Abstract
The distinguishing property of remotely sensed data is the multivariate information coupled with a two-dimensional pictorial representation amenable to visual interpretation. The contribution of this work is the design and implementation of various schemes that exploit this property. This dissertation comprises two distinct parts. The essence of Part One is the algebraic solution for the partition function of a high-order lattice model of a two dimensional binary particle system. The contribution of Part Two is the development of a procedural framework to guide multispectral image analysis. The characterization of binary (black and white) images with little semantic content is discussed in Part One. Measures of certain observable properties of binary images are proposed. A lattice model is introduced, the solution to which yields functional mappings from the model parameters to the measurements on the image. Simulation of the model is explained, as is its usage in the design of Bayesian priors to bias classification analysis of spectral data. The implication of such a bias is that spatially adjacent remote sensing data are identified as belonging to the same class with a high likelihood. Experiments illustrating the benefit of using the model in multispectral image analysis are also discussed. The second part of this dissertation presents a procedural schema for remote sensing data analysis. It is believed that the data crucial to a succc~ssful analysis is provided by the human, as an interpretation of the image representation of the remote sensing spectral data. Subsequently, emphasis is laid on the design of an intelligent implementation of existing algorithms, rather than the development of new algorithms for analysis. The development introduces hyperspectral analysis as a problem requiring multi-source data fusion and presents a process model to guide the design of a solution. Part Two concludes with an illustration of the schema as used in the classification analysis of a given hyperspectral data set.
Date of this Version
August 1999