The importance of dynamic and complex interactions among urban land use, land-cover change, and global environmental change is well recognized. Despite significant advances in geographic information science and technology, however, effectively categorizing digital remote sensing data into detailed urban land categories remains a challenge. This research project aims to develop a frequency-based, multi-scale classification algorithm using overcomplete wavelet transforms that can generate higher-level spatial arrangements of objects and features for detailed urban land categorization. The investigator will seek to enhance spatial modeling and concepts that describe spatial association, spatial pattern, spatial regression, and segregation by adding decomposition procedures that can extract spatial features in different directions at infinite scale. Selected wavelet transforms to be used in the project include a series of Daubechies and Coiflets. The project's multi-faceted approach will permit the new algorithm to be used for any level of scale, from large-scale air photos to coarse-resolution MODIS and AVHRR data. Five distinct normalization procedures will be used to prevent large-range features from dominating the distance measure. The performance of a minimum distance classifier will be evaluated for texture classification using the computed texture feature value of the sub-images. In addition, the Mahalanobis distance rule will be employed to account for the variability of classes. Finally, three classification decision rules will be developed. The project is expected to result in a new wavelet-based framework that provides scientific evidence for the role of spatial properties and frequencies of geo-objects in different directions at infinite scale. This framework and algorithm will be made publicly available for performing image classification effectively in remotely sensed imagery.
This project will enable researchers, modelers, and analysts to differentiate among urban land-cover and land-use types and to categorize detailed urban land data. It is expected to generate maps that are more accurate, thereby improving mapping and analysis procedures. The new geospatial frequency-based framework and methods will help improve semi-automated and automated analysis procedures. The tools from this project will be versatile and can be applied to a wide variety of other land cover, land-use types (i.e., agriculture, rangeland, forest, wetlands, and coastal zones) and conditions (i.e., drought, fire fuel concentrations, desertification, flood risk, and coastal erosion), thereby making it very useful for more than just urban planning. Furthermore, it will advance geographic information science by building a foundation for methodological innovations in the field.