基于计算机视觉的钢结构螺栓病害智能检测方法研究OA

To address issues in steel structure bolt defect detection such as low efficiency,high missed detection and false judgment rate of manual visual inspection,poor detection accuracy,weak anti-noise capability,and large number of computational parameters of traditional deep learning models,this paper proposes a bolt defect detection model based on improved YOLOv8.The model improves the backbone using poly-scale convolution and enhances spatial feature extraction capability through multi-dilation rate convolution kernels.It introduces CSPStage and DySample modules to optimize the neck,strengthens the efficiency and quality of multi-scale feature fusion,and improves the model's anti-noise capability in complex scenarios.For dataset acquisition,DJI Mavic 4 Pro unmanned aerial vehicle was used to collect images of steel truss joint bolts from the Pingyin Yellow River Bridge.Combined with CLAHE preprocessing and data augmentation strategies to improve image data quality,a steel structure bolt defect detection dataset containing four states:rusty,loose,missing,and normal was constructed.Experimental results show that the improved YOLOv8-GDFPN model achieves a mean average precision(mAP)of 83.5%and an accuracy of 88.1%on the self-built dataset,which are 4.9%and 4.5%higher than the YOLOv8 model,respectively.The model has 3.263 M computational parameters and an FPS value of 181.8 frames per second,achieving a balance between detection accuracy and inference efficiency.This model can provide a reference for intelligent detection of bolt defects in complex outdoor scenarios.