In this paper we develop a novel classification method for multibeam sonar images based on the Weyl transform. The texture descriptor based on Weyl coefficients describes effectively the multiscale correlation features appearing in the sonar images. Our classification approach combines the Weyl coefficients with statistical features that are commonly used in the analysis of seabed sonar images and captures the morphological variation and geoacoustic characteristics of the seafloor. We employ a neural network as a classifier. The proposed combined feature extraction method demonstrates better performance than the commonly used statistical methods in this application.