Jürgen T. Hahn

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.

Target Discrimination and Classification in Through-the-Wall Radar Imaging

In this paper, a scheme for target discrimination and classification is proposed. The proposed scheme is applied to through-the-wall microwave images obtained by using a wideband radar implementing frequency-domain back-projection. We consider stationary targets where Doppler and change-detection based techniques are inapplicable. The proposed scheme applies image segmentation, followed by feature extraction. We map target returns to a feature space, where discrimination among different targets and clutter is performed. To achieve target-clutter discriminations independent of target location in range and cross-range, we use compensation methods to account for varying system resolution within the perimeter of the scene imaged. Real data collected using an indoor radar imaging scanner is used for validation of performance.

Compressive sensing and adaptive direct sampling in hyperspectral imaging

Hyperspectral imaging (HSI) is an emerging technique, which provides the continuous acquisition of electro-magnetic waves, usually covering the visible as well as the infrared light range. Many materials can be easily discriminated by means of their spectra rendering HSI an interesting method for the reliable classification of contents in a scene. Due to the high amount of data generated by HSI, effective compression algorithms are required. The computational complexity as well as the potentially high number of sensors render HSI an expensive technology. It is thus of practical interest to reduce the number of required sensor elements as well as computational complexity - either for cost or for energy reasons. In this paper, we present two different systems that acquire hyperspectral images with less samples than the actual number of pixels, i.e. in a low dimensional representation. First, a design based on compressive sensing (CS) is explained. Second, adaptive direct sampling (ADS) is utilized to obtain coefficients of hyperspectral images in the 3D (Haar) wavelet domain, simplifying the reconstruction process significantly. Both approaches are compared with conventionally captured images with respect to image quality and classification accuracy. Our results based on real data show that in most cases only 40% of the samples suffice to obtain high quality images. Using ADS, the rate can be reduced even to a greater extent. Further results confirm that, although the number of acquired samples is dramatically reduced, we can still obtain high classification rates.

Feature extraction in Through-the-Wall radar imaging

This paper deals with the problem of automatic target classification or Through-the-Wall radar imaging. The proposed scheme considers stationary objects in enclosed structures and works on the SAR image rather than the raw data. It comprises segmentation, feature extraction based on superquadrics, and classification. We present a recursive splitting tree to obtain optimum parameters for feature extraction. Support vector machines and nearest neighbor classifiers are then applied to successfully classify among different indoor targets. The classification methods are tested and evaluated using real data generated from synthetic aperture Through-the-Wall radar imaging experiments.

Inverse Reinforcement Learning using Expectation Maximization in mixture models

Reinforcement Learning (RL) is an attractive tool for learning optimal controllers in the sense of a given reward function. In conventional RL, usually an expert is required to design the reward function as the efficiency of RL strongly depends on the latter. An alternative has been presented by the concept of Inverse Reinforcement Learning (IRL), where the reward function is estimated from observed data. In this work, we propose a novel approach for IRL based on a generative probabilistic model of RL. We derive an Expectation Maximization algorithm that is able to simultaneously estimate the reward and the optimal policy for finite state and action spaces, which can be easily extended for the infinite cases. By means of two toy examples, we show that the proposed algorithm works well even with a low number of observations and converges after only a few iterations.

Bootstrap-based SVM aggregation for class imbalance problems

Support Vector Machines (SVMs) are considered to be one of the most powerful classification tools, widely used in many applications. However, in numerous scenarios the classes are not equally represented and the predictive performance of SVMs on such data can drop dramatically. Different methods have been proposed to address moderate class imbalance issues, but there are few methods that can be successful at detecting the minority class while also keeping high accuracy, especially when applied to datasets with significant level of imbalance. In this paper, we consider SVM ensembles that are built by using a bootstrap-based undersampling technique. We target to reduce the bias induced by class imbalances via multiple undersampling procedures and then decrease the variance using SVM ensembles. For combining the SVMs, we propose a new technique that deals with class imbalance problems of varying levels. Experiments on several datasets demonstrate the performance of the proposed scheme compared to state-of-the-art balancing methods.

Adaptive Compressed Classification for hyperspectral imagery

Hyperspectral imaging (HSI) is a useful tool for the classification of vast areas. High accuracy is achieved by means of spectral information for each pixel, which inherently leads to a huge amount of data and, thus, requires costly processing. We present an Adaptive Compressed Classification (ACC) framework for HSI that allows a compressive acquisition of the scene of interest. Since classification is performed in the compressive domain, expensive reconstruction is avoided, significantly reducing computational requirements. For ACC, we propose an adaptive probabilistic approach to optimize the measurement and basis matrices. Based on real data sets, we show that Compressed Classification yields high classification accuracy close to results obtained for the complete data. Using the proposed adaptive approach, even higher accuracies are achieved in all tested cases.

Risk-sensitive decision making via constrained expected returns

Decision making based on Markov decision processes (MDPs) is an emerging research area as MDPs provide a convenient formalism to learn an optimal behavior in terms of a given reward. In many applications there are critical states that might harm the agent or the environment and should therefore be avoided. In practice, those states are often simply penalized with a negative reward where the penalty is set in a trial-and-error approach. For this reason, we propose a modification of the well-known value iteration algorithm that guarantees that critical states are visited with a pre-set probability only. Since this leads to an infeasible problem, we investigate the effect of nonlinear and linear approximations and discuss the effects. Two examples demonstrate the effectiveness of the proposed approach.