基于改进YOLOv8面向无人机巡检的起重机损伤检测研究OA

Crane key structures are constantly exposed to complex environments characterized by elevated temperatures,high humidity,and severe corrosion,rendering them susceptible to surface defects such as cracks and peeling that pose significant threats to construction safety.In light of the limitations inherent to traditional inspection methods,including low efficiency,poor real-time performance,and unstable accuracy,a lightweight and efficient crane surface defect detection model for UAV inspections,designated YOLOv8-CDS,was proposed.Built upon the YOLOv8 framework,this model incorporated three structural enhancements:the original SPPF module in the backbone was replaced with SPPELAN to strengthen multi-scale feature fusion;a Dynamic Head detector was integrated to improve recognition of weak-textured targets;and a lightweight fusion module,CCFM,was incorporated to reduce computational complexity for edge deployment.Experimental results demonstrate that YOLOv8-CDS achieves identification accuracies of 95.7%for weld lines,91.5%for corrosion-related defects,and 89.5%for cracks.The overall mAP@0.5 score reaches 85.3%,with the model's inference speed attaining approximately 163 FPS,a computational load reduction of roughly 6.2%compared to the original YOLOv8 model.These performance characteristics satisfy the real-time requirements of UAV inspections and exhibit substantial industrial application value.