부산대학교 도서관

Sonar has been the Navy's workhorse for detection and discrimination of underwater objects from clutter through analysis of image phenomena. However, new phenomena need to be identified to accommodate a growing class of unexploded ordnance targets with more varied shapes and sizes, and difficulties increase for buried ordnance because the wave attenuation and inhomogeneity in ocean sediments make detection less predictable and high-resolution imaging more difficult. Even when imaging can be done, important image phenomena (e.g., highlight/shadow details) are lost. Therefore, modeling, data collection, and data processing/analysis has been performed at NSWC PCD to build a target response database that could be exploited both for statistical automated target recognition studies based on existing feature types and for isolation of promising physical phenomena that classifiable features could be extracted from without requiring high image resolution. Several methods to build this database have been pursued, including data collection in full-scale field experiments in the Gulf of Mexico, data collection in NSWC PCD's freshwater pond facility, data collection within a small freshwater tank using scaled targets on scaled glass bead sediments, and data generation through finite element simulations. In this paper, results from each of these methods are presented and compared to assess the tradeoffs inherent in them.