Fusion of Multisensor Remote Sensing Data for Earth Observation
S.B.Serpico
Dept. of Biophysical and Electronic Eng. – Univ. of Genoa (Italy)
Abstract
Recent and forthcoming missions will provide remotely-sensed data of the earth that allow to obtain information about the soil or the sea surface acquired by sensors of different and complementary nature, such as, SAR sensors, optical sensors, and hyperspectral cameras. Appropriate signal processing and analysis, based on a synergistic use of these data, will represent powerful and flexible tools for the investigation of the state of the earth surface and its changes, and for object detection.
Data fusion techniques will be of primary importance in order to exploit the complementary nature of the sensors involved. The purposes may be different, such as a composite display of the information derived from different sensors, target detection, classification, change detection, etc.
Optical and SAR data, for instance, provide quite complementary information, due to the intrinsic differences of the sensors. Many approaches have been proposed in the literature to fuse these two kinds of data for classification purposes. Recently, some proposals have also been presented to integrate the optical data acquired at one date with the SAR data acquired at another date, taking into account that changes may have been occurred between the two dates. This situation is interesting for example when one wants to utilize SAR data acquired during night time or in adverse weather conditions, but such data on themselves do not allow a sufficient classification accuracy to be reached. So the integration with optical data of another date may provide an important contribution to improve the quality of results.
Other examples of exploitation of complementary data are the combined display and analysis of high-spatial resolution multispectral (or panchromatic) data with hyperspectral data of lower spatial resolution. In the case of display, the objective is to present an image to the user that exhibits the same spatial resolution of the multispectral data with the spectral information of the hyperspectral data. Quite impressive results have been presented in the literature on this subject. The information derived from this kind of data fusion may be useful also in the case the purpose is data analysis, for example, when considering the so-called "spectral unmixing" problem, whose solution allows a quantitative estimation of the materials present at ground to be obtained.
The above mentioned data fusion methodologies will be briefly described and discussed. Experimental results obtained with real data will also be presented.