基于机器视觉的铝型材表面瑕疵检测方法  

Method for detecting surface defects of aluminumprofile based on machine vision

作　　者：赵文宏[1] 周神特 吕建标 张潇 王宇宇 ZHAO Wenhong;ZHOU Shente;Lü Jianbiao;ZHANG Xiao;WANG Yuyu(College of Mechanical Engineering,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学机械工程学院,浙江杭州310023

出　　处：《浙江工业大学学报》2021年第1期76-81,共6页Journal of Zhejiang University of Technology

基　　金：浙江省科技计划项目-公益技术(LGG18E050025)。

摘　　要：铝型材的表面瑕疵影响其质量、外观以及安全性,如何准确、快速和高效地识别铝型材的表面瑕疵至关重要。为解决检测铝型材表面瑕疵的问题,提出一种基于机器视觉的铝型材表面瑕疵检测方法。该方法基于铝型材的瑕疵种类和特性,采用非线性的双边滤波,并对其定义域核函数作出空间域改进,改进后的滤波算法能够较好地抑制噪声,保留瑕疵边缘信息。预处理后采用改进的Canny算法对图像进行梯度计算并且通过非极大值抑制得到候选边缘,对缺陷进行定位。为了有效地对瑕疵进行分类,对其数据集进行数据增强,防止过拟合情况的发生,结合光照等外界因素对瑕疵特征的影响,采用HOG算法进行特征提取,并将卷积神经网络多层的特征与HOG特征进行融合,对卷积核采用相关系数法优化,解决传统HOG+SVM模型泛化能力和鲁棒性差的问题,铝型材表面瑕疵检测准确率提升至90.26%,平均每张测试集仅需58.34 ms,符合目前对铝型材表面质量检测的需求。The surface defects of the aluminum profile affect its quality,safety and appearance.How to identify the surface defects of the aluminum profile accurately,quickly and efficiently is crucial.To solve this problem,the detection method of the aluminum profile surface defects based on machine vision is proposed.Based on the defect types and characteristics of aluminum profiles,the nonlinear bilateral filtering is used and the spatial function of its definition domain kernel function is improved in this method.The improved filtering algorithm can better suppress noise and retain defect edge information.After preprocessing,the Canny algorithm is used to calculate the gradient of the image and the candidate edges are obtained by non-maximum suppression to locate the defects.In order to effectively classify the defects,the data set is enhanced with data to effectively prevent the occurrence of overfitting.Combining the influence of external factors such as illumination on flaw features,the HOG algorithm is used to extract features,and the multi-layer features of the convolutional neural network are combined with the HOG features to solve the problem of poor generalization ability and poor robustness of the traditional HOG+SVM model.The accuracy rate of surface defect detection of aluminum profiles is increased to 90.26%,and the average test set time is only 58.34ms,which meets the current requirements for surface quality inspection of aluminum profiles.

关 键 词：机器视觉 铝型材 HOG算法 表面瑕疵 卷积神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]