EVALUATING HUMAN/MACHINE INTERACTION FOR UNDERWATER THREAT CLASSIFICATION: A CASE STUDY

In this paper, expert deminers and automated algorithms are charged with the task of analysing sonar images collected during real mine countermeasures exercises in order to classify targets. Images are collected using synthetic aperture sonar (SAS) and side scan sonar (SSS), covering a test area on the Belgian Continental Shelf. A total of 1241 images (with 847 detection opportunities) collected from different sonar systems, each of them covering the entire area, are used. Image resolution is divided in three categories: (1) up to 5cm pixel size, (2) over 5cm until 10cm pixel size, (3) larger than 10cm pixel size. Data are analysed in different ways by the expert operators and the algorithms. Results demonstrate how challenging underwater threat recognition still is, and highlight the utility of considering the human operator as an integral part of the automatic underwater object recognition process, as well as how automated algorithms can extend and complement human performances.