System of detecting underwater threats in side scan sonar images

This paper proposes a promising new approach to detect underwater threats in side scan sonar (SSS) images without machine learning procedure. Although object detection requires high reliability, the maritime environment changes unpredictably and dynamically. In order to accomplish high reliability for object detection systems, a huge number of the samples under various different environments are required to be learned. However, samples of underwater objects are quite hard to be collected. For this reason, we have already proposed a promising method called “One Shot Detector (OSD)”, which accomplishes object recognition without any machine learning procedure. In this paper, we show that OSD is effectively applicable for detecting underwater threats in SSS images. Finally, through the comparison to a promising previous approach, we show that our approach is suitable under the condition that any machine learning procedure cannot be used.