Guided spatial segmentation and knowledge-based remotely sensed data classification

The incompatibility between Earth's natural spectral patterns and human defined information categories makes the automated classification of remotely sensed data difficult. Spatial information contained in Geographic Information System (GIS) can play an important role in resolving the incompatibility between the spectral patterns and information categories. Road network information was used to spatially segment the image into simpler parameters. A subsequent knowledge-based analysis and classification using these parameters resulted in a classification map that agrees with the cartographic standard more closely than traditional remotely sensed image classifications.; Topologically Integrated Geographic Encoding and Referencing (TIGER) line file is a rich source for geographic research and application. The geometric and positional accuracy of TIGER greatly restricted the potential applications of TIGER data. Image road line information was related to TIGER road lines through the Hough Transform. Combining the accurate geometric information of image roads with the topological information of TIGER, a road file with both accurate position and complete topology was created. That new road network file was then used as a spatial segmentation tool.; Resolution merge images provide a better visual effect than its component image sources. The visual quality improvement is very useful in urban structure and reference sample identification. The statistical analysis of this research revealed that the data redundancy was increased after the resolution merge process. The land use or land cover spectral patterns possessed by resolution merge data also differed from the component data sources, demonstrating that resolution merge imagery is a artificial image dataset that did not possess natural spectral characteristics.; Spatial segmentation using the polygons formed by bounding roads contributed to a classification accuracy greater than that of layered classification alone. A relationship between the size of the regions bounded by roads and the information class(es) within these regions was revealed. Finally, a knowledge-based analysis process was designed to implement a road guided spatial segmentation and knowledge-based classification task. The results were accepted both visually and statistically.