融合视觉感知机理与知识模型的射线检测缺陷智能识别技术

项目名称： 融合视觉感知机理与知识模型的射线检测缺陷智能识别技术

项目编号： No.51205265

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 机械工程学科

项目作者： 殷鹰

作者单位： 四川大学

项目金额： 25万元

中文摘要： 射线检测图像中缺陷识别的视觉感知、几何特征分析与基于知识的判定之间存在密切的联系，基于知识模型的协同缺陷识别是促进射线检测技术发展的关键问题之一。本项目针对机械产品射线检测图像中微小缺陷智能识别所面临的科学问题，探讨射线检测领域专家对检测图像中缺陷识别的决策原理和模式，研究融合仿人视觉感知机理与知识模型的射线检测图像中微小缺陷智能识别技术。提出一种基于注意机制的提取边缘、局部、拓扑关系特征使大背景图像中微小缺陷目标凸显的仿人视觉感知模型，建立描述缺陷形成机理、缺陷特征、检测知识、评判规则、缺陷与产品关联背景知识等内容所组成的缺陷识别知识模型，形成为一种新的基于图像处理算法、仿人视觉感知与知识模型的微小缺陷识别推理和分类决策方法，开发出射线检测图像中缺陷智能识别软件原型系统并在企业应用验证。研究成果可为解决目前机械产品射线检测图像中缺陷识别还主要依靠人工识别评定问题提供一种新的技术方法。

中文关键词： 射线检测图像；智能识别；仿人视觉感知；缺陷检测；图像分割

英文摘要： There is a close link between visual perception, analysis of geometric features and decision-making based Knowledge for defect identification in radiographic detection images. The collaboration defect identification based knowledge model is one of the key issues to promote the development of radiographic detection technology. For the scientific problems in the intelligent identification defects of radiographic detection images in mechanical products, the decision-making principle and mode of the defect identification in the detection images by radiographic testing experts is explored. The intelligent identification technology of small defects in radiography image is researched, and which is integrated the humanoid visual perception model and the knowledge model. A humanoid visual perception model is proposed based on attention mechanism of highlighting small defects by edge extraction, local, and topological relation features in the large background image. The defect identification knowledge model is build up, which is consisting of the defect formation mechanism, defect characteristics, inspection knowledge, judgment rules and associating products background knowledge. A new small defects identification reasoning and classification decision-making method is developed based on the image processing algorithms, th

英文关键词： Radiographic testing image；Intelligent recognition；Humanoid visual perception；Defect Detection；Image segmentation