Abstract: In the harsh corrosive environments characterized by high temperatures,high humidity and high salt spray,corrosion issues frequently arise in the structures of aircraft wings and fuselage.Rapid and accurate detection of surface metal corrosion,along with the implementation of appropriate repair and maintenance measures was crucial for ensuring flight safety,extending service life and reducing aircraft downtime for maintenance.The characteristics of metal corrosion on aircraft wing surfaces,highlighting typical corrosion sites and types were analyzed.Through a review and analysis of relevant literature from both domestic and international sources,the visual detection and identification methods for detecting metal corrosion on aircraft wings were discussed.The current state of research was summarized across three key approaches:traditional image processing,machine learning and deep learning.Particular emphasis was placed on deep learning-based corrosion identification methods,with an analysis of the principles and applications of techniques such as YOLO and Detection Transformer (DETR).To address the issue of insufficient datasets for aircraft wing surface metal corrosion that negatively impacts model training performance,the sample augmentation methods based on data enhancement and transfer learning were explored,and the relevant datasets and performance evaluation metrics for corrosion identification were compiled.Finally,the key challenges in identifying metal corrosion on aircraft wing surfaces and their corresponding solutions were summarized,while providing insights and predictions for future research directions and trends in this field.

Key words: aircraft wings, corrosion identification, deep learning, small sample size, transfer learning

摘要： 针对高温、高湿、高盐雾等严苛腐蚀环境下飞机机翼及机体结构腐蚀频发,快速、准确地识别腐蚀,并采取相应的修复与维护措施,对保障飞机飞行安全、延长使用寿命及缩短停机维修时间具有重要意义。分析了飞机机翼表面金属腐蚀的特性,列举了典型的腐蚀部位及类型。通过对国内外相关技术文献的调研与分析,重点探讨了基于视觉检测与识别的飞机机翼表面金属腐蚀识别方法,从传统图像处理、机器学习和深度学习3个方面综述了研究现状。特别关注了基于深度学习的腐蚀识别方法,分析了YOLO,DETR等方法的识别原理与应用场景。针对飞机机翼表面金属腐蚀数据集数量不足所导致的模型训练效果不佳的问题,探讨了基于数据增强和迁移学习的样本扩充方法,梳理了相关数据集及腐蚀识别性能评价指标。最后,总结了飞机机翼表面金属腐蚀识别的关键问题及其解决方案,并对未来研究方向与发展趋势进行了展望与预测。

关键词: 飞机机翼, 腐蚀识别, 深度学习, 小样本, 迁移学习