Resolving Spectrally Similar Landuse/Landcover Class Conflict in Remote Sensing Images Using Rough Sets

Major source of the geospatial information for a variety of applications including mission critical analysis, land cover mapping, etc. is the satellite based sensors facilitating different manifestations of the terrain. Yet, the spectral signatures between distinct features may sometimes be too subtle to discriminate, given the spectral and radiometric resolution of the sensor systems. The human expert understands the presence of spectrally indiscernible distinct classes based on his/her expertise and that may be defined as the class conflict. We note that the class conflict problem is usually avoided in selecting the landuse/landcover. In this study, the results indicate that the model developed by rough sets facilitate us an insight into the expert´s knowledge and provide an efficient and transparent mechanism to resolve such a real world problem of class conflict. Also, it speaks on the pointer to the extent to which the training dataset may be used