Application of an Object Classification Method for Determining the Spatial Distribution of Sea Bottom Structures and Their Cover Using Images from a Side Scan Sonar

Abstract

Side Scan Sonars combined to positioning systems constitute an exceptionally valuable tool for collecting information about seabed characteristics. The main objective of this study is to apply the new methods of object classification, based on Side Scan Sonar images for determining the bottom structure and coverage. The aim is to adapt land image analysis for acoustic data and to produce a reliable model that can be applied in further analysis. Software development tailored for this type of data enables wide-range analysis and visualization of spatial data. The main tool used in postprocessing of sonar images was the Trimble eCognition™ Developer. The classification process is a categorization of objects and phenomena in the analyzed area. Classification uses information contained in a single pixel creating more general scheme. An automated mode of the classification method is used in this study OBIA (Object-Based Image Analysis). Classification is based on the segmentation process, in which pixels are divided into homogeneous regions with respect to neighboring pixels, which determines the final shape of each class object. Objects and parameters, controlling the processing of supervised classification, are selected by user. Representative functions of every object are defined by the user to depict each class, helping to assign them to specific classes. The adopted approach for mixed type of classification processes, is based on the reference objects and training samples and subsequently classifies them by class discriminants. As a result a sonar mosaic map of the seabed image classes is provided in relationship to the bottom sediments and the bottom surface roughness. The presented results could help to transform a protected area HELCOM BSPA into the undersea sanctuary in the Gulf of Gdansk (Baltic Sea) and support this very important interdisciplinary environmental study. In this paper, I have demonstrated that a multistage image segmentation methodology can be more efficiently used for geomorphometry and terrain classification for underwater areas obtained from a side scan sonar mosaic. The methodology introduced in this paper can be applied for many other applications where relationships between topographic features and other components are to be assessed (samples etc.). In this way, we hope to address a time-consuming task to produce underwater maps with bigger scales but still with an appropriate resolution. © Springer International Publishing Switzerland 2014.