红外技术2025，Vol.47Issue(5)：640-647,8.
基于改进YOLOv7-FSE算法的飞机复合材料缺陷红外检测
Infrared Detection of Defects in Aircraft Composite Materials Based on Improved YOLOv7-FSE Algorithm
摘要
Abstract
This study proposes an improved detection algorithm,YOLOv7-FSE(YOLOv7 with FReLU-SiLU-EIOU enhancements),to address the challenges of low resolution and poor detection accuracy in infrared images of composite material defects in aircraft.These limitations make it difficult to accurately characterize defect features.The proposed algorithm introduces several key modifications to the original YOLOv7 architecture.First,the SiLU activation function is replaced with the funnel activation function FReLU to improve spatial sensitivity to defect features.Subsequently,space-to-depth convolution(SPD Convolution)is employed to improve the feature extraction process,thereby enhancing the algorithm's ability to characterize complex defect features in low resolution infrared images.Finally,the EIOU loss function is replaced by the CIOU loss function,and the boundary box recognition weights are optimized to generate higher quality anchor boxes,further improving overall detection performance.Comparison results demonstrate that YOLOv7-FSE outperforms traditional detection methods such as Faster RCNN and YOLOv3.Specifically,it achieves a mean average precision(mAP)improvement of 10.8％over Faster R-CNN and 10.1％over YOLOv3.Compared to the original YOLOv7,the precision(P)increases from 88.3％to 94.9％,while the mAP rises from 90.1％to 97.7％.The YOLOv7-FSE algorithm is well-suited for infrared detection of composite material defects on aircraft surfaces and holds significant potential for integration with embedded devices for rapid,on-site defect detection.关键词
YOLOv7-FSE/复合材料缺陷/红外检测/低分辨率/mAP精度Key words
YOLOv7-FSE/defects in composite materials/infrared detection/low resolution/mAP分类
信息技术与安全科学引用本文复制引用
张华忠,邓旭,李飞,杨荣,钟勉..基于改进YOLOv7-FSE算法的飞机复合材料缺陷红外检测[J].红外技术,2025,47(5):640-647,8.基金项目
四川省通用航空器维修工程技术研究中心项目(GAMRC2021YB12,GAMRC2023ZD01),民航飞行技术与飞行安全重点实验室自主研究项目(FZ2022ZZ03). （GAMRC2021YB12,GAMRC2023ZD01）
