摘要 针对铆钉表面缺陷纹理形态复杂多变,传统纹理特征提取方法难以获取准确纹理信息、铆钉缺陷识别率较低的问题,提出一种自适应阈值抗噪LBP的铆钉表面缺陷检测算法AT_NRLBP。首先,将铆钉图像均匀分块后提取铆钉子块;然后,基于PCA分解铆钉子块的协方差矩阵,估计子块图像的噪声水平。根据图像噪声强度计算NRLBP阈值,编码铆钉子块得到NRLBP纹理特征。最后,训练SVM单分类器分类铆钉子块,检测并标记出缺陷子块。实验结果表明,这里算法能有效地检测出铆钉表面缺陷,误检率明显降低;与其他纹理分类算法相比,这里算法在KTH-TIPS数据集上的分类准确率较高。 Aiming at the traditional texture feature extraction methods cannot obtain the texture information of rivets accurately and the defect recognition rate is low,which are caused by the surface texture of rivets is complex and variable.A rivet surface defect detection algorithm based on adaptive threshold noise-resistant LBP was proposed,which named AT_NRLBP.Firstly,the rivet images were evenly divided into pieces to extract the rivet sub-blocks.Secondly,the covariance matrix of the rivet sub-blocks were decomposed by PCA and the noise level of the sub-blocks image were estimated.The NRLBP threshold was calculated based on the image noise level,and the rivet sub-blocks were encoded to NRLBP features.Finally,the SVM single classifier was trained by the features and the rivet sub-blocks were classified.And then,the defective sub-blocks were detected and marked.The experimental results show that the proposed algorithm can detect the defects of the rivet surface effectively,and the false detection rate is significantly reduced.Compared with other texture classification algorithms in the KTH-TIPS dataset,the proposed algorithm is better.
机构地区 西南交通大学机械工程学院
出处 《机械设计与制造》 北大核心 2021年第4期56-61,共6页 Machinery Design & Manufacture
基金 四川省科技支撑计划(2016GZ0194)。
关键词 铆钉检测 纹理分类 抗噪LBP 噪声估计 Rivet Detection Texture Classification Noise-Resistant Local Binary Pattern Noise Estimation
