Perception of the seabed environment is an important capability of autonomous underwater vehicles. This paper focuses on defining and extracting robust texture features from visual images that lead to useful and practical automated identification of the types of seabed sediments. The visual texture features are described by using a gray-level co-occurrence matrix (GLCM) and fractal dimension, after which an unsupervised learning method, self-organizing map (SOM), is adopted to evaluate the validity of features descriptors on three types of seabed sediments. Subsequently, a kernel-based approach that exhibits robustness versus low numbers of high-dimensional samples, named support vector domain description (SVDD), is applied to classify the types of seabed sediments. In comparison with state-of-the-art classifiers, the experimental results demonstrated the effectiveness of the SVDD on the classification of seabed sediments. •The visual images of seabed sediments are characterized by the texture features which are extracted based on the GLCM and fractal theory.•A multi-class classification strategy for seabed sediments is proposed by adding a distance measure after SVDD implementation.•The experimental results demonstrate that the proposed classification strategy is feasible in recognizing the type of seabed sediments.