Automatic target recognition in synthetic aperture sonar images for autonomous mine hunting

The future of Mine Countermeasures (MCM) operations lies with unmanned platforms where Automatic Target Recognition (ATR) is an essential step in making the mine hunting process autonomous. At TNO, a new ATR method is currently being developed for use on an Autonomous Underwater Vehicle (AUV), using SAS images of known targets in an operational environment as input. The focus is set on high resolution Synthetic Aperture Sonar (SAS) input images to follow modern sensor developments. This paper discusses preliminary results achieved with both trial datasets and simulated data. Various preprocessing methods, feature sets and classification algorithms are compared on the basis of their classification performance. We conclude that, even though fine-tuning should further improve the results, the performance of the developed ATR processing chain is already quite encouraging