An Underwater Defect Instance Segmentation Method for a Bridge Pier Inspection Crawling Robot

摘要

The safety and stability of bridge piers, which are a crucial element of bridge structures, are of utmost significance due to the rapid advancement in bridge construction. Nevertheless, the intricate and imperceptible nature of the underwater environment makes inspecting the underwater section of bridge piers a challenging task. This study presents the development of a C2f-HG attention-depthwise separable convolution network (CADNet) for accurate identification and evaluation of defects in the underwater structure of bridge piers. The primary components of CADNet's infrastructure are the CADblocks and an SPPF module, which possess the capability of conducting multilevel and deep feature learning. The CADNet is employed in the advanced crawling robot for bridge pier inspection to enable the automated analysis and processing of high-definition underwater images. The advanced bridge pier inspection crawling robot uses the CADNet to automate the analysis and processing of high-definition underwater photos of bridge piers. The system can precisely detect cracks and spalling, effectively separating them from the complicated background. This provides inspectors with clear and easily understandable detection data. In the segmentation and detection tasks, CADNet achieved precision of 79.8% and 77.2% and recall of 78.7% and 76.6%, respectively. The use of the suggested underwater defect instance segmentation technique in the developed bridge pier inspection crawling robot holds significant practical importance and promising application potential.