Kernel-based reclassification algorithm applied on very high spatial resolution satellite imagery of complex ecosystems

摘要

Kernel-based reclassification algorithm derives information on specific thematic classes on the basis of the frequency and spatial arrangement of land cover classes within a square kernel. This algorithm has been originally developed and validated for the urban environment. The present work investigates the potential of projecting this technique to the classification of very high spatial resolution satellite imagery of natural ecosystems. For that purpose a software tool has been developed. The output, apart from the reclassified image, includes a post-classification probability map which shows the areas where the kernel reclassification algorithm has given valid results. The software was tested on an IKONOS image of Lake Kerkini (Greece), a wetland of great ecological value, included in the NATURA 2000 list of ecosystems. The results show that the algorithm has responded successfully in most cases overcoming problems previously encountered by pixel-based classifiers, such as pixel noise.