Remote Sensing Image Fusion for Unsupervised Land Cover Classification

Abstract

1.1 Context With the development of new satellite systems and the accessibility of data from public through web services like Google Earth, remote sensing imagery, knows today an important growing which advanced and still advances researches in this area on different aspects. Especially in cartography, many studies have been conducted for multi-source satellite images classification. These studies aim to develop automatic tools in order to facilitate the interpretation and provide a semantic land cover classification. Classical tools based on satellite images deal essentially with one category of satellite images which allows a partial interpretation. Multi-sensor or multi-source image fusion have been applied in the field of remote sensing since 20 years and continues today to provide efficient solutions to problems related to detection and classification. The work presented in this chapter is a part of multi-source fusion research efforts to have reliable and automatic satellite image interpretation. We propose to apply the new fusion concepts and theories for multi-source satellite images. Our main motivation is to measure the real contribution of multi-source image fusion according to the exploitation of satellite images separately. Recent studies suggest that the combination of imagery from satellites with different spectral, spatial, and temporal information may improve land cover classification performance. The use of multi-source satellites images fully take into account the complementary and supplementary information provided by different data sources and considerably optimize the classification of cartographic objects. Particularly, combination of optical and radar remote sensing data may improve the classification results because of the complementarities of these two sources. Spectral features extracted form optical data may remove some difficulties faced when using only radar images. However, radar images present the following massive advantage: the possibility of penetrating the clouds. Thus, data fusion technique is applied to combine these two kinds of information.