Search

The classification of hyperspectral imagery, using multiple classifier systems is discussed and an SVM-based ensemble is introduced. The data set is separated into separate feature subsets using the correlation between the different spectral bands as a criterion. Afterwards, each source is classified separately by an SVM classifier. Finally, the different outputs are used as inputs for final decision fusion that is based on an additional SVM classifier. The results using the proposed strategy are compared to classification results achieved by a single SVM and other well known classifier ensembles, such as random forests, boosting and bagging. [ABSTRACT FROM AUTHOR]

The classification of remote sensing data with imbalanced training data is addressed. The classification accuracy of a supervised method is affected by several factors, such as the classifier algorithm, the input data and the available training data. The use of an imbalanced training set, i.e., the number of training samples from one class is much smaller than from other classes, often results in low classification accuracies for the small classes. In the present study support vector machines (SVM) are trained with imbalanced training data. To handle the imbalanced training data, the training data are resampled (i.e., bagging) and a multiple classifier system, with SVM as base classifier, is generated. In addition to the classifier ensemble a single SVM is applied to the data, using the original balanced and the imbalanced training data sets. The results underline that the SVM classification is affected by imbalanced data sets, resulting in dominant lower classification accuracies for classes with fewer training data. Moreover the detailed accuracy assessment demonstrates that the proposed approach significantly improves the class accuracies achieved by a single SVM, which is trained on the whole imbalanced training data set. [ABSTRACT FROM AUTHOR]

Several widely used methods have been proposed for fusing high resolution panchromatic data and lower resolution multi-channel data. However, many of these methods fail to maintain spectral consistency of the fused high resolution image, which is of high importance to many of the applications based on satellite data. Additionally, most conventional methods are loosely connected to the image forming physics of the satellite image, giving these methods an ad hoc feel. In this chapter, a method for image fusion of satellite images is given. The method is based on the properties of imaging physics in a statistically meaningful way. Based on our analysis, it is seen that spectral consistency is a direct consequence of imaging physics and hence guaranteed by our method. This is achieved while exploiting the high resolution single-channel data in what can be seen as a statistical optimal way, yielding a framework to which additional constraints can be added in a straight forward manner. In this chapter we exploit this framework and add some simple optimization terms for smoothing the fused image. Specifically, the method is based on the observation that any given channel of the satellites imaging device can be seen as an inner-product between the radiated light arriving at the sensor and the spectral response function of that channel. This gives a simple inner product space encompassing the relationship between the different channels as well as imposing spectral consistency. Normal distributed statistics - inducing the same norm as the above mentioned inner product - are used for regularization. This yields a framework to which additional constraints are added in a straight forward manner. Here we add a simple term to the discussed framework for smoothing the image, although more elaborate terms might be preferable. Computationally, we achieve a solution for the derived objective function via stochastic optimization, i.e., using the Metropolis algorithm in conjunction with simulated annealing. Apart from contributing with a novel analysis giving more insight into the image fusion problem, the method proposed in this chapter has been applied to images from the IKONOS satellite. These experimental results validate the proposed method. [ABSTRACT FROM AUTHOR]

The need to optimize the classification accuracy of remotely sensed imagery has led to an increasing use of Earth observation data with different characteristics collected from a variety of sensors from different parts of the electromagnetic spectrum. Combining multisource data is believed to offer enhanced capabilities for the classification of target surfaces. In the paper several single and multiple classifiers which are appropriate for classification of multisource remote sensing and geographic data are considered. The focus is on multiple classifiers: bagging algorithms, boosting algorithms, and consensus theoretic classifiers. These multiple classifiers have different characteristics. The performance of the algorithms in terms of accuracies is compared for a multisource remote sensing and geographic data set. [ABSTRACT FROM AUTHOR]