Land use/cover mapping using multisensor image fusion technique


Abdikan S., Sanli F. B. , ESETLİLİ M. T. , KURUCU Y.

28th European-Association-of-Remote-Sensing-Laboratories (EARSeL) Symposium and Workshops on Remote Sensing for a Changing Europe, İstanbul, Türkiye, 2 - 05 Haziran 2008, ss.157-164 identifier

Özet

Remote Sensing is an important technique for mapping land use and land cover in the vast acreages. In this sense, the fusion of optical and radar remote sensing data offers the opportunity to combine complementary sensors with different features. In this study, beside the capability of the combined multi source imagery, the contribution of SAR images to the optical images for identifying land use/cover types was investigated. For this purpose, using the synergy between SAR and Optical data, the improvement in the classification accuracy was analyzed. The study area, covering urban and agricultural areas, lies in the Menemen Plain to the west of Gediz Basin in the Aegean Region of Turkey. The satellite data used in this study are multispectral SPOT, ENVISAT-ASAR, and ALOS-PALSAR images. The 3-2-1 band combination of a SPOT-2 image was fused with C band ASAR imagery and with the new mission L band PALSAR imagery. The land use/cover types were defined from both of the fused images. In this case, since the SAR images have different bands (C band and L band) the penetration property is the key factor to see the affects on extracting information from fused images. Before the fusion application, the speckle reducing filter techniques were used for the preprocessing of SAR images. For the filtering of SAR images, kernel windows with different size were tried. Then the SPOT image was registered to SAR images. For the registration of SAR images, image to image registration method was used with a root mean square error of less than 1 pixel. A pixel based fusion method was carried out. Both of the fused images (SPOT-ASAR and SPOT-PALSAR) were classified to determine the land use/cover map. The results were compared with a classified SPOT image, which is commonly used to define land cover types. While processing the classification, the training areas were selected covering a large portion of the individual fields and were away from the field boundaries to reduce the mixed pixels. The ground truth data were used for the accuracy assessment process.