成都大学学报（自然科学版）2023，Vol.42Issue(4)：365-371,7.DOI:10.3969/j.issn.1004-5422.2023.04.006
基于融合神经网络的飞机蒙皮缺陷检测的研究
Investigation of Aircraft Skin Defect Detection Based on Fusion Neural Network
摘要
Abstract
The aircraft skin is damaged by multiple factors,such as atmospheric environment erosion,bird strike and so on.Flight safety is threatened by those factors.A skin defect detection method based on fu-sion neural network is proposed to solve the problems,such as time-consuming and insufficient manual in-spection in aircraft skin in this paper.The Xception architecture is integrated into the YOLOv5 network,and the global channel attention mechanism is added to Backbone,and the channel space attention mech-anism is added in Neck and Output so as to form a new fusion neural network based on the YOLOv5 net-work.The 8 503 images of aircraft surface defects collected are divided into training sets and test sets.Af-ter training,the new fusion neural network is verified by the test set,and the average accuracy of the five kinds of defects detection,including rivet peeling,rivet corrosion,skin peeling,skin crack and skin im-pact,are 0.960,0.928,0.931,0.934,0.948 respectively.And the overall recognition accuracy of the whole aircraft skin defects to the new fusion neural network is 0.950,the recall rate is 0.964,and the av-erage accuracy rate is 0.957.The experimental results show that the new fusion neural network is effective for aircraft skin defect recognition.关键词
飞机蒙皮缺陷/注意力机制/深度学习/融合神经网络/目标检测Key words
aircraft skin defects/attention mechanism/deep learning/fusion neural network/object detection分类
航空航天引用本文复制引用
张德银,黄少晗,赵志恒,李俊佟,张裕尧..基于融合神经网络的飞机蒙皮缺陷检测的研究[J].成都大学学报（自然科学版）,2023,42(4):365-371,7.基金项目
中国民航局科技项目(MHRD1229) （MHRD1229）
中央高校基本科研业务费专项资金项目(ZJ2022-007) （ZJ2022-007）
中国民用航空飞行学院大创立项(S202210624219) （S202210624219）
