Supervised and unsupervised landuse map generation from remotely sensed images using ant based systems

The landuse or land-cover map depicts the physical coverage of the Earth's terrestrial surface according to its use. Landuse map generation from remotely sensed images is one of the challenging tasks of remote sensing technology. In this article, motivated from group forming behavior of real ants, we have proposed two novel ant based (one supervised and one unsupervised) algorithms to automatically generate landuse map from multispectral remotely sensed images. Here supervised landuse map generation is treated as a classification task which requires some labeled patterns/pixels beforehand, whereas the unsupervised landuse map generation is treated as a clustering based image segmentation problem in the multispectral space. Investigations are carried out on four remotely sensed image data. Experimental results of the proposed algorithms are compared with corresponding popular state of the art techniques using various evaluation measures. Potentiality of the proposed algorithms are justified from the experimental outcome on a number of images. (C) 2011 Elsevier B.V. All rights reserved.