An Interactive Segmentation and Generalisation of Satellite Imagery to Built

要旨

This paper provides overview the past and current developments in generalisation of raster data and to present the authors’ efforts in developing a new methodological framework for segmentation and generalisation of raster data which is known as “Interactive Automated Segmentation and Raster Generalisation Framework” (IASRGF). IASRGF eliminates some of the drawbacks associated with the supervised classification and raster generalisation. The test result of IASGRF shows that all objects derived by generalising land use data from Landsat 7 imagery over Canberra in Australia are satisfactorily classified and mapped. The classification accuracy was 85.5%, whereas the commission errors were relatively high at 38.5%. The maximum likelihood classifier using training sites and associated ground truth data produced an overall accuracy of 85.5% and 0.798 Kappa coefficients. To enhance the supervised classification results, post-classification was carried out. It was found that this improved the overall classification accuracy slightly, but commission errors significantly increased (by 6%).