Interpreting synthetic ground-penetrating radar using object-oriented, neural, fuzzy, and genetic processing.

Abstract

This project, funded by NASA through the University of Arizona Space Engineering Research Center, is an extension of earlier work, and is aimed at developing the base technology for continuous profiling geophysical systems that will be able to determine not only where anomalous features lie, but also what they look like and, ultimately, what has caused them. A hybrid approach was used that employed object oriented and procedural programming, neural networks, fuzzy systems theory, genetic algorithms, and symbolic processing initially within a distributed computing architecture, and later on an 80486 PC platform. Neural networks were used to map synthetic GPR patterns to geometric models. Object oriented programming was combined with fuzzy theory to map these geometric models to a database of real world objects. GPR objects were then combined to build extended objects and systems. Genetic algorithms were used to fine tune the system. Testing was done solely with synthetic data, with future intent of progressing to laboratory and field geophysical patterns. The field context assigned to the vision system for this research corresponds to a buried prehistoric archaeological site comprising occupation and utility/storage rooms of various sizes. This context was chosen to take advantage of the rich suite of GPR patterns already collected, and the GPR modeling underway to characterize archaeological signatures on GPR. A room system was selected as the GPR target because it incorporated most of the basic characteristics expected for GPR systems in general. The interpretation system was able to handle, (1) Multiple components; (2) Offset components; (3) Non-continuous components; (4) Nonmonotonic states; (5) Fuzziness.