改进MB-LBP特征与LPP算法在带钢表面缺陷识别上的应用  

Application of improved MB-LBP features and LPP algorithm in strip surface defect identification

作　　者：陈小芳 于凤芹[1] 陈莹[1] CHEN Xiaofang;YU Fengqin;CHEN Ying(College of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China)

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《传感器与微系统》2020年第4期156-160,共5页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61573168);中央高校基本科研业务费专项资金资助项目(JUSRP51733B)。

摘　　要：针对带钢表面缺陷识别率受到光照变化、纹理复杂多样以及噪声干扰而导致误识别率高的问题,提出一种新的带钢表面缺陷识别算法。首先从增加邻域联系的角度改进多块局部二值模式(MB-LBP)特征,缓解提取过程中因所选子窗口尺寸大小不同而造成的保留图像细节与去除噪声之间的平衡性问题;其次将改进的MB-LBP特征与梯度方向直方图(HOG)特征线性加权得到融合特征,弥补MB-LBP特征没有表征缺陷边缘和方向的缺点,从而更全面地表征复杂的缺陷纹理;最后通过同时增加全局信息和监督信息改善的局部保持投影(LPP)算法将高维的融合特征非线性映射到低维的本质特征空间中,减少融合特征冗余对分类器识别率的影响。在NEU数据集上仿真实验结果表明:算法对光照变化、纹理复杂多样、以及噪声具有一定的鲁棒性,在信噪比为50 d B情况下将带钢表面缺陷识别准确率提高了5. 17%。Aiming at the problem of high false recognition rate caused by the complex texture of the strip surface defect,the noise,and uneven illumination,a new strip surface defect identification algorithm is proposed. Firstly,the multi-block local binary patterns( MB-LBP) feature is improved from the perspective of increasing the neighborhood connection,which alleviates the balance between the retained image details and the noise removal caused by the selected sub-window size in the MB-LBP feature extraction process. Secondly,the improved MB-LBP feature and the histogram of oriented gradient( HOG) feature are linearly weighted to obtain the fusion feature,which makes up for the shortcomings of the MB-LBP feature without characterizing the defect edge and direction,so as to more fully characterize the complex defect texture. Finally,the locality preserving projection( LPP)algorithm which improves the global information and the supervised information is used to nonlinearly map the high-dimensional fusion features into the low-dimensional essential feature space,and reduce the influence of the fusion feature redundancy on the classifier recognition rate. The simulation results on NEU dataset show that the proposed algorithm has certain robustness to illumination changes,complex texture and noise,and improves the accuracy of strip surface defect identification by 5. 17 %,in case of signal to noise ratio( SNR) of 50 d B.

关 键 词：机器视觉 改进多块局部二值模式特征 融合特征 局部保持投影 带钢表面缺陷识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]