Feature-Based Compression and Browsing of Tactical Data

Wavelets are the topic of much excitement in mathematics, engineering, and scientific circles, particularly in the field of data compression. We propose to develop a suite of advanced wavelet compression algorithms especially designed for low frequency acoustic and transient signals. These algorithms will incorporate the latest wavelet transforms, including wavelet packets and cosine packets. The techniques will be designed to maximize classification performance. To develop and evaluate the data compression methods, we will build a realistic testbed of low frequency acoustic and transient data. This testbed includes both real data, extracted from the "Hudson Canyon experiment" and simulated data generated from the KRAKEN model. A range of methods will be used to evaluate the effects of compression, including a mode coupling analysis and assessment of the misclassification rate. This research proposal involves a partnership between the StatSci Division of MathSoft, Inc. and the Applied Physics Laboratory (APL) of the University of Washington. StatSci has an active wavelet and signal processing research group, and in 1994, released the S+WAVELETS software toolkit for wavelet analysis. APL is one of the leading research centers emphasizing Navy applications in ocean science, acoustics, and engineering.