Erzsébet Merényi

Neural maps in remote sensing image analysis

We study the application of self-organizing maps (SOMs) for the analyses of remote sensing spectral images. Advanced airborne and satellite-based imaging spectrometers produce very high-dimensional spectral signatures that provide key information to many scientific investigations about the surface and atmosphere of Earth and other planets. These new, sophisticated data demand new and advanced approaches to cluster detection, visualization, and supervised classification. In this article we concentrate on the issue of faithful topological mapping in order to avoid false interpretations of cluster maps created by an SOM. We describe several new extensions of the standard SOM, developed in the past few years: the growing SOM, magnification control, and generalized relevance learning vector quantization, and demonstrate their effect on both low-dimensional traditional multi-spectral imagery and approximately 200-dimensional hyperspectral imagery.

Exploiting Data Topology in Visualization and Clustering of Self-Organizing Maps

The self-organizing map (SOM) is a powerful method for visualization, cluster extraction, and data mining. It has been used successfully for data of high dimensionality and complexity where traditional methods may often be insufficient. In order to analyze data structure and capture cluster boundaries from the SOM, one common approach is to represent the SOMs knowledge by visualization methods. Different aspects of the information learned by the SOM are presented by existing methods, but data topology, which is present in the SOMs knowledge, is greatly underutilized. We show in this paper that data topology can be integrated into the visualization of the SOM and thereby provide a more elaborate view of the cluster structure than existing schemes. We achieve this by introducing a weighted Delaunay triangulation (a connectivity matrix) and draping it over the SOM. This new visualization, CONNvis, also shows both forward and backward topology violations along with the severity of forward ones, which indicate the quality of the SOM learning and the data complexity. CONNvis greatly assists in detailed identification of cluster boundaries. We demonstrate the capabilities on synthetic data sets and on a real 8D remote sensing spectral image.

Retrieval of apparent surface reflectance from AVIRIS data: A comparison of empirical line, radiative transfer, and spectral mixture methods☆

Abstract Three methods for converting Airborne Visible / Infrared Imaging Spectrometer (AVIRIS) radiance data to apparent surface reflectance were compared using data collected over the Lunar Crater Volcanic Field in Nevada and the Pavant Butte tuff cone in Utah. The methods examined were the empirical line method, radiative transfer modeling (using LOWTRAN 7), and spectral mixture analysis using reference endmembers. Of the three, the empirical line and spectral mixture methods both provided good results. The approach utilizing LOWTRAN 7 accentuates noise inherent in AVIRIS data and requires a very accurate estimate of atmospheric water.

Classification of asteroid spectra using a neural network

The 52-color asteroid survey (Bell et al., 1988) together with the 8-color asteroid survey (Zellner et al., 1985) provide a data set of asteroid spectra spanning 0.3–2.5 μm. An artificial neural network clusters these asteroid spectra based on their similarity to each other. We have also trained the neural network with a categorization learning output layer in a supervised mode to associate the established clusters with taxonomic classes. Results of our classification agree with Tholens classification based on the 8-color data alone. When extending the spectral range using the 52-color survey data, we find that some modification of the Tholen classes is indicated to produce a cleaner, self-consistent set of taxonomic classes. After supervised training using our modified classes, the network correctly classifies both the training examples, and additional spectra into the correct class with an average of 90% accuracy. Our classification supports the separation of the K class from the S class, as suggested by Bell et al. (1987), based on the near-infrared spectrum. We define two end-member subclasses which seem to have compositional significance within the S class: the So class, which is olivine-rich and red, and the Sp class, which is pyroxene-rich and less red. The remaining S-class asteroids have intermediate compositions of both olivine and pyroxene and moderately red continua. The network clustering suggests some additional structure within the E-, M-, and P-class asteroids, even in the absence of albedo information, which is the only discriminant between these in the Tholen classification. New relationships are seen between the C class and related G, B, and F classes. However, in both cases, the number of spectra is too small to interpret or determine the significance of these separations.

Relevance-Based Feature Extraction for Hyperspectral Images

Hyperspectral imagery affords researchers all discriminating details needed for fine delineation of many material classes. This delineation is essential for scientific research ranging from geologic to environmental impact studies. In a data mining scenario, one cannot blindly discard information because it can destroy discovery potential. In a supervised classification scenario, however, the preselection of classes presents one with an opportunity to extract a reduced set of meaningful features without degrading classification performance. Given the complex correlations found in hyperspectral data and the potentially large number of classes, meaningful feature extraction is a difficult task. We turn to the recent neural paradigm of generalized relevance learning vector quantization (GRLVQ) [B. Hammer and T. Villmann, Neural Networks, vol. 15, pp. 1059-1068, 2002], which is based on, and substantially extends, learning vector quantization (LVQ) [T. Kohonen, Self-Organizing Maps, Berlin, Germany: Springer-Verlag, 2001] by learning relevant input dimensions while incorporating classification accuracy in the cost function. By addressing deficiencies in GRLVQ, we produce an improved version, GRLVQI, which is an effective analysis tool for high-dimensional data such as remotely sensed hyperspectral data. With an independent classifier, we show that the spectral features deemed relevant by our improved GRLVQI result in a better classification for a predefined set of surface materials than using all available spectral channels.

A Validity Index for Prototype-Based Clustering of Data Sets With Complex Cluster Structures

Evaluation of how well the extracted clusters fit the true partitions of a data set is one of the fundamental challenges in unsupervised clustering because the data structure and the number of clusters are unknown a priori. Cluster validity indices are commonly used to select the best partitioning from different clustering results; however, they are often inadequate unless clusters are well separated or have parametrical shapes. Prototype-based clustering (finding of clusters by grouping the prototypes obtained by vector quantization of the data), which is becoming increasingly important for its effectiveness in the analysis of large high-dimensional data sets, adds another dimension to this challenge. For validity assessment of prototype-based clusterings, previously proposed indexes-mostly devised for the evaluation of point-based clusterings-usually perform poorly. The poor performance is made worse when the validity indexes are applied to large data sets with complicated cluster structure. In this paper, we propose a new index, Conn_Index, which can be applied to data sets with a wide variety of clusters of different shapes, sizes, densities, or overlaps. We construct Conn_Index based on inter- and intra-cluster connectivities of prototypes. Connectivities are defined through a “connectivity matrix”, which is a weighted Delaunay graph where the weights indicate the local data distribution. Experiments on synthetic and real data indicate that Conn_Index outperforms existing validity indices, used in this paper, for the evaluation of prototype-based clustering results.

Learning Highly Structured Manifolds: Harnessing the Power of SOMs

In this paper we elaborate on the challenges of learning manifolds that have many relevant clusters, and where the clusters can have widely varying statistics. We call such data manifolds highly structured . We describe approaches to structure identification through self-organized learning, in the context of such data. We present some of our recently developed methods to show that self-organizing neural maps contain a great deal of information that can be unleashed and put to use to achieve detailed and accurate learning of highly structured manifolds, and we also offer some comparisons with existing clustering methods on real data.

Automated Labeling of Materials in Hyperspectral Imagery

We present a technique for automatically labeling segmented hyperspectral imagery with semantically meaningful material labels. The technique compares the mean signatures of each image segment to a spectral library of known materials, and material labels are assigned to image segments according to the most similar library entry. The similarity between spectral signatures is evaluated using our recently proposed CICRd similarity measure designed specifically for hyperspectral imagery. This measure considers both the continuum-intact reflectance spectrum and its continuum-removed representation. We provide a thorough assessment of this measure by comparison to several commonly used similarity measures on a well-studied low-altitude Airborne Visible/Infrared Imaging Spectrometer image of an urban area. We evaluate our results using both information-theoretic techniques and visual validation of the resulting spectral matches.

Extensions and modifications of the Kohenen-SOM and applications in remote sensing image analysis

Utilization of remote sensing multi- and hyperspectral imagery has shown a rapid increase in many areas of economic and scientific significance over the past ten years. Hyperspectral sensors, in particular, are capable of capturing the detailed spectral signatures that uniquely characterize a great number of diverse surface materials. Interpretation of these very high-dimensional signatures, however, has proved an insurmountable challenge for many traditional classification, clustering and visualization methods. This chapter presents spectral image analyses with Self-Organizing Maps (SOMs). Several recent extensions to the original Kohonen SOM are discussed, emphasizing the necessity of faithful topological mapping for correct interpretation. The effectiveness of the presented approaches is demonstrated through case studies on real-life multi- and hyperspectral images.

The landscape of Comet Halley

Abstract In this paper the three-dimensional model of P/Halleys nucleus is constructed based on the results of the imaging experiments. It has the following major characteristics: the overall sizes are 7.2, 7.22, and 15.3 km; its volume is 365 km 3 ; the ratios of the inertial momenta are 3.5:3.4:1, the longest inertial axis is slightly inclined to the geometrical axis of the body, 3.3° latitude and 1.1° longitude if the geometrical axis is at zero latitude and longitude. The shape is very irregular; it is not close to any regular body, e.g., to a tri-axial ellipsoid. Several features on the images are interpreted as local slope variations of the nucleus.