Seafloor Sediments Classification of Side-scan Sonar Images Based on Dual-path Network

Abstract

Seafloor sediments classification is of great significance in underwater remote sensing and ocean engineering. Although previous researchers have contributed greatly to the seafloor sediments classification, many have neglected the effect of the noise present in sonar images on sediment classification. In order to improve the classification accuracy of seafloor sediments, a classification model of seafloor sediments is established using side-scan sonar images. First, we denoise the images using median filtering. Then we build a novel dual-path network for seafloor sediments classification of side-scan sonar images. The network has two paths with different kernel sizes, which can extract features of seafloor sediment images at different scales and thus improve the classification accuracy of the network. The experimental results show that the proposed seafloor sediments classification model trained using the filtered images has better classification performance.