基于非下采样Shearlet变换的磁瓦表面缺陷检测

发布时间：2018-05-18 08:49

本文选题：磁瓦 + 非下采样Shearlet变换　；参考：《工程科学与技术》2017年02期

【摘要】：针对磁瓦表面缺陷对比度低,图像受不均匀背景和磨削纹理干扰大等问题,提出了一种基于非下采样Shearlet变换(non-subsampled Shearlet transform,NSST)的磁瓦表面缺陷检测方法。首先,对磁瓦图像进行多尺度多方向NSST分解,得到一个低频子带图像及多个频率和方向变化的高频子带图像。然后对缺陷在高低频域表现出的不同特征进行针对性的处理,在低频子带中分别计算行均值线图像和列均值线图像,将列均值线图像沿行均值线图像扩展,构造基于均值的自适应阈值对低频子带进行滤波,以去除不均匀背景;同时,利用同一分解尺度下各高频子带系数中微弱缺陷信号的方差较大,显著缺陷信号的能量较大,而噪声和背景干扰信号的方差和能量均较小的差异,构造基于Shearlet高频分解系数方差和能量的综合高频缺陷识别算子,滤除高频子带中的噪声和背景干扰。最后,对修正后的分解系数进行逆NSST重构,得到背景均匀,磨削纹理和噪声干扰被充分抑制的高对比度图像,并采用自适应阈值分割方法提取出缺陷区域。实验结果表明,该方法的假阳性率、假阴性率和检测准确率分别达到8.8%、5.0%和93.1%;本文算法在MATLAB仿真平台中平均运行时间为0.629 s;相较于现有的磁瓦表面缺陷检测算法,该方法能够有效地去除不均匀背景、磨削纹理和噪声干扰,检测结果更加准确,鲁棒性更强。
[Abstract]:Aiming at the problems of low contrast of surface defect of magnetic tile, large interference of image by non-uniform background and grinding texture, a method of detecting surface defect of magnetic tile based on non-subsampled Shearlet transform (non-subsampled Shearlet transform NSST) is proposed. Firstly, a low frequency subband image and a high frequency subband image with multiple frequencies and directions are obtained by multi-scale and multi-directional NSST decomposition. Then we deal with the different characteristics of defects in high and low frequency domain, calculate the row mean line image and column average line image in the low frequency subband, and extend the column mean line image along the row mean line image. The adaptive threshold based on mean value is constructed to filter the low frequency subband to remove the non-uniform background. At the same time, the variance of the weak defect signal in each high frequency subband coefficient under the same decomposition scale is large, and the energy of the significant defect signal is larger. However, the variance and energy of noise and background interference signal are small, so a comprehensive high frequency defect recognition operator based on variance and energy of Shearlet high frequency decomposition coefficient is constructed to filter the noise and background interference in high frequency subband. Finally, the modified decomposition coefficient is reconstructed by inverse NSST, and a high contrast image with uniform background and sufficient suppression of grinding texture and noise interference is obtained, and the defect region is extracted by adaptive threshold segmentation method. The experimental results show that the false positive rate, false negative rate and detection accuracy of this method are 8.8% and 93.1%, respectively. The average running time of this algorithm in MATLAB simulation platform is 0.629 s. This method can effectively remove uneven background, grinding texture and noise interference, and the detection results are more accurate and robust.
【作者单位】：四川大学制造科学与工程学院;
【基金】：国家科技支撑计划资助项目(2015BAF27B01) 四川省科技支撑计划资助项目(2016GZ0160)
【分类号】：TP391.41

【参考文献】