Fine-Grained Segmentation of High-Resolution Bridge Crack Images Using Rendering Technology

Abstract: Drawing on insights from computer graphics, this study introduces the Crack Boundary Point Rendering Network (CBPRN), an innovative high-resolution (HR) crack segmentation framework designed to improve UAV bridge inspections. We developed an edge-guided branch and an uneven sampling strategy, enhancing detail preservation and directing computational resources toward critical crack boundary areas effectively. Through comprehensive ablation experiments, the efficacy of the CBPRN was validated, demonstrating its superior performance with remarkable outcomes for images beyond 2K resolution. The CBPRN establishes a new standard in HR crack image segmentation.