期刊文献+

基于深度学习特征匹配的铸件微小缺陷自动定位方法 被引量:29

Automatic localization method of small casting defect based on deep learning feature
下载PDF
导出
摘要 针对射线实时成像检测中精密铸件微小缺陷自动定位的需要,提出一种基于深度学习特征匹配的铸件缺陷三维定位方法。模拟选择注意机制的中央-周边差算法,提出以视觉显著度为尺度,从射线图像复杂背景中检测出微小缺陷及其区域,以定义的区域中央点为待匹配点;然后,提出构造深度卷积神经网络自动提取微小缺陷区域的深度学习特征,通过深度学习特征矢量的相似度,实现在不同视角下投影图像中的同一微小缺陷点的自动匹配;最后,基于平移视差测距原理计算缺陷匹配点的三维空间坐标。实验表明,基于深度学习特征匹配的方法能够正确搜索平移前后投影图像中的同一缺陷点,以此为基础,利用视差测距原理实现了微小缺陷匹配点的自动准确定位,深度定位误差小于5.52%,能够满足对精密铸件微小缺陷智能评判的需要。 Aiming at the requirement of the automatic localization of precise casting small defect in radiographic real-time imaging detection, a three-dimensional localization method for casting defects based on deep learning feature matching is proposed. Simulating the central-peripheral difference algorithm of selective attention mechanism, taking the visual saliency as the scale, the small defect and its region are detected from complex background of the ray images. The central point defined in the defect region is taken as the point to be matched. Then the deep convolution neural network is constructed to automatically extract the deep learning features of the small defect region. Through the similarity of the deep learning feature vector, the automatic matching of the same small defect point in the projection images at different viewing angles is achieved. Finally, based on the principle of translation parallax distance measurement, the 3D spatial coordinates of the defect matching points are calculated. Experiment results show that the method of deep learning feature matching can correctly search the same defect point in the projection images before and after translation. On the basis of this feature matching, the automatic and accurate localization of the small defect matching point is realized using the principle of parallax distance measurement. The depth localization error is less than 5.52% , which can meet the requirement of the small defect intelligent evaluation in precise casting.
出处 《仪器仪表学报》 EI CAS CSCD 北大核心 2016年第6期1364-1370,共7页 Chinese Journal of Scientific Instrument
基金 重庆市基础与前沿研究计划(cstc2013jcyj A70009) 国家自然科学基金青年基金(51075419)项目资助
关键词 射线图像 缺陷检测 深度学习 自动定位 神经网络 radiographic image defect detection deep learning automatic localization neural network
  • 相关文献

参考文献11

二级参考文献122

  • 1郑世才.射线实时成象检验技术[J].无损检测,1996,18(7):202-206. 被引量:17
  • 2同济大学数学教研室.高等数学[M].北京:高等教育出版社,1987..
  • 3刘发鸿 杜玉金.射线体视检测原理及其应用[J].陕西机械学院学报,1985,(4):23-37.
  • 4洛班诺夫AH.航空摄影测量地形学[M].北京:测绘出版社,1983.139-150.
  • 5王之卓.摄影测量原理[M].北京:测绘出版社,1985.20-38.
  • 6诸梅芳.计算方法[M].北京:化学工业出版社,1989.55-70.
  • 7张勤俭.射线三维检测数学模型的建立及实验研究[M].西安:西北纺织工学院,1998..
  • 8Bikash Ghose, Kankane D K. Estimation of location of defects in propellant grain by X-ray radiography[J]. NDT&E International, 2008(41): 125-128.
  • 9Sun C, Tang G, Wang Z. Realization and application of simulation technology in Direct Radiography [C]. CD-Proceedings of 17th World Conference on Nonde- structive Testing. Shanghai: Shanghai Chinese Society of Nondestructive Testing, 2008 : 25-28.
  • 10WANG Z F, ZHENG ZH G. A region based stereo matching algorithm using cooperative optimization [ C ]. Computer Vision and Pattern Recognition. CVPR 2008. 2008 : 1-8.

共引文献231

同被引文献288

引证文献29

二级引证文献391

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部