Robust classification of sector-scan sonar image sequences

The paper reports the development of a system for the automated identification of sonar targets. It uses a novel set of feature measures, derived from sequences of sonar scans, to characterise the behaviour of targets over time. Previous research used grey-scale and shape descriptors derived from images of single sonar scans. However, problems were experienced with targets whose return varied significantly over time (such as divers, ROVs, and ships' wakes). Hence a novel set of temporal feature measures have been developed, to provide a quantitative description of a sonar target's behaviour over several sonar scans. The research showed that the classification accuracy was significantly improved by the use of these new features. While the use of "static" feature measures (derived from a single scan) gave classification errors between 20% and 40% for certain targets, the use of temporal measures reduced this error rate to zero in some cases.<
>