Comparison of pixel-based, object-based and sequential masking classification procedures for land use and land cover mapping using multiple sensor SAR in SWEDEN

Multiple sensor applications have become increasingly common in recent years and offer new opportunities to the remote sensing community to extract better information about the earth surface. However, the processing of multiple sensor SAR for land use and land cover mapping is not straightforward and still needs more investigation in order to become operational. This study investigates the efficiency of three different types of classification procedures, namely pixel-based, object-based and sequential masking to extract land use and land cover information from multiple sensor SAR images using the same training and validation areas. Four sensors (RADARSAT finebeam, RADARSAT standard-beam, ERS-2, and JERS-1) in different combinations were investigated in two study areas, to compare their effectiveness for accurate land cover mapping. The results indicate that the pixel-based classifier namely ANN is more accurate (around 90% overall accuracy and 0.90 Kappa coefficient) compared with object-based classification for extracting land use and land cover information from multiple sensor SAR. Overall it was found that the best performance (more than 90% overall accuracy and more than. 90 Kappa coefficient) can be achieved using a sequential masking approach because of its step by step classification technique.