Information fusion for rural land-use classification with high-resolution satellite imagery

Abstract

We propose an information fusion method for the extraction of land-use information based on both the panchromatic and multispectral Indian Remote Sensing Satellite 1C (IRS-1C) satellite imagery. It integrates spectral, spatial and structural information existing in the image. A thematic map was first produced with a maximum-likelihood classification (MLC) applied to the multispectral imagery. Probabilistic relaxation (PR) was then performed on the thematic map to refine the classification with neighborhood information. Furthermore, we incorporated edges extracted from the higher resolution panchromatic imagery in the classification. An edge map was generated using operations such as edge detection, edge thresholding and edge thinning. Finally, a modified region-growing approach was used to improve image classification. The procedure proved to be more effective in land-use classification than conventional methods based only on multispectral data. The improved land-use map is characterized with sharp interregional boundaries, reduced number of mixed pixels and more homogeneous regions. The overall kappa statistics increased considerably from 0.52 before the fusion to 0.75 after.