Hybrid classification methods based on consensus from several data sources are considered. Each data source is at first treated separately and modeled using statistical methods. Then weighting mechanisms are used to control the influence... more
Hybrid classification methods based on consensus from several data sources are considered. Each data source is at first treated separately and modeled using statistical methods. Then weighting mechanisms are used to control the influence of each data source in the combined classification. The weights are optimized in order to improve the combined classification accuracies. Both linear and nonlinear optimization methods
Research Interests:
The combination of multisource remote sensing and geographic data is believed to offer improved accuracies in land cover classification. For such classification, the conventional parametric statistical classifiers, which have been applied... more
The combination of multisource remote sensing and geographic data is believed to offer improved accuracies in land cover classification. For such classification, the conventional parametric statistical classifiers, which have been applied successfully in remote sensing for the last two decades, are not appropriate, since a convenient multivariate statistical model does not exist for the data. In this paper, several single
Research Interests: Geography, Geophysics, Statistics, Remote Sensing, Multivariate Statistics, and 18 morePattern Recognition, Neural Networks, Image Classification, Models, Classification, Data Fusion, Boosting, Sensor Fusion, Consensus Theory, Geomatic Engineering, Land cover classification, Accuracy, Boosting Algorithm, Multiple Classifiers, THERMAL INFRARED REMOTE SENSING DATA, Precision, Electrical And Electronic Engineering, and Bagging
Page 1. Support Vector Machines in Multisource Classification Gisli Hreinn Halldorsson, Jon Atli Benediktsson and Johannes R. Sveinsson Engineering Research Institute, University of Iceland, Hjardarhaga 2-6, 107 Reykjavik, Iceland e-mail:... more
Page 1. Support Vector Machines in Multisource Classification Gisli Hreinn Halldorsson, Jon Atli Benediktsson and Johannes R. Sveinsson Engineering Research Institute, University of Iceland, Hjardarhaga 2-6, 107 Reykjavik, Iceland e-mail: {ghh, benedikt, sveinsso}@hi.is ...
Research Interests:
Three methods are investigated for speckle reduction and enhancement of synthetic aperture radar (SAR) images in the orthogonal wavelet domain. The first method is a nonlinear method based on soft thresholding the wavelet coefficients for... more
Three methods are investigated for speckle reduction and enhancement of synthetic aperture radar (SAR) images in the orthogonal wavelet domain. The first method is a nonlinear method based on soft thresholding the wavelet coefficients for logarithmically transformed SAR image data (Guo et al. 1994). The second method uses an enhanced adaptive Lee filter (Lopes et al., 1990) of the wavelet
Research Interests:
Research Interests:
15 Random Forest Classification of Remote Sensing Data Sveinn R. Joelsson, Jon A. Benediktsson, and Johannes R. Sveinsson CONTENTS 15.1 Introduction...................................... more
15 Random Forest Classification of Remote Sensing Data Sveinn R. Joelsson, Jon A. Benediktsson, and Johannes R. Sveinsson CONTENTS 15.1 Introduction................................... .................................................................................. 327 15.2 The Random Forest Classifier........ ...
The purpose of this paper is to develop a method for denoising images corrupted with additive white Gaussian noise (AWGN). The noise degrades quality of the images and makes interpretations, analysis and segmentation of images harder. The... more
The purpose of this paper is to develop a method for denoising images corrupted with additive white Gaussian noise (AWGN). The noise degrades quality of the images and makes interpretations, analysis and segmentation of images harder. The discrete curvelet transform is a new image representation approach that codes image edges more efficiently than the wavelet transform. On the other hand,
Research Interests:
Morphologicalfeature extraction (MFE) has been success-fully used to increase classification accuracy and reduce the noise levelfor classification ofaerial images. In this paper we explore feature selection and extraction for MFE using... more
Morphologicalfeature extraction (MFE) has been success-fully used to increase classification accuracy and reduce the noise levelfor classification ofaerial images. In this paper we explore feature selection and extraction for MFE using random forests (RFs) for classification and ...
Research Interests:
Research Interests: Remote Sensing, Pattern Recognition, Statistical Analysis, Neural Networks, Neural Network, and 13 moreMultidisciplinary, Classification, Data Fusion, Sensor Fusion, Data Transformation, Time Frequency Analysis, Backpropagation, Statistical Pattern Recognition, THERMAL INFRARED REMOTE SENSING DATA, PROBABILITY DENSITY, Classification Accuracy, Conjugate Gradient, and Wavelet packet
Research Interests:
A new artificial neural network (ANN) architecture for classification of event-related potential (ERP) waveforms is proposed. The new architecture is called the parallel principal component neural network (PPCNN). The use of PPCNN is... more
A new artificial neural network (ANN) architecture for classification of event-related potential (ERP) waveforms is proposed. The new architecture is called the parallel principal component neural network (PPCNN). The use of PPCNN is discussed in terms of classification of ERP data obtained from chronic schizophrenic patients and from normal volunteers
Research Interests:
Classification of hyperspectral data with high spatial resolution is discussed. A method based on mathematical morphology for pre-processing of the hyperspectral data is investigated. In this approach, opening and closing morphological... more
Classification of hyperspectral data with high spatial resolution is discussed. A method based on mathematical morphology for pre-processing of the hyperspectral data is investigated. In this approach, opening and closing morphological transforms are used in order to isolate bright (opening) and dark (closing) structures in images, where bright/dark means brighter/darker than the surrounding features in the images. Then, a morphological profile is constructed based on the repeated use of openings and closings with a differently sized structuring element. In order to apply the morphological approach to hyperspectral data, principal components are computed. Then, the principal components are used as base images for the morphological profiles. The use of extended morphological profiles, based on more than one principal component is proposed. In experiments, two data sets are classified. The proposed method performs well in terms of classification accuracies. It gives similar overall accuracies to statistical approaches.