Abstract
A system for classifying objects in sector-scanning sonar images which uses temporal feature measures to provide a quantitative description of a target's behaviour over several scans is described. Results show that this method provides better classification accuracy than analysing single images.