Evidential Approach for Multisensor Fusion Using Beta Distribution

This paper has dealt with a data fusion for the problem of land-cover classification using multisensor imagery. Dempster-Shafer evidence theory has been employed to combine the information extracted from the multiple data of same site. The Dempster-Shafer’s approach has two important advantages for remote sensing application: one is that it enables to consider a compound class which consists of several land-cover types and the other is that the incompleteness of each sensor data due to cloud-cover can be modeled for the fusion process. The image classification based on the Dempster-Shafer theory usually assumes that each sensor is represented by a single channel. The evidential approach to image classification, which utilizes a mass function obtained under the assumption of class-independent beta distribution, has been discussed for the multiple sets of mutichannel data acquired from different sensors. The proposed method has applied to the KOMPSAT-EOC panchromatic imagery and LANDSAT ETM+ data, which were acquired over Yongin/Nuengpyung area of Korean peninsula. The experiment has shown that it is greatly effective on the applications in which it is hard to find homogeneous regions represented by a single land-cover type in training process. INTRODUCTION During the last decade sensor technology has been greatly advanced and continues its revolution, resulting in increasing the number of sensors for space-borne data acquisition of Earth observational applications. Generally images of the same site acquired by different sensors are partially redundant, as they represent the same scene, and partially complementary since the sensors have different characteristics and physical interaction mechanisms. In image classification, data from a single sensor may be insufficient to provide accurate description of a ground scene and a combination of information provided by multiple sensors can then make reduce misclassification due to the redundant and complementary nature of mutisensor data. As an alternative to Bayesian probability, Shafer (1976) first proposed Dempster-Shafer (DS) mathematical theory of evidence by extending the Dempster’s research. Both imprecision and uncertainty can be represented by the evidence theory, which appears more flexible and general than Bayesian approach for decision processes, and Dempster’s rule of combination provides the mechanism for combining independent sources of information. In many problems of land-cover classification using space-borne data, it is important to design the method able to adapt to missing information usually resulting from the presence of clouds in optical sensing and to the case of a compound region which consists of several different land-cover types. In the fusion process of land-cover classification, the DS theory enables to consider not only a simple region of homogenous type but also a compound one as a class. Another ability of the DS theory for multisensor data fusion is to model the incompleteness of each sensor data, either due to clouds or bad observations. In image classification, the DS evidence theory defines a mass function as the basic probability assignment for every hypothesis about pixel class. Lee (2004) presented an approach to estimate Gaussian mass function based on the results of the unsupervised classification based on spatial region growing segmentation, which makes use of hierarchical clustering (2001). This study extends to estimate a mass function based on the Beta distribution which is defined in the value range of limited interval and whose probability density function has a more general form than that of Gaussian distribution used in (Lee, 2004). Although most sensors of space-borne system collect the data through multiple channels of different spectral wave lengths, the fusion process of DS evidence theory usually assumes that each sensor/source is represented by one single channel. Under this assumption, the multichannel data of a single sensor may be transformed to a spectral index (Lee, 2004; Jouan and Allard, 2004), or the principal component may be used for a single channel