摘要
针对流水线动态检测汽车法兰表面缺陷效率较低的问题,提出一种基于改进YOLOv3算法和DMPHN算法的汽车法兰表面缺陷检测方法。在图像预处理阶段使用DMPHN算法,可以解决汽车法兰表面缺陷图片模糊、失真的问题。目标识别网络使用YOLOv3算法,将DarkNet53替换为轻量级模型MobileNetV3,并进行改进,用于检测去模糊后的图像。这一方法的平均精度均值为97.74%,比改进之前提高1.03个百分点,检测速度达到每秒6.8帧。与传统表面缺陷检测方法相比,这一方法通过优化网络结构,解决了特定情况下模糊图像的检测精度问题。
Aiming at the low efficiency of pipeline dynamic detection of automobile flange surface defect,a method for automobile flange surface defect detection based on improved YOLOv3 algorithm and DMPHN algorithm was proposed.Using the DMPHN algorithm in the image preprocessing stage can solve the problem of blurred and distorted image of automobile flange surface defect.The object recognition network uses the YOLOv3 algorithm,which replaces DarkNet53 with a lightweight model MobileNetV3 and improves it for detecting deblurred image.The mean average precision of this method is 97.74%,which is 1.03 percentage points higher than the mean average precision before the improvement,and the detection speed reaches 6.8 frames per second.Compared with traditional surface defect detection methods,this method solves the issue of detection accuracy of blurred image in specific situation by optimizing the network structure.
出处
《机械制造》
2022年第8期74-80,共7页
Machinery
基金
国家自然科学基金资助项目(编号:61703217,51505238)
浙江省自然科学基金资助项目(编号:LY21F030003)
宁波市科技创新2025重大专项(编号:2019B10100)
关键词
汽车
法兰
表面缺陷
检测
Automobile
Flange
Surface Defect
Detection
作者简介
唐金亮(1996—),男,硕士研究生,主要研究方向为深度学习和缺陷检测。
