
科学技术与工程

中国技术经济学会
基于改进YOLOv11n的飞机铆钉损伤视距自适应检测方法
夏正洪 , 钟吉飞 , 张军 , 胡佳鑫 , 杨磊 , 黄明泽
科学技术与工程 ›› 2026, Vol. 26 ›› Issue (23) :10072-10079
基于改进YOLOv11n的飞机铆钉损伤视距自适应检测方法
Aircraft Rivet Damage Detection Algorithm at Different Viewing Distances Based on Improved YOLOv11n
针对飞机铆钉损伤检测中因成像视距动态变化引发的特征分辨率衰减、尺寸变化及识别率显著下降等问题,提出一种基于改进YOLOv11n的飞机铆钉损伤视距自适应检测方法。首先,依托AirNet框架构建正负样本对学习退化表征空间,并利用退化先验信息引导修复网络,实现退化图像的特征分辨率恢复;其次,借助SCConv的空间与通道双路协同重构机制,抑制近距铆钉损伤特征的非关键背景,同时增强远距铆钉损伤的关键特征;然后,构建160×160高分辨率P2检测层,解决铆钉因拍摄距离波动导致的尺寸变化问题;最后,提出DySample距离感知卷积核形变映射函数,实现近距离局部采样对细节的保留,远距离铆钉损伤特征的上下文语义增强,从而有效提升铆钉损伤的识别率。通过自主构建的多视距飞机铆钉损伤数据集对改进算法效果进行验证,结果显示:改进算法精确率达92.1%、召回率达83.2%、mAP@0.5达89.6%,较基准模型YOLOv11n分别提升3.4%、6.4%、5.3%,验证了所提飞机铆钉损伤视距自适应检测方法的有效性。
To address the issues of feature resolution attenuation, size variation, and significant decline in recognition rate caused by dynamic changes in imaging distance during aircraft rivet damage detection, a viewing-distance adaptive detection method based on improved YOLOv11n was proposed. Firstly, leveraging the AirNet framework, a degraded representation space was learned by constructing positive-negative sample pairs, and a restoration network guided by degradation priors was employed to recover feature resolution in degraded images. Secondly, the dual-path spatial and channel cooperative reconstruction mechanism of SCConv was utilized, non-critical backgrounds in close-range rivet damage features were suppressed while simultaneously enhancing critical features in long-range rivet damage. Thirdly, a high-resolution 160×160 P2 detection layer was constructed to resolve the problem of rivet size variations resulting from fluctuating shooting distances. Finally, the DySample distance-aware convolution kernel deformation mapping function was introduced, enabling local sampling at close range to preserve details while enhancing the contextual semantics of long-range rivet damage features. This effectively improved the recognition rate for rivet damage. The effectiveness of the improved algorithm was validated using a self-built multi-viewing-distance aircraft rivet damage dataset. Results demonstrate that the improved algorithm achieved a precision of 92.1%, a recall of 83.2%, and an mAP@0.5 of 89.6%, representing improvements of 3.4%, 6.4%, and 5.3% respectively compared to the baseline YOLOv11n model. These results verify the effectiveness of the proposed viewing-distance adaptive detection method for aircraft rivet damage.
航空安全 / 目标检测 / 铆钉损伤 / 卷积核形变映射 / 改进YOLOv11n
aviation safety / object detection / rivet damage / convolutional kernel deformation mapping / improved YOLOv11n
| [1] |
谢辉, 高文翔, 张立新, |
| [2] |
|
| [3] |
|
| [4] |
孟芳芳, 田孝壮, 方薇, |
| [5] |
|
| [6] |
杨春晖, 戴家隆, 张森, |
| [7] |
|
| [8] |
李淑凤, 曹秀. 基于飞机表面检测的无人机全覆盖路径规划[J]. 科学技术与工程, 2025, 25(29): 12699-12708. |
| [9] |
|
| [10] |
|
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
崔俊佳, 张军, 莦茹茹, |
| [16] |
|
| [17] |
魏英姿, 苏迈. YOLOv5框架下的飞机蒙皮铆接缺陷检测[J]. 组合机床与自动化加工技术, 2023(7): 106-109. |
| [18] |
|
| [19] |
李耀华, 陈杰. 融合注意力机制的飞机蒙皮损伤LHN-YOLO小目标检测算法[J]. 北京航空航天大学学报, 2025(3): 1-13. |
| [20] |
|
| [21] |
董芳媛, 韩天龙. 基于深度学习的飞机蒙皮铆接质量检测系统设计[J]. 现代制造技术与装备, 2025, 61(9): 44-46. |
| [22] |
|
| [23] |
夏正洪, 何琥, 杨磊, |
| [24] |
|
国家重点研发计划(2024YFC3014400)
中央高校基本科研业务费(26CAFUC03075)
中央高校基本科研业务费(26CAFUC10023)
四川省民航飞行技术与飞行安全工程技术研究中心项目(GY2025-23D)
/
| 〈 |
|
〉 |