A neural network based approach to detecting urban land cover changes using Landsat TM and IKONOS imagery

Much attention has been drawn to the new applications and opportunities afforded by high-resolution satellite imagery, such as IKONOS and QuickBird. The purpose of this paper is to examine the extent to which high-resolution change detection can be performed using a combination of high and medium-resolution satellite imagery. This combination is important for detecting changes during the time before and after the high-resolution satellite imagery was made available. In particular, the analysis is oriented towards smaller cities and municipalities. Many change detection algorithms and methods have been evaluated. The post-classification change detection algorithm was deemed to be the most suitable technique for this project. Landsat 5 TM and IKONOS MS images of Fredericton, New Brunswick, Canada, were used as source data for the change detection. The results tend to suggest that it is possible to extract reliable change detection information pertaining to small streets, and new rows of residential housing with a medium-resolution benchmark. However, the detection of change in individual houses and small buildings proved to be beyond the capabilities of this procedure.