The U.S. Army Corps of Engineers and the Navy have identified more than 400 underwater sites that are potentially contaminated with munitions, and most are in shallow water posing a threat to human safety and the environment. Similar to dry land UXO, successful and economical remediation requires a well-coordinated combination of sensor technologies, and effective classification algorithms to detect the munitions but with minimal false alarm rate.

Our objective is to improve low-frequency sonar detection and classification of buried underwater munitions by developing new features that are highly diagnostic of target shape and composition. These features are T-matrix coefficients, and for the case of single or multiple targets embedded in 1D layered media, they completely characterize the target scattering response. A key innovation is our method to invert for these coefficients from experimental observations. Ordinarily, they are obtained via numerical methods which, due to numerical instabilities, usually fail for targets with aspect ratios greater than 2:1. Our method bypasses this problem, and is potentially higher fidelity as well since the coefficients are obtained for an actual target exemplar rather than an idealized CAD model. The reference coefficients for known targets (obtained from laboratory data) and the estimated coefficients from field data form the basis of our proposed classification algorithm. These features augment image-based features derived from 3D beamformed Buried Object Scanning Sonar (BOSS) data used in the SERDP effort MR-1533, and the combined image-morphology and physics-based feature set will yield much improved classification performance. Finally, the 1D medium inversion in combination with our green function formalism will enable higher resolution beamforming for improved image quality.

The objectives are achieved by (i) processing of free-field and buried-target acoustic field data previously collected by the Naval Research Laboratory for various exemplar targets; (ii) application of an innovative inversion algorithm to this data to obtain high-fidelity estimates of the target T-matrix coefficients; (iii) development and application of a 1D inversion algorithm to estimate the sound speed and sediment bottom properties (necessary for field-estimation of the coefficients); (iv) incorporation of the new feature class into our relevance vector machine classifier for BOSS data that previously used image-morphology features alone; and (v) demonstration of the new classifier with the FY08 multi-sensor database (MSDB) collected by the Naval Surface Warfare Center Panama City Division (NSWC PCD) under SERDP effort MR-1507, and for which the BOSS system was used to collect the sonar component of the data.

The benefit to the DoD user community is a new processing scheme that will reduce current false alarm rates while using currently deployed low-frequency systems such as the BOSS system. The scientific benefits are (i) the development of a new experimental database with key, physics-based features available for use in classification schemes; (ii) a new estimation methodology to estimate these coefficients from either lab or field data; (iii) the ability to perform rapid, high-fidelity forward simulations of the acoustic field using the coefficients; (iv) a 1D background model inversion scheme for higher accuracy beamforming and use of (iii); (v) a new advanced classifier; and (vi) a demonstration of the advances using the FY08 MSDB BOSS data collect.