Underwater vision-enhanced image segmentation for supporting automated inspection of underwater bridge components

Abstract

Vision-based robotic systems for automated bridge inspection are limited in analyzing underwater inspection images, which present a set of unique visual challenges caused by light scattering, light attenuation, and low-light conditions in underwater environments. To address this limitation, this paper proposes an underwater vision-enhanced image segmentation method: (1) underwater vision-based quality enhancement is proposed to simultaneously mitigate quality degradations of underwater inspection images caused by light scattering, light attenuation, and low-light conditions; and (2) semantic segmentation is proposed to analyze quality-enhanced underwater images to localize bridge components, enabling effective component localization for subsequent damage detection and characterization in underwater inspection images. Baseline and ablation experiments were conducted for performance evaluation. The results showed that the proposed method achieved a mean, structure, and background IoUs of 91.7 %, 88.5 % and 94.8 % – outperforming state-of-the-art methods in segmenting underwater inspection images and demonstrating its potential to enable vision-based robotic systems for cost-effective underwater inspection.