基于深度学习的焊缝X射线图像缺陷识别  

Weld X-ray Image Defect Recognition Based on Deep Learning

作　　者：程希莹 何云凯 吴晗宇 CHENG Xiying;HE Yunkai;WU Hanyu(College of Architecture,Anhui Science and Technology University,Bengbu Anhui 233030,China;Key Laboratory of Nondestructive Testing(Nanchang Hangkong University),Ministry of Education,Nanchang Jiangxi 330063,China)

机构地区：[1]安徽科技学院建筑学院,安徽蚌埠233030 [2]无损检测技术教育部重点实验室(南昌航空大学),江西南昌330063

出　　处：《淮阴师范学院学报（自然科学版）》2024年第4期319-328,共10页Journal of Huaiyin Teachers College(Natural Science Edition)

基　　金：安徽省教育厅自然科学重点项目(2024AH050329,2023AH051881);安徽省社会科学创新发展研究课题(2023KY008);国家级大学生创新训练项目(S202210879006);南昌航空大学大学生创新创业训练项目(2023ZD215)。

摘　　要：随着工业制造对焊接质量要求的日益提升,焊缝缺陷的精准检测与识别已成为保障结构安全的重要环节.传统检测方法在处理复杂焊缝图像时效率低下,且准确率有限.本文提出一种基于X射线图像的焊缝缺陷识别方法,聚焦于缺陷特征提取和随机森林模型的应用。通过图像预处理技术,提升图像质量并准确提取焊道区域;随后,采用多种特征提取方法对缺陷进行分类识别,构建并优化随机森林模型.实验结果显示,该模型在识别裂纹和气孔缺陷方面表现优异,准确率分别达到91.0%和84.6%,显著优于传统的BP神经网络方法.本研究为焊缝质量评估提供了一种高效、可靠的解决方案,具有广泛的工业应用前景,能够有效提升焊接工程的安全性和可靠性。With the increasing requirements for welding quality in industrial manufacturing,accurate detection and identification of weld defects become the important parts to ensure the safety of the structure.Traditional detection methods are inefficient and inaccuracy in processing complex weld images.This paper proposes a method of weld defect identification based on X-ray images,with a focus on the defect feature extraction and the application of random forest model.We use image preprocessing to improve image quality,and accurate extraction of weld bead areas.Subsequently,by using various feature extraction techniques to defect classification and identification,we construct and optimize the random forest model to evaluate its performance.The experiment shows that the model performs excellently in identifying cracks and pores,with accuracy rates of 91.0%and 84.6%,respectively,which is superior to the BP neural network.This study provides a comprehensive and effective solution for the evaluation of weld quality,which is of great significance to welding engineering.

关 键 词：焊缝缺陷识别 随机森林模型 X射线检测 阈值分割 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]