【作者】 王静；

【导师】 原培新；

【作者基本信息】 东北大学 ， 机械电子工程， 2008， 硕士

【摘要】 随着X射线检测和图像处理技术的发展,焊缝缺陷的检测也逐步从人工评片过渡到计算机智能识别。利用计算机对数字化焊缝图像进行分析和识别在检测效率、经济效益、方便实用等方面得到人们的认可。本课题以实际工程需求为背景,以焊缝X射线检测底片为研究对象,综合运用图像处理的方法给出焊缝故障的定量及定性描述,应用Visual C++建立基于统计决策树的焊缝图像识别系统,实现故障类型的自动评判。根据焊缝图像的特点,图像处理系统主要内容有降噪滤波、图像增强、边缘检测和图像分割。通过对各种图像处理方法的分析比较,选择自适应中值滤波方法对原始焊缝图像进行降噪滤波；再引入了直方图均衡化以及模糊增强等方法对图像作增强处理；在边缘检测方面,在梯度算子中引入遗传算法理论提取焊缝的边缘；最后通过对图像分割方法中的基于阈值分割方法、边缘检测方法及数学形态学分水岭方法的分析讨论,选择类间、类内方差比分割法和数学形态学方法并用进行焊缝图像分割,以提取出有效的焊缝区域。为了更好的进行图像识别,又对图像分割后的图像进行缺陷标记和缺陷跟踪；其次特征参数的选择是缺陷识别的前提条件。本文通过对缺陷成像特点的分析,确定能够反映缺陷本质特征的特征参数,最后使用统计决策树的方法对缺陷进行识别和分类。更多还原

【Abstract】 With the development of X-ray detection and image processing technology, welding defects detection gradually transits from artificial detection to computer intelligence recognition. Analysis and recognition digitized welding image by computer is approved by the people in detection efficiency,economic profits,convenient and pragmatic, etc.The paper is based on the actual project demand and takes the X-ray detection film as the research object, useing the method of image processing give the quantitative and qualitative description of welding defects, applied Visual C++ build up welding image recognition system based on the statistics decision tree, carry out the assessment automatically.According to chatacteristic of the weld image, divided the image processing system into noisy reducing, image enhancement, edge detection and image segmentation. After various methods of image processing are compared, the method of the adaptive median filtering is used for decreasing the noise and filtering; the methods of the histogram equalization and the blur enhancement are introduced to make the image enhancement; in the part of the edge detection, the genetic algorithm is introduced to the gradient operators to extract the edge of the welding; at last, the threshold segmentation, edge detection and mathematical morphology of image segmentations are compared, the variance ratio of interclass and intraclass segmentation, and mathematical morphology are used for the image segmentation of the welding to extract the effective area of the welding.After the image segmentation, multi-defect tracking and filling are designed for carrying out image recognition; Choosing feature parameter is a premise of the defects recognition. The paper determines the feature parameter which can reflect the nature of the defects by analyzing the image characteristic of the defects.at last, the defects recognition and classification are realized by the tree classifier.更多还原