Low-Cost Mechanical Sonar Mapping With Artifact Removal in Confined Spaces

Exploration of confined underwater environments— such as polar sub-ice regions, submerged caves, shipwreck interiors, flooded water-conveyance tunnels and enclosed test tanks—has garnered increasing research interest. Conventional sensing modalities, including optical imaging, passive acoustics, and tethered active sonar systems, encounter substantial limitations in such settings. Low-cost mechanically scanning imaging sonar (MSIS) systems offer a promising alternative. However, MSIS measurements are severely affected by multipath artifacts inherent to acoustic propagation in enclosed spaces, which obscure true structural boundaries and pose significant challenges to robust underwater perception. To address these challenges, a complete MSIS-based mapping framework is developed for confined environments. By reconstructing physically interpretable angle-range representations from discrete time-domain echo signals parsed directly from the MSIS serial stream, we propose two boundary extraction strategies: an adaptive thresholding-based method and a lightweight deep segmentation network termed MSIS-Net. To eliminate multipath artifacts and recover true structural contours, a novel graph-based acoustic front extraction algorithm is developed, leveraging acoustic propagation priors for directionally guided path tracing. Experimental validation is conducted in real-world tank environments of varying scale, demonstrating the robustness and decimeter-level accuracy of the proposed framework. In support of further research, we release the Confined-MSIS Dataset, the first open-access dataset tailored to confined-space acoustic mapping using MSIS.