Robust Semiautomated Image Recognition for Laminated Rubber Bearings of Bridges in Complex Environmental Conditions

Laminated rubber bearings in bridges often face environments with occlusion, insufficient illumination, and clutter from nearby objects, which makes it challenging to recognize the bearing components from images. To address this issue, this study proposes a semiautomatic bearing identification method that integrates multilevel filtering with interactive manual point selection. First, multiobject segmentation is performed on the bearing images to obtain multiple regions, including the bearings. Oversegmented regions are then merged to refine the segmentation results. Subsequently, the segmented regions undergo progressive filtering based on spatial location, shape, texture, and color features. This multilevel filtering strategy effectively excludes interfering regions. Finally, an interactive manual point selection mechanism is introduced to assist in handling extremely complex scenarios. Experimental analysis of 400 real-world bridge bearing images confirms effective identification of bearing components under standard conditions using the fully automated method. This approach achieves 0.815 precision and 0.663 recall while reducing workload by approximately 70%. However, recognition performance deteriorates significantly in complex environments with severe occlusion, extreme illumination, or critical defects. Under these conditions, precision and recall drop substantially to 0.226 and 0.236. Introducing interactive manual point selection addresses such challenging scenarios effectively. The method elevates precision to 0.760 and recall to 0.849, demonstrating substantial performance improvements. The robust image recognition method of laminated rubber bearings could reduce manual recognition workload during the inspection of bearings in complex field conditions.