Advances in shallow water seabed classification: acoustic, optical, and multi-source data fusion

Shallow water seabed classification plays a critical role in marine resource development, ecological conservation, and coastal engineering. This paper provides a comprehensive narrative review of recent advances in this field, with a particular focus on three core technical domains: acoustic techniques, optical techniques, and multi-source data fusion. To provide an integrated perspective beyond modality-specific reviews, the literature is organised through a unified data–method framework and an evolution-oriented synthesis from point-based sampling and conventional feature analysis to area-wide mapping, learning-based classification, and multimodal integration. Acoustic methods provide broad spatial coverage and remain applicable under turbid-water conditions, but classification performance is affected by acoustic scattering ambiguity, system configuration, and observation scale. Optical methods offer high-resolution spectral and visual information in optically shallow waters, while water-column attenuation, turbidity, illumination variability, and limited penetration constrain their applicability. Multi-source fusion can exploit complementary acoustic, optical, bathymetric, environmental, and in situ information, but its effectiveness depends on accurate spatial registration, scale consistency, data quality, and robust validation; uncertainty propagation across sensors and processing stages remains a key challenge. Recent advances in deep learning, self-supervised learning, small-sample learning, and physics-guided fusion are extending feature representation and integration strategies, yet limited labelled data and cross-region transferability remain unresolved. Future work should prioritise robust multi-source integration, uncertainty-aware validation, dynamic monitoring, and standardised classification and evaluation frameworks.