A Domain-Adaptive Visual Enhancement Approach for Underwater Simultaneous Localization and Mapping

Objectives To address the problems of unstable visual features, insufficient trajectory continuity, and localization drift in close-range structural scenes for underwater robots operating in weak-texture, low-illumination, and suspended-particle scattering environments. Methods A multi-sensor tightly coupled underwater simultaneous localization and mapping (SLAM) method based on domain-adaptive visual enhancement is proposed. In the visual frontend, UFEN is fine-tuned via teacher–student distillation on HoloOcean simulated underwater images, and sign-bit quantization converts 256-dimensional floating-point descriptors into 32-byte binary descriptors. Visual reprojection, IMU preintegration, and joint DVL–IMU preintegration constraints are incorporated into local factor graph optimization. Results On the eight sequences of the TANK dataset, the proposed method achieves the lowest absolute trajectory error root-mean-square error on seven sequences. Compared with the AQUA algorithm, the localization error is substantially reduced by 30.3%–84.2% in degraded scenarios such as low illumination, medium-to-high-speed motion, and large-scale loop trajectories, and the accuracy gain progressively increases with the severity of image degradation. Conclusions Experimental results demonstrate that the proposed method, through domain-adaptive fine-tuning in the frontend, effectively mitigates the degradation of visual constraints caused by underwater environments, significantly enhancing trajectory continuity and scale stability in complex large-scale scenes. This provides a highly reliable multi-sensor tightly coupled SLAM solution for underwater close-range structural inspection.