基于改进K-means算法的钢管表面缺陷视觉检测方法

Visual inspection method for surface defects of steel pipes based on improved K-means algorithm

为了利用机器视觉技术检测钢管表面缺陷,设计并搭建了钢管表面图像采集实验平台,针对钢管表面覆盖有氧化铁皮以及弧形外表面易造成光照不均等问题,提出一种基于改进K-means灰度正反求和的检测方法。首先采用垂直投影法获取钢管区域图像,计算得到其灰度反转图像,参照Frankle-McCann Retinex算法原理分别对钢管区域图像及灰度反转图像进行增强,获得各自背景均匀的高对比度图像,再采用改进的K-means算法进行图像分割,得到两个缺陷检测结果,并对二者求和,最后通过图像后处理优化检测结果。构建了不同光照环境下包含凹坑、翘皮、划伤和辊痕等多类缺陷的钢管表面图像样本集进行实验,结果表明本文方法的检测精度较高,对光照不均匀具有良好的抗干扰能力。

In order to detect the surface defects of steel pipes by machine vision technology,an experimental platform for image acquisition of steel pipe surface was designed and constructed.Aimed at the problem of uneven illumination caused by steel pipe surface with iron oxide scale and arc-shaped surface,a detection method based on improved K-means algorithm and the sum of positive and negative gray-scale images was proposed.Firstly,the vertical projection method was used to extract the image of steel pipe area,and the image after gray-scale inversion was calculated.By reference to the principle of Franklin-McCann Retinex algorithm,the original and gray-scale inversion images of the steel pipe area were enhanced respectively to get images with uniform background and high contrast.Then the improved K-means algorithm was used to segment the surface defects from the enhanced images,and the sum of two segmentation results was obtained.Finally,image post-processing was performed to optimize the detection results.Image sample sets of steel tube surfaces containing such defects as pit,warping,scratch and roll mark under different illumination conditions were built for experiment.The test results show that the proposed method has high detection accuracy and good robustness to uneven illumination.