Automatic classification for MCM systems

Abstract

Sonar systems, designed to detect stealthy mines, will now also hand over to the classification process many more small non-mine bottom objects (NOMBO), with mine-like target strength. In these conditions the global search effort is transferred to the classification function by reducing the number of pointless false detection alarms, i.e. by classifying a NOMBO as a non mine like contact (NON MILCO). This paper then aims at giving demonstrative results and means to perform automatic classification while keeping low false classification rate. The considered method is divided into two steps: a first coarse long range pre-classification only provides relevant mine like echoes (MILEC) to be then processed by a shorter range classification. The false alarm reduction of the first step is performed with a long range detection sonar by analysing the echo signal structure and its evolution given the sonar motion. Then, the second step is image based classification and needs a high frequency classification sonar to provide a more detailed acoustic image of the object (especially concerning its cast shadow shape).