Classification of hazelnut orchards by self-organizing maps

Land cover identification from remote sensing images by using automatic methods has been essential for agricultural management and monitoring. In particular, accurate detection of lands covered with nuts orchards from VHR images is an ongoing challenge mainly due to varying statistics of orchards, such as crown sizes, overlapping crowns, distances between the orchards, existence of different tree species and surface structure. An innovative approach, becoming more popular everyday to overcome similar problems, is to merge spectral and spatial information for utilizing their advantages. In this paper, we propose a self-organizing map that exploits these two information without additional computation of texture. Experimental results on detection of hazelnut orchards from Quickbird imagery show that the proposed method outperforms methods based only on spectral or on spatial information.