基于改进Otsu算法的金属器件镀锌表面缺陷识别方法  

Improved Otsu algorithm for galvanized surface defect recognition of metal devices

作　　者：马栎[1] 冯占荣[2] Ma Li;Feng Zhanrong(Jiangxi Industry Polytechnic College,Nanchang 330000,China;School of Aeronautical Manufacturing Engineering,Nanchang Hangkong University,Nanchang 330063,China)

机构地区：[1]江西工业职业技术学院,江西南昌330000 [2]南昌航空大学航空制造工程学院,江西南昌330063

出　　处：《电镀与精饰》2025年第2期46-53,共8页Plating & Finishing

基　　金：江西省教育厅科学技术研究项目(GJJ160701)。

摘　　要：镀锌表面纹理、颜色以及亮度变化的复杂度往往较高,且不同的光照条件会对金属表面的反射和阴影产生显著影响,当前固定的阈值选择方式难以适应这种复杂多变的识别环境,影响当前人工智能领域中表面缺陷的识别效果,故提出了基于改进Otsu算法的金属器件镀锌表面缺陷识别方法。首先,针对金属器件镀锌表面图像,根据结构张量提取图像的轮廓信息,利用Itti模型提取图像颜色和亮度信息,并分别生成各通道显著图。经规范化处理后,通过线性组合构成视觉显著图,用于初步判断图像中是否存在表面缺陷;然后,在常规的Otsu算法中,引入二阶振荡粒子群优化算法多次调整灰度阈值,利用最优的灰度阈值分割出缺陷区域;最后,利用加权马氏距离表示协方差距离,突出缺陷边缘像素特征,使缺陷兴趣区域更加显著,再采用连通区域标记的方式准确识别表面缺陷。实验结果表明,在金属器件镀锌表面缺陷人工智能识别中,该方法可以准确检索到缺陷区域,识别结果的敏感度和特异性较高。由此可以说明,该方法具有良好的应用效果。The complexity of texture,color,and brightness changes on galvanized surfaces is often high,and different lighting conditions can have a significant impact on the reflection and shadow of metal surfaces.The current fixed threshold selection method is difficult to adapt to this complex and changing recognition environment,which affects the recognition effect of surface defects in the field of artificial intelligence.Therefore,a method for identifying surface defects on galvanized metal devices based on an improved Otsu algorithm was proposed.Firstly,for the galvanized surface image of metal devices,the contour information of the image was extracted based on the structural tensor.The Itti model was used to extract the color and brightness information of the image,and the saliency maps of each channel were generated separately.After standardization,a visual saliency map was constructed through linear combination to preliminarily determine whether there were surface defects in the image;Then,in the conventional Otsu algorithm,a second-order oscillation particle swarm optimization algorithm was introduced to adjust the grayscale threshold multiple times,and the optimal grayscale threshold was used to segment the defect area;Finally,the weighted Mahalanobis distance was used to represent the covariance distance,highlighting the pixel features of defect edges to make the defect interest region more prominent.Then,the connected region labeling method was used to accurately identify surface defects.The experimental results showed that in the artificial intelligence recognition of surface defects on galvanized metal devices,this method could accurately retrieve the defect area,and the sensitivity and specificity of the recognition results were high.This indicated that the method had good application effects.

关 键 词：OTSU算法 金属器件 镀锌表面 缺陷识别 二阶振荡粒子群优化算法 最优灰度阈值 GABOR小波变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]