基于机器视觉的铸件缺陷动态检测方法  

Dynamic detection method of casting defects based on computer vision

作　　者：畅凡 陈富民[1] 唐凯旋 段童童 CHANG Fan;CHEN Fumin;TANG Kaixuan;DUAN Tongtong(School of Mechanical Engineering,Xi'an Jiaotong University,Xi'an Shaanxi 710049,China;Xi'an XD Switchgear Electric Corporation Limited,Xi'an Shaanxi 710077,China)

机构地区：[1]西安交通大学机械工程学院,西安710049 [2]西安西电开关电气有限公司,西安710077

出　　处：《计算机应用》2021年第S02期258-264,共7页journal of Computer Applications

摘　　要：针对数字化X射线实时成像系统自动化检测问题,提出了一种基于机器视觉的铸件缺陷动态检测方法。首先,通过计算射线图像的行灰度曲线的波峰波谷快速检测出缺陷帧;然后,利用帧间差分法获得前景区域即缺陷区域,融合边缘检测算法与形态学方法在缺陷区域中准确分割出缺陷;最后,使用卡尔曼滤波器与匈牙利算法获得缺陷在连续多帧图像中的位置,通过挖掘缺陷在多帧图像中的位置关联关系修正抑制错检并修正漏检,提升了正确率。实验结果表明,与人工检测和基于单幅图像的缺陷检测方法相比,所提方法显著提高了缺陷检测的正确率与效率。In order to realize automatic detection using digital X-ray real-time imaging system,a dynamic detection method of casting defects based on machine vision was proposed. Firstly,the defect frames were quickly detected by calculating the peaks and valleys of the line gray curve of the radiographic image. Secondly,the inter-frame difference method was used to obtain the foreground area which was the defect area. Then,the edge detection algorithm and the morphological method were merged to accurately segment the defect in defect area. Finally,the Kalman filter and the Hungarian algorithm were used to obtain the position of the defect in the continuous multi-frame images,and the missing detection was corrected and the wrong detection was suppressed by mining the positional relationship of the defect position in the image sequences,the accuracy was improved. Experimental results show that,compared with manual detection and defect detection methods based on single image,the proposed method has significantly improved the accuracy and efficiency of defect detection.

关 键 词：X射线实时成像 机器视觉 缺陷分割 错漏检修正 无损检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]