Anomaly detection in Sonar images: application of saliency filters

Sea mines are still a concrete menace both for military and civilian ships and detecting them is necessary to ensure the safety of the navigation. In this work we explored the possibility of automatically detect mines by using unmanned Autonomous Underwater Vehicles equipped with Side Scan Sonar (SSS) Sensors. To accomplish the detection task, we considered saliency detection algorithms coming from RGB and radar fields to highlight the mines with respect to the background. The algorithms were tested on a valuable dataset of images collected by the Italian Navy under operational conditions during several activities conducted in the Mediterranean Sea. We evaluated the performance according to broadly used performance indices such as ROC curves and MAE scores. Furthermore, a new performance analysis score called FAR@95%Pd is presented.