基于改进Faster R-CNN的飞机蒙皮缺陷检测方法  

Aircraft Skin Defect Detection Method Based on Improved Faster R-CNN

作　　者：潘甜 惠开发 PAN Tian;HUI Kaifa(School of Mechanical Engineering,Jiangsu Aviation Vocational and Technical College,Zhenjiang 212134,China;School of Information,Yancheng Institute of Technology,Yancheng 224000,China)

机构地区：[1]江苏航空职业技术学院机械工程学院,江苏镇江212134 [2]盐城工学院信息学院,江苏盐城224000

出　　处：《机械工程与自动化》2025年第2期139-141,共3页Mechanical Engineering & Automation

基　　金：2022年度江苏航院院级重点课题资助项目(JATC22010107)。

摘　　要：针对传统的检测算法对目标缺陷定位精度不高的问题,提出了一种基于改进Faster R-CNN的飞机蒙皮缺陷检测方法。在对缺陷图像进行预处理获得数据增强后,首先在主干网络中融合特征金字塔网络(FPN),加强对小缺陷的特征提取;接着使用ROI Align算法进行双线性插值,更加准确地定位目标缺陷;最后使用K-means算法得到适应飞机蒙皮缺陷的锚框。实验结果表明:优化后的Faster R-CNN算法具有较强的目标定位处理能力,对蒙皮缺陷检测的平均精度MAP达到91%,较传统算法提升了5%;在复杂缺陷混合的情况下对小目标缺陷的检测准确率高达95%,较传统算法提升了7%,为飞机蒙皮表面的缺陷检测提供了一种新的研究方向。Aiming at the problem that the traditional detection algorithm is not high in locating target defects,an aircraft skin defect detection method based on improved Faster R-CNN is proposed.After the defect image is preprocessed to obtain data enhancement,the feature pyramid network(FPN)is fused into the backbone network to enhance the feature extraction of small defects.Then the ROI Align algorithm is used for bilinear interpolation to locate the target defects more accurately.Finally,the anchor frame adapted to aircraft skin defects is obtained by using K-means algorithm.The experimental results show that the optimized Faster R-CNN algorithm has better target location processing capability,and the average precision MAP of skin defect detection reaches 91%,which is 5% higher than the traditional algorithm.In the case of mixed complex defects,the detection rate of small target defects is as high as 95%,which is 7% higher than the traditional algorithm,and provides a new research direction for the defect detection of aircraft skin surface.

关 键 词：飞机蒙皮 改进的Faster R-CNN 缺陷检测 FPN ROI Align K-MEANS 

分 类 号：TH164[机械工程—机械制造及自动化]