Wavelet-Based Detection and Classification for Shallow-Water Impulsive-Source Sonar

In recent years the anti-submarine-warfare mission of the U. S. Navy has been expanded to include operations against stealthy, diesel-electric submarines operating in near-shore, shallow water regions of the world. Acoustic conditions in such regions are characterized by significant spatial and temporal variability. This variability reduces the performance of traditional sonar detection and classification algorithms. Impulsive-source active sonar systems can provide sufficient acoustic energy to detect these targets, but also produce significant clutter in a shallow--water environment. Detection of too many false targets can quickly overload sonar operators, so that true enemy targets are not detected and properly classified. Our proposed research will use a combination of wavelet and nonparametric bootstrap methods to achieve improved performance for detection and classification. Bootstrap methods allow one to adaptively estimate the underlying probability distribution of interference in the non--Gaussian colored-noise environment typical of shallow-water active sonar, and to set detection thresholds to maintain a desired false alarm rate. Subsequent target classification will be performed using wavelet representations of the sonar echoes, taking advantage of the time--frequency structure of targets and clutter. These representations may be combined with existing discriminant analysis methods. This work will demonstrate the feasibility of these techniques for sonar target detection and guide our phase II development of improved methods for detecting and classifying targets in the presence of clutter.