Unsupervised urban land-cover classification using WorldView-2 data and self-organizing maps

Fully automated land-cover classification from commercial remote sensing satellite imagery has had limited success in part due to their limited spectral bands. New promise of unsupervised analysis has emerged with the recent launch of the eight-band high resolution satellite WorldView-2. In this paper, a fully unsupervised classification algorithm is proposed based on self-organizing maps and watershed segmentation. The results demonstrate that the proposed algorithm performs better in classifying homogeneous regions while achieving better accuracy than k-means.