Roel Heremans

Hyperspectral and LiDAR Data Fusion: Outcome of the 2013 GRSS Data Fusion Contest

The 2013 Data Fusion Contest organized by the Data Fusion Technical Committee (DFTC) of the IEEE Geoscience and Remote Sensing Society aimed at investigating the synergistic use of hyperspectral and Light Detection And Ranging (LiDAR) data. The data sets distributed to the participants during the Contest, a hyperspectral imagery and the corresponding LiDAR-derived digital surface model (DSM), were acquired by the NSF-funded Center for Airborne Laser Mapping over the University of Houston campus and its neighboring area in the summer of 2012. This paper highlights the two awarded research contributions, which investigated different approaches for the fusion of hyperspectral and LiDAR data, including a combined unsupervised and supervised classification scheme, and a graph-based method for the fusion of spectral, spatial, and elevation information.

Fusion of PolSAR and PolInSAR data for land cover classification.

Abstract The main research goal of this study is to investigate the complementarity and fusion of different frequencies (L- and P-band), polarimetric SAR (PolSAR) and polarimetric interferometric (PolInSAR) data for land cover classification. A large feature set was derived from each of these four modalities and a two-level fusion method was developed: Logistic regression (LR) as ‘feature-level fusion’ and the neural-network (NN) method for higher level fusion. For comparison, a support vector machine (SVM) was also applied. NN and SVM were applied on various combinations of the feature sets. The results show that for both NN and SVM, the overall accuracy for each of the fused sets is better than the accuracy for the separate feature sets. Moreover, that fused features from different SAR frequencies are complementary and adequate for land cover classification and that PolInSAR is complementary to PolSAR information and that both are essential for producing accurate land cover classification.

Ensemble Learning in Hyperspectral Image Classification: Toward Selecting a Favorable Bias-Variance Tradeoff

Automated classification of hyperspectral images is a fast growing field with numerous applications in the areas of security and surveillance, agriculture, urban management, and environmental monitoring. Although significant progress has been achieved in various aspects of hyperspectral classification (e.g., feature extraction, feature selection, classification, and post-classification processing), the problem has not been addressed so far from a bias-variance decomposition point of view. In this work, we introduce a consistent unified framework that jointly considers all steps in the hyperspectral image classification chain from a bias-variance decomposition perspective. Additionally, we show how state-of-the-art techniques in feature extraction, ensemble-based classification, and post-classification segmentation are related to the bias-variance tradeoff and how this relation can be used to improve classification accuracy. An important outcome of our analysis is that all the steps of the classification chain should be optimized jointly as this unified optimization can guide toward a more favorable bias-variance tradeoff. Experimental results of the proposed framework in the case of four hyperspectral datasets prove the effectiveness of our approach.

Synthetic aperture imaging extended towards novel THz sensors

The purpose of this paper is to extend the technology of synthetic aperture (SA) imaging used in radar (SAR) and sonar (SAS) to the THz range (SAT). This novel approach in THz will be applied in the domain of non-destructive testing (NDT). The paper presents simulated SAT images which are demonstrating the performance of a synthetic aperture time domain reconstruction processing. The generated SAT target responses of broadband radiation show high resolution and high signal-to-noise ratio (SNR) which lead to an improvement of non-destructive defect identification.

Automatic fusion and classification of hyperspectral and LiDAR data using random forests

In this paper we discuss the use of the random forest algorithm for automatic fusion and classification of hyperspectral and LiDAR data. We demonstrate how relative feature relevance can be used in random forests to perform automatic and unsupervised feature selection. This allows using a large number of features without suffering from the curse of dimensionality. The effectiveness of the proposed approach is demonstrated on two datasets. The first dataset features a combination of hyperspectral and LiDAR data for urban classification whereas the second dataset is the well-known Indian Pines dataset featuring pure hyperspectral imagery. We show that by using the proposed approach classification accuracies can be improved significantly.

Multisensor data fusion for IED threat detection

In this paper we present the multi-sensor registration and fusion algorithms that were developed for a force protection research project in order to detect threats against military patrol vehicles. The fusion is performed at object level, using a hierarchical evidence aggregation approach. It first uses expert domain knowledge about the features used to characterize the detected threats, that is implemented in the form of a fuzzy expert system. The next level consists in fusing intra-sensor and inter-sensor information. Here an ordered weighted averaging operator is used. The object level fusion between candidate threats that are detected asynchronously on a moving vehicle by sensors with different imaging geometries, requires an accurate sensor to world coordinate transformation. This image registration will also be discussed in this paper.

Space-time versus frequency domain signal processing for 3D THz imaging

For the first time, to the best of the authors knowledge, a 3D image reconstruction is developed using wide beam THz radiation. The reconstruction method finds its origins in the domain of radar and sonar where it is known as Synthetic Aperture Radar (SAR) and Synthetic Aperture Sonar (SAS) respectively. The extension to the SAR/SAS reconstruction algorithms on wide beam terahertz radiation results in a high-resolution 3D image by combining the depth information (due to the penetration aspect of THz radiation) with a 2D scanning setup. Two 3D reconstruction algorithms have been developed, one in the space-time domain and one in the frequency domain. They both have been validated and analysed using simulated data: the azimuth resolution dependence - on the transmitted frequency and on the opening angle - is compared between the space-time and the frequency domain algorithm as well as their respective computational load. The application of the proposed imaging techniques lays in the domain of non-destructive testing (NDT) in particular for composite aircraft samples.

Motion compensation on synthetic aperture sonar images

High resolution sonars are required to detect and classify mines on the sea-bed. Synthetic aperture sonar increases the sonar cross range resolution by several orders of magnitudes while maintaining or increasing the area search rate. The resolution is however strongly dependent on the precision with which the motion errors of the platform can be estimated. The term micro-navigation is used to describe this very special requirement for sub-wavelength relative positioning of the platform. Therefore algorithms were designed to estimate those motion errors and to correct for them during the (ω, k)-reconstruction phase. To validate the quality of the motion estimation algorithms a single transmitter/multiple receiver simulator was build, allowing to generate multiple point targets with or without surge and/or sway and/or yaw motion errors. The surge motion estimation is shown on real data, which were taken during a sea trial in November of 2003 with the low frequency (12 kHz) side scan sonar (LFSS) moving on a rail positioned on the sea-bed near Marciana Marina on the Elba Island, Italy.

3B-6 Motion Compensation on Synthetic Aperture Sonar Images

The problem of horizontal displacement errors or sway errors in the tow-path of the synthetic aperture sonar (SAS) platform has a devastating effect on the quality of the reconstructed image if they are left uncorrected. The Displaced Phase Center Array algorithm is the most effective of the micro-navigation algorithms developed to date to obtain these path errors, which exploits in a unique way the spatial and temporal coherence properties of the sea-floor backscatter. This DPCA algorithm is described bellow in detail as well as the way how to compensate for this motion

Multilevel ensembling for local climate zones classification

This paper presents an end-to-end system for automatic local climate zones classification of various types of urban environment. For that we perform fusion of multispectral images from Landsat-8 and Sentinel-2 satellites with site description extracted from OpenStreetMap layers. The proposed classification approach is based on a multi-level ensemble scheme that combines Convolutional Neural Networks, Random Forests and Gradient Boosting Machines.