Research On Surface Recognition Technology Of Weld Defects In Pipelines Based On Machine Vision

Weld defect detection is an important part of ensuring welding quality and safety.Traditional methods of in-pipeline weld defect detection mainly rely on manual inspection,which requires engineers with relevant professional knowledge and experience to enter the pipeline for inspection.However,due to the complex internal environment of the pipeline and the subjective nature of manual inspection,the accuracy of the results can be affected.In recent years,with the development of computer technology and image processing technology,the weld defect detection method based on machine vision has gradually become a research hotspot.In this paper,the defect recognition of stomata by machine vision method based on image processing is first studied.Image preprocessing is performed by filtering,image enhancement,image segmentation,and edge detection,and the Hof circle transform is used to detect the pores in the image.Because this method also depends on the size and shape of the defect,the recognition effect is relatively limited.Therefore,this paper studies the deep learning algorithm,constructs a new weld defect dataset,and segments and annotates the weld image.Aiming at the problem that the accuracy of the One-stage object detection algorithm is not high,the self-attention mechanism Co TNet is added to the YOLOv5 s model,and the Neck part is improved,and the feature fusion algorithm of Bi FPN-weighted bidirectional feature pyramid network structure is introduced to improve the detection performance.Through ablation experiments,it is concluded that when the attention mechanism and feature fusion algorithm are improved,the m AP value is 8.2% higher than that of the original YOLOv5 s model,and through comparative experiments,the m AP value of the improved YOLOv5 s network is 8.5% higher than that of the Faster-RCNN network in the Two-stage algorithm and22.3% higher than that of the SSD in the One-stage algorithm.Therefore,it shows that the improved YOLOv5 s model proposed in this paper can effectively improve the network performance and improve the detection accuracy of weld porosity and slag inclusion defects.Finally,this paper verifies the improved YOLOv5 s algorithm by building a weld defect detection system.In the pipe with an inner diameter of 325 mm,the algorithm is used to detect defects by building a weld defect detection system,and the porous defects and slag inclusion defects on the surface of the weld are successfully detected,which proves that the improved algorithm can accurately identify weld defects.