完美论文网提供代写毕业论文、代写本科论文、代写硕士论文、论文发表及查重服务!QQ:505676555

电话:18796337551
当前位置:完美论文网信息系统 → 文章正文

基于视觉显著性的钢板表面缺陷检测研究

作者:完美论文网  来源:www.wmlunwen.com  发布时间:2019/10/10 9:00:28  

摘要:中国作为多年以来钢铁生产总量稳居世界第一的国家,其钢铁企业在产业结构及技术水平等方面都得到了改善。随着技术水平不断更新,各行各业对钢铁制品的品质要求也越来越高,而钢板表面的缺陷是影响钢铁制品质量的关键因素之一。因此,开展对钢板表面缺陷检测方法的研究具有较大的实用价值。

由于钢板表面的缺陷区域是图像中显著的前景目标并且具有稀疏性,符合视觉显著性检测算法的基本理论特征,则可通过基于视觉显著性的方法来解决钢板表面缺陷检测率低和检测速度慢的问题。本课题针对现有视觉显著性检测算法的优缺点、性质以及特点,设计了两种改进算法,并通过大量对比实验验证两种算法的优越性。本课题的主要研究工作有以下两点:

第一,针对钢板表面缺陷不易检测的难题,提出基于IFT-SMD的钢板表面缺陷检测算法。该算法提出采用自适应全变分模型替代频率调谐(FT)显著性算法的高斯滤波器和融入H分量的方法来构建改进型FT(IFT)算法,采用结构化矩阵分解(SMD)算法提取缺陷区域的结构信息。并由提出的自适应融合算法融合IFT和SMD算法生成的显著图,得到缺陷区域最终的显著图。以钢板表面缺陷图像为实验样本,与几种经典的显著性检测方法进行比较,IFT-SMD算法能够有效地抑制钢板表面背景的表达,实现钢板表面缺陷与背景的分离。IFT-SMD算法提取出的缺陷区域较为完整,其检测精确率也有显著提升。

第二,为进一步提高钢板表面缺陷的检测效率,提出采用基于谱图加权的低秩矩阵分解(SWLRD)算法。该算法在低秩矩阵分解算法的基础上引入了谱图先验知识,通过谱图先验来校正由位置、颜色和边界等传统先验引起的不合理先验分配,以获取谱图加权特征矩阵。根据谱图加权特征矩阵来构建谱图加权的低秩矩阵分解算法,对其进行优化求解以实现钢板表面缺陷区域与背景区域的分离。最后通过回归优化算法和激活函数的后处理,以提高表面缺陷区域的完整性和均匀性,获得凸显钢板表面缺陷的高质量显著图。通过对比实验可以看出,SWLRD算法在检测精确率上与IFT-SMD算法大致相同,但在检测速度上有明显提高。

As the country with the largest steelproduction in the world for many years, its steel companies have improved interms of industrial structure and technical level. With the continuous updatingof the technical level, the quality requirements of steel products in variousindustries are getting higher and higher, and the defects on the surface ofsteel plates are one of the key factors affecting the quality of steelproducts. Therefore, research on the detection method of steel surface defectshas great practical value.

Since the defect area on the surface of thesteel sheet is a significant foreground target in the image and has sparseness,it conforms to the basic theoretical characteristics of the visual saliencydetection algorithm. The problem of low surface defect detection rate and slowdetection speed can be solved by a method based on visual saliency. Based onthe advantages, disadvantages, properties and characteristics of the existingvisual saliency detection algorithms, this paper designs two improvedalgorithms and verifies the superiority of the two algorithms through a largenumber of comparative experiments. The main research work of this subject hasthe following two points:

Firstly, based on the problem that thesurface defects of steel sheets are not easy to detect, an algorithm based onIFT-SMD for surface defect detection of steel sheets is proposed. The improvedFT (IFT) algorithm is constructed by using adaptive total variation modelinstead of the Gauss filter of frequency tuning (FT) saliency algorithm andincorporating H component. Structural matrix decomposition (SMD) algorithm isused to extract the structural information of defect areas. The proposedadaptive fusion algorithm is used to fuse the saliency map generated by the IFTand SMD algorithms to obtain the final saliency map of the defect area. Takingthe surface defect image of the steel sheet as the experimental sample,compared with several classical significant detection methods, the IFT-SMDalgorithm can effectively suppress the expression of the surface background ofthe steel sheet, and realize the separation of the surface defects and thebackground of the steel plate. The defect area extracted by IFT-SMD algorithmis relatively complete, and its detection accuracy is also significantlyimproved.

Secondly, in order to further improve thedetection efficiency of surface defects of steel plates, a low rank matrixdecomposition algorithm (SWLRD) based on spectral weighting is proposed. Basedon the low rank matrix decomposition algorithm, the algorithm introduces theprior knowledge of the spectrum, and corrects the unreasonable priordistribution caused by traditional priors such as position, color and boundaryby spectral prior to obtain the weighted features of the spectrum matrix.According to the spectral weighted feature matrix, a low-rank matrixdecomposition algorithm based on spectral weighting is constructed andoptimized to solve the separation of the surface defect area and the backgroundarea of the steel sheet. Finally, through the post-processing of the regressionoptimization algorithm and the activation function, the integrity anduniformity of the surface defect area are improved, and a high-quality saliencymap highlighting the surface defects of the steel sheet is obtained. It can beseen from the comparison experiments that the SWLRD algorithm is roughly thesame as the IFT-SMD algorithm in detecting accuracy, but the detection speed issignificantly improved.

关键词:钢板表面缺陷检测;视觉显著性;低秩矩阵分解;FT算法;SWLRD算法

steel sheet surface defect detection;visual saliency; low rank matrix decomposition; FT algorithm; SWLRD algorithm

联系方式

客服QQ 505676555
客服热线18796337551
网站地址 www.wmlunwen.com
郑重承诺 原创,包修改,包通过!
完美论文网真诚欢迎新老客户的光临与惠顾!