The generation of three-dimensional (3-D) images and map building are essential components in the development of an autonomous underwater system. Although the direct generation of 3-D images is more efficient than the recovery of 3-D data from 2-D information, at present for underwater applications where sonar is the main form of remote sensing, the generation of 3-D images can only be achieved by either complex sonar systems or with systems which have a rather low resolution. In this paper an overview is presented on the type of sonar systems that are available for underwater remote sensing, and then a technique is presented which demonstrates how through simple geometric reasoning procedures, 3-D information can be recovered from side scan-type (2-D) data. Also presented is the procedure to perform map building on the estimated 3-D data.
Optimizing the Sediment Classification of Small Side-Scan Sonar Images Based on Deep Learning