【作者】 禹建东；

【导师】 孔月萍；

【作者基本信息】 西安建筑科技大学 ， 计算机应用技术， 2008， 硕士

【摘要】 针对目前自动化方法在带钢表面缺陷检测时准确度不高的问题,本文探讨了基于计算机视觉的智能检测系统总体设计方案及软、硬件构成,重点设计了其中的缺陷初检、分割和识别步骤解决方案。针对实际生产中缺陷出现概率较低的现状,利用对图像分块差值图提取的灰度范围、方差等特征量,构造了缺陷初检判别函数,以提高检测算法的实时陛。对初检含缺陷的图像,因光照不均匀而引起传统的单一阈值分割结果十分不理想,提出了一种基于B样条拟合阈值曲面的分割算法,用于获得分割图像的阈值曲面。在分割出缺陷目标的基础上,选择了几何形状、不变矩等缺陷种类判别特征量,构建了相应的BP神经网络分类器完成对缺陷具体类别的识别。在上述带钢表面缺陷智能检测系统的设计思路下,本文对其中的缺陷初检、分割和识别等关键步骤进行了相应的算法实现与测试,取得良好效果。更多还原

【Abstract】 This paper discussed the overall design of intelligent defects detection system, and focused on the primary detection, segmentation and recognition.For the low probability of defects in actual production, the discriminant function which is used to detect whether the image contains defects or not has been constructed by extracting the features, such as gray range, variance and so on. The image that contains defects can not avoid the impact of uneven illumination, so the traditional segmentation also can not get the ideal division results. A segmentation method using threshold surface based on surface fitting is proposed to overcome such inadequacy. In the basis of the defect area was segmented, the features of the shape and the same moment are selected in order to construct the BP neural network classifiers for recognizing defects. The algorithms of primary detection, segmentation and recognition have been implemented and tested, achieved good results.更多还原