphological Profiles). It is clear that no single technology can be sufficient for a reliable classification, but combining many of them can lead to problems like the curse of dimensionality, excessive computation time and so on. Applying feature reduction techniques on all the features together is not a good idea, because this does not take into account the differences in structure of the feature spaces. Decision fusion on the other hand has difficulties with modeling correlations between the different data sources. In this paper, we propose a graph-based fusion method, where the high-dimensional features is projected on a lower dimen- sional feature space. This projection is learned based on a fused graph that takes into account the differences between the data sources. Experimental results on the classification of fusing real HS image and LiDAR data demonstrate effectiveness of the proposed method both visually and quantitatively.


2’nd Place

Title: A TWO-STREAM FUSION FRAMEWORK FOR LIDAR AND HYPERSPECTRAL IMAGERY

Authors: Christian Debes†, Andreas Merentitis†, Roel Heremans†, Jurgen Hahn‡, Nikolaos Frangiadakis† and Tim van Kasteren†

Affiliations: †AGT International, Darmstadt, Germany; ‡Signal Processing Group, Institute of Telecommunications, Germany.

Contact Email: cdebes@agtinternational.com

Abstract: Data fusion can significantly increase accuracy of automated classification in remote sensing applications by combining data from different types of sensors. Particularly for hyperspectral imagery (HSI), complementing the hyperspectral data with topographical information in the form of a Digital Surface Model (DSM) generated by LiDAR data is promising to address problems with artifacts or distortion in difficult areas. In this paper we introduce a novel framework for combining the HSI and LiDAR data, which enables handling identified objects as uniform entities rather than as independent pixels. Further contributions include an initial spectral unmixing step that segregates noise and significantly improves the benefit of adding LiDAR, as well as the application of ensemble learning in the form of Random Forest algorithms that inherently support feature selection.


3’d Place

Title: OPERATOR ANALYSIS AND DIFFUSION BASED EMBEDDINGS FOR HETEROGENEOUS DATA FUSION

Authors: Alexander Cloninger, Wojciech Czaja, Timothy Doster

Affiliations: Norbert Wiener Center, Department of Mathematics, University of Maryland, College Park, USA

Contact Email: tdoster@math.umd.edu

Abstract: As new sensing modalities emerge and the presence of multiple sensors per platform becomes widespread, it is vital to develop new algorithms and techniques which can fuse this data. Previous attempts to deal with the problem of heterogeneous data integration for the applications in data classification were either highly data dependent or relied on simply fusing classifier outputs. In this paper we examine several related approaches: graph fusion, operator fusion, and feature space fusion. They are all associated with graph diffusion processes generated by appropriately designed operators. Our results do not make any assumptions about the data and can be eas- ily extended to new additional modalities.


4’th Place

Title: FUSION OF HYPERSPECTRAL AND LIDAR DATA FOR LANDSCAPE VISUAL QUALITY ASSESSMENT

Authors: Tomohiro Matsuki, Shinji Nakazawa, Naoto Yokoya, and Akira Iwasaki

Affiliations: University of Tokyo

Contact Email: nao.1021.yky@gmail.com

Abstract: Visual quality assessment of landscape is an important factor for habitants to select rooms to live in. In this work, the authors present a fusion of hyperspectral and LiDAR data for landscape visual quality assessment. From the fused hyperspectral and LiDAR data, classification and depth images at any location could be obtained, enabling physical features such as material properties and openness to be quantified. The relationship between physical features and landscape preferences is learned using the lasso regression. The proposed method is applied to the hyperspectral and LiDAR datasets provided in the 2013 IEEE GRSS Data Fusion Contest. The results showed that the proposed method successfully learned the model that enables the prediction of scenic quality at any viewpoint. This work contributes to automatic landscape assessment and optimal spatial planning using remote sensing data.


5’th Place

Title:  FUSION OF AIRBORNE HYPERSPECTRAL AND LIDAR DATA TO MODEL THE THERMAL CHARACTERISTICS OF URBAN ENVIRONMENTS

Authors: C. Berger1, F. Riedel1, J. Rosentreter1, E. Stein2, S. Hese1, C. Schmullius1

Affiliations: 1: Department for Earth Observation, University of Jena, Jena, Germany; 2: Earth Observation Center, German Aerospace Center (DLR), Wessling, Germany

Contact Email: christian.berger@uni-jena.de

Abstract: This study focuses on the derivation of an urban surface material map to parameterize a 3D numerical microclimate model. For this purpose, fusion of airborne hyperspectral and light detection and ranging (LiDAR) data is performed. In a first step, surface materials are extracted from the preprocessed input datasets using a hybrid, three-stage classification approach. The resulting map is then utilized in combination with the LiDAR object height information data to parameterize the microclimate model. To demonstrate the potential of data-driven microclimate modeling, two case studies are presented for selected test sites in the City of Houston, TX. The results of this study highlight that the synergistic combination of hyper- spectral and LiDAR data enables reliable mapping of some of the key input parameters required for urban microclimate modeling. Moreover, classification-based microclimate simula- tions can reveal the thermal properties of urban neighborhoods under varying conditions and, thus, facilitate the identification of hot spot areas and critical land cover configurations.


6’th Place

Title: CLOUDS SHADOW REMOVING IN A HYPERSPECTRAL IMAGE: THEORY AND APPLICATIONS

Authors: Doron E. Bar and Svetlana Raboy

Contact Email: doron.e.bar@gmail.com

Abstract: A novel approach for irradiance calculations for hyperspectral images is presented. It is proved that the solar scattered irradiance through the clouds is not neglected and Mie scattering theory plays a significant role. For the given data set, a linear relation between sunny and shady spectra as a function of wavelength in the visible region was found. Using the general theory of Mie for matching practice and theory, this linear relation is explained as an outcome of the solar irradiance scattering through the clouds, the cloud’s droplet size is found, and other cloud’s parameters are calculated. From the image the penumbra width was measured and the cloud’s base height was estimated. The penumbra width and the relation between sunny and shady spectra are used for a shadow removing algorithm. The new approach enables image reconstruction, and provides a way to better calibrate images under variable illumination conditions.


7’th Place

Title: A NOVEL DATA FUSION FRAMEWORK FOR ELABORATE URBAN CLASSIFICATION USING HYPERSPECTRAL IMAGERY AND LIDAR DATA

Authors: Bei Zhao, Yanfei Zhong, Ji Zhao, and Liangpei Zhang

Affiliations: State Key Laboratory of Information Engineering in Surveying, Mapping, and Remote Sensing, Wuhan University, P. R. China

Contact Email: zhaoys@whu.edu.cn

Abstract: In order to make full use of the complementary characteristics and advantages of hyperspectral imagery and LIDAR data, a novel data fusion framework is proposed in this paper to deal with the elaborate classification of an urban area. The proposed framework contains two main processing chains. One chain employs a divide-and-conquer approach which classifies ground/non-ground pixels obtained by terrain filtering into a subset of the class set and then integrates the classification results. Another chain classifies all the pixels into one of the whole class sets. After the majority voting for the classification maps, and the post- classification with LIDAR data, the final classification map can be obtained. In the proposed framework, the 3-D LIDAR-derived data can be sufficiently integrated into the classification procedure by utilizing the LIDAR data as a feature for the classifiers, to generate the decision of a ground class or non-ground class for the divide-and-conquer approach, and to regularize the buildings and highways for post-processing. The result of the experiment using datasets provided by the Data Fusion Technical Committee (DFTC) of the IEEE Geoscience and Remote Sensing Society shows that the framework relieves the negative effects of shadows and outperforms many other strategies (ranking second).


8’th Place

Title: SPECTRAL AND SPATIAL HYPERSPECTRAL CLASSIFICATION BY ORTHOGONAL PROJECTION WITH PROBABILISTIC DATA FUSION

Authors: Xavier Hadoux, Nathalie Gorretta, Jean-Michel Roger, Gilles Rabatel

Affiliations: UMR ITAP, France

Contact Email: xavier.hadoux@gmail.com

Abstract: To create a highly reliable remote sensing classification map, the optimal synergy between hyperspectral image and LiDAR DSM have to be realized with advanced fusion methods. This paper introduces a data fusion based on pixel level membership degree. The fusion is realized after feature extraction, classification and spatial regularization. The classification method uses orthogonal projection includes an embedded spectrum correction. Every membership degree map can be treated independently, which provides a adjustable method that can be adapted to optimally fit the data. Experimental results demonstrated the good performance of this method.


9’th Place

Title: DATA FUSION AND LAND-USE CLASSIFICATION: AN OPEN SOURCE APPROACH

Authors: P. Kempeneers1, D. McInerney2

Affiliations: 1: VITO, TAP – Remote Sensing department, Belgium; CEA-LSCE, 2: Laboratoire des Sciences du Climat et l’Environnement, Orme des Merisiers, France

Contact Email: kempenep@gmail.com

Abstract: This paper presents a methodology to classify Earth observation data for a study area in Houston, Texas. It was developed from a reproducible research perspective, using open source software only. Data were obtained from the IEEE Geoscience and Remote Sensing Society Data Fusion Contest. Hyperspectral satellite imagery from the CASI airborne scanner was combined with a LiDAR derived digital surface model in a feature fusion approach. The fused dataset was subsequently classified using a support vector machine with a Markov random field based approach using a 15 class land-cover/land-use nomenclature. A morphological filter was applied in a post-processing step to better discriminate specific features.


10’th Place

Title: CORRELATING THE HYPERSPECTRAL IMAGE BANDS WITH THE NORMALIZED TOTAL INSOLATION INDEX (NTII)

Authors: Kazimierz Becek and Andrzej Borkowski

Affiliations: Wroclaw University of Environmental and Life Sciences, Wroclaw, Poland

Contact Email: kbecek@gmail.com

Abstract: In this paper, we examine a dependency of the normalized total insolation index (NTII) and the pixel intensity of a hyperspectral image. The NTII over a test site has been estimated using the light detection and ranging (LiDAR)- derived digital surface model (DSM) and some basic metadata available for the LiDAR and hyperspectral data acquisition. Our preliminary investigations indicate that the NTII may be useful to identify shadows within a scene of a remote sensing image. We have confirmed that there is a relationship between the NTII and the pixel intensities of bands of a hyperspectral image. We claim that these findings may be useful to preprocess a remote sensing multi- or hyperspectral image to adjust for variable levels of insolation. This may lead to improvement of the classification results of multi- or hyperspectral remote sensing datasets.


Comments are closed.