Sonar scattering from the sea bottom near the Norwegian coast

Abstract

High false alarm rates on active sonar systems in shallow waters is a well known problem which may be a limiting factor for the sonar performance. One way to reduce the false alarm rate is through supervised learning and machine learning algorithms applying relevant classification features. Here we propose backscatter intensity as a classification feature. It is calculated from a topographical map under the assumption of omnidirectional scattering for sound, an unrefracted path and iso-velocity sound speed profile. Analysis of historic data for a towed array sonar near the Norwegian coast showed that clutter echoes to a large extent was located at cliffs, escarpments and ridges where the calculated backscatter intensity was high. With a simple thresholding of the estimated backscatter intensity one can correctly classify 60% of terrain echoes at the expense of 10% potential target echo misclassification rate. The classification rate rises to 75% for the echoes of highest signal-to-noise ratio. The proposed classification feature is computationally inexpensive and does not depend on free parameters.