A novel visual classification method of seabed sediments

This study aims at the autonomous seafloor surveillance by underwater vehicles based on computer vision techniques. A novel scheme of seabed image classification is proposed to identify three types of seabed sediments. The texture features of seabed sediments were described by using gray-level co-occurrence matrix and fractal dimension. Subsequently, an unsupervised learning method, Self-Organizing Map, was applied to analyze the seabed images with the extracted texture features. The experimental results demonstrated that the proposed texture feature descriptors were feasible and effective to category the three types of seabed images.