摘要
针对钢材表面裂纹缺陷在二维图像中缺乏深度视觉信息、形貌特征检测困难的问题,设计了一种基于加权的多特征融合的钢材表面裂纹缺陷特征提取算法。首先通过双目视觉检测平台采集钢材表面三维点云图像,然后通过体素化滤波进行初步去噪,并利用RANSAC算法分割平面。其次采用kd树对分割后的点云进行DBSCAN聚类,从而初步定位缺陷区域。最后通过引入曲率和法向量的加权因子提升高度差乘积算法的鲁棒性,有效地放大缺陷点云特征,结合区域生长法实现完整裂纹提取。实验结果表明,改进方法的裂纹提取准确率、召回率、F值分别为97.9%、95.3%、96.6%,较RANSAC算法分别提高了4.1%、11.1%、7.9%,算法运行时间为1.51 s,误差为4.7%,有效提高了对复杂平面的裂纹实时特征提取能力。
To solve the problem of lack of depth visual information and difficulty in detecting surface features of steel with cracks in two-dimensional images,a steel surface crack defect feature extraction algorithm based on weigh-ted multi-feature fusion was designed.First,the steel surface three-dimensional point cloud image was collected through a binocular visual detection platform,and then preliminary noise reduction was performed through voxelization filtering.The plane was segmented using the RANSAC algorithm.Second,the DBSCAN clustering algorithm was used to cluster the segmented point cloud using the kd tree,thereby initializing the defect region.Finally,the robustness of the height difference product algorithm was improved by introducing weighting factors for curvature and normal vectors,effectively enhancing the defect point cloud features,and the complete crack extraction was achieved through the region growing method.The improved method achieved accuracy,recall,and F-score of 97.9%,95.3%,and 96.6%in crack extraction,respectively.These are 4.1%,11.1%,and 7.9%higher than those of the RANSAC algorithm.With a runtime of 1.51s and an error of 4.7%,it significantly enhances real-time crack feature extraction on complex planes.
作者
曾凯
夏梓博
钱俊磊
杜学强
王跃林
朱立光
ZENG Kai;XIA Zibo;QIAN Junlei;DU Xueqiang;WANG Yuelin;ZHU Liguang(College of Electrical Engineering,North China University of Science and Technology,Tangshan Hebei 063210,China;College of Metallurgy and Energy,North China University of Science and Technology,Tangshan Hebei 063210,China;Tangshan Iron and Steel Enterprise Process Control and Optimization Technology Innovation Center(Tangshan ANODE Automation Co.,Ltd.);College of Materials Science and Engineering,Hebei University of Science and Technology,Shijiazhuang 050018,China)
出处
《激光杂志》
北大核心
2025年第8期89-96,共8页
Laser Journal
基金
国家自然科学基金资助项目(No.52374335)
唐山市科技计划项目(No.22130204G、No.22130220G)。
关键词
区域生长
多特征融合
高度差乘积
特征提取
region growth
multiple features fusion
product of height difference
feature extraction
