| Operations on Remotely Sensed Imagery, a form of spatial data, pose unique challenges. The dataset consists of hundreds of millions of pixels, structured as one pixel per tuple, but of usually less than ten attributes. Therefore, scalability in terms of cardinality is an issue. However, cardinality problems can be resolved by preprocessing data, i.e., pruning the dataset. To this end, we propose attribute-level classification methods--based on the data's spatial and spectral characteristics--to address spatial data mining scalability issues, thus aiding in visualizing the dataset for subsequent data mining by a human expert, since the human brain is the most complex, sophisticated, and powerful information-processing device known.; In order to accomplish this, we define attribute-level operations as functionals and extend the concept of functional contours to include contour-pruning preprocessing; we use the byte as the basic unit of information storage and retrieval. For visualizing n-dimensional data, we study the Jewell Diagram and develop the Augmented Himalayan Chain. Finally, we implement our methods on an experimental testbed, TM-Mine, and test our algorithms in a mixed-language environment using Assembly language subroutines called from a C++ shell.; This work is presented in a multi-paper format: Paper 1: Visualizations of High-Dimensional Space; Paper 2: Spatial Proximity of Structural Attributes in Analyzing RSI: A Survey of Analysis Methods; Paper 3: TM-LAB: An Experimental Program for Viewing and Investigating Thematic Mapper Imagery on the PC; and Paper 4: A C++ Graphics Programming Toolkit for Developing Remote Sensing Applications. The first two papers were presented at CATA 2007 and the second two papers at PECORA 13.; This dissertation is organized as follows: a General Introduction that describes the datasets, including an extended overview of the technology and their attribute-level organizations, briefly discusses classification and summarizes classification methods, introduces a classification method that defines attribute-level operations as functionals and utilizes contour-pruning preprocessing, and looks at functionals in terms of visualization diagrams that could aid as preliminary identification of clusters and outliers; the four published papers; General Conclusions; and descriptions of the Testbed and Benchmark environments. |
