Unsupervised detection of mine-like objects in seabed imagery from autonomous underwater vehicles

摘要

Autonomous image processing of sonar images from stable underwater platforms such as autonomous underwater vehicles (AUVs) provides a means of rapidly detecting mine-like objects on the seabed, while avoiding the delays and human demands associated with manual processing. The Defence Science & Technology Organisation has developed software using an unsupervised processing technique to detect mine-like objects in high-resolution sidescan sonar images. The software enables the user to process large volumes of data from AUV operations and report detection results. In the present study, the software detected 86% of mine-like objects in the imagery, with 0.13 false alarms per image (approximately one false alarm per eight minutes of survey). The results and analysis provide insight into the reasons for non-detections and false alarms, and strategies for improving the object detection performance. These techniques are suitable for application in post-processing of AUV data, for on-board processing applications and for the prediction of performance in the detection of objects on the seabed.