Profile Fitting-based Small Target Detection in Water for Side-scan Sonar Image

Abstract

Side-scan sonars have seen wide deployment in underwater imaging, but the substantially lower visual quality of sonar images leads to a few questions, such as bright spots-like small target with a few pixels, the difficult distinguish of bright spot and speckle noise. In order to solve this problem, we propose a two-stage detection method without obtaining sample information in advance. In the first stage, we fit across-track signal to obtain the peak points of the signal profile, which indicates the position where the target or shadow may be located at this across-track signal. In the second stage, the disordered peak points are clustered using a tree-growing clustering method. A cluster represents the place where the target or shadow may appear in the side-scan sonars image. The target is then detected by matching lights and shadows with possible locations. Experimental results verified that the proposed method is efficient in small target detection from lower visual quality images.