Abstract: To avoid the subjectivity and the inefficiency in traditional metal industrial component detection by means of X-ray image, an intelligent detection method based on Gaussian Mixture Model(GMM) is put forward. Image sequence of the same component is learned online, and each pixel is composed of multiple Gaussian distribution. During ordinary process, the pixels are classified to different patterns by the learned Gaussian. During ordinary process, the pixels are classified to different patterns by the learned Gaussian distribution, and the pixels not belonging to any existed Gaussian distribution are regarded as perspective objects(damage points). The overall damage region is confirmed by connecting damage sector based on seed growth method. Experimental results show the method can locate the damage position in a component accurately and automatically. Furthermore, detection efficiency is increased obviously.

Key words: X-ray, Gaussian Mixture Model(GMM), image sequence, model learning , image sample, intelligent detection

摘要： 传统金属工业构件X射线图像检测手段主观性过强、检测效率低下。为此，提出一种基于高斯混合模型(GMM)的智能检测方法。对同一构件的图像序列进行在线学习，每一像素点由多个高斯分布分量组成。正常工作时对每一像素点用学习到的高斯分量进行模式分 类，若不符合任一现有高斯分量就视为前景目标(损伤点)，采用种子生长法连通损伤区域，确定整个损伤区域。实验结果表明，该方法可精确定位构件损伤部位，实现金属构件损伤的自动检测，检测效率较高。

关键词: X射线, 混合高斯模型, 图像序列, 模型学习, 图像样本, 智能检测

CLC Number: 

• N945