基于改进Canny算子的齿轮边缘缺陷检测方法  

Image Edge Detection Method Based on Improved Canny Algorithm

作　　者：高昕 甄国涌 储成群 王子硕 Gao Xin;Zhen Guoyong;Chu Chengqun;Wang Zishuo

机构地区：[1]中北大学仪器与电子学院

出　　处：《工具技术》2024年第9期145-151,共7页Tool Engineering

基　　金：国家自然科学基金重点项目(62131018);山西省基础研究计划(202103021222012)。

摘　　要：针对Canny算子在齿轮边缘缺陷检测中存在识别相似度低、因图像噪声影响导致微弱边缘难以提取、边缘连续性与鲁棒适应性差的问题,提出了一种改进Canny算子的齿轮边缘缺陷检测技术。改进滤波器,使用梯度双边滤波进行图片预处理,平滑图像并减少图像噪声;改进了卷积核,在45°和135°梯度方向上对像素点进行非极大值抑制,增加了微弱边缘被保留的概率;采用最大类间方差法(Otsu算法)计算图像的高阈值,通过双阈值的办法自适应寻找图像强弱边缘;将Prewitt算子和Canny算法以及改进算法进行对比实验验证。结果表明,改进算法可以提取更完整的齿轮边缘,处理后图片的峰值信噪比(PSNR)相比Canny算法提升了16%,检测效果提升了30%,重叠系数高达81.9%,提升了25.1%,为齿轮边缘缺陷检测提供了一定的参考价值。A modified Canny edge detection technique is proposed to address the issues of low recognition similarity,difficulty in extracting weak edges due to image noise,and poor edge continuity and robustness in gear edge defect detection.The filter is improved by gradient bilateral filtering for image preprocessing,which smoothens the image and reduces image noise.The convolution kernel is enhanced by adding a non-maximum suppression process for pixel points in the 45°and 135°gradient directions,increasing the probability of retaining weak edges.The maximum between-class variance method(Otsu’s algorithm)is employed to compute the high threshold of the image,and an adaptive dual-threshold method is used to locate strong and weak edges in the image.A comparative experiment is conducted between the Prewitt operator,the traditional Canny algorithm,and the modified algorithm.The results show that the modified algorithm can extract more complete gear edges,with a 16%improvement in peak signal-to-noise ratio(PSNR)compared to Canny,a 30%improvement in detection effectiveness,and an overlap coefficient of 81.9%,which is a 25.1%enhancement.This provides valuable reference for gear edge defect detection.

关 键 词：梯度双边滤波 齿轮边缘缺陷检测 最大类间方差法 峰值信噪比 

分 类 号：TG806[金属学及工艺—公差测量技术] TH161.1[机械工程—机械制造及自动化]