基于GCSA-YOLOv8的航空材料表面缺陷检测算法  

Aerospace Material Surface Defect Detection Algorithm Based on GCSA-YOLOv8

作　　者：陈鑫 于云飞 丁相玉 喻姝桐 CHEN Xin;YU Yunfei;DING Xiangyu;YU Shutong(AECC Shenyang Engine Research Institution,Shenyang 110015,China;School of Materials Science and Engineering,Shenyang Aerospace University,Shenyang 110015,China;School of Power and Energy,Nanchang Hangkong University,Nanchang 330034,China)

机构地区：[1]沈阳发动机研究所,沈阳100095 [2]沈阳航空航天大学材料科学与工程学院,沈阳110015 [3]南昌航空大学动力与能源学院,南昌330034

出　　处：《失效分析与预防》2025年第1期39-47,82,共10页Failure Analysis and Prevention

基　　金：中国航发自主创新专项资金项目(ZZCX-2020-026)。

摘　　要：为提升航空发动机缺陷检测任务的精度与鲁棒性,本文提出了一种基于改进YOLOv8航空材料缺陷检测的算法。首先,针对航空材料中表面缺陷特征难以提取、目标区分度低及尺度不一的问题,在YOLOv8中设计全局通道与空间注意力(GCSA)模块。GCSA模块结合了通道注意力、通道洗牌和空间注意力机制,旨在增强模型对全局上下文依赖关系的建模能力,从而有效捕捉微小缺陷与不规则形状损伤。基于自建航空缺陷数据集进行实验验证,结果表明:相较于原始YOLOv8模型,加入GCSA模块后的改进算法在自建航空材料缺陷数据集上的mAP@0.5和mAP@0.5:0.95分别提升了1.5%和1.1%,改进算法在精度与速度间实现良好平衡,显著增强对航空发动机关键部件缺陷的检测能力,为复杂工业场景下的缺陷检测任务提供更可靠的解决方案。To enhance the accuracy and robustness of defect detection in aerospace engines,this paper proposes an improved YOLOv8-based algorithm for detecting aerospace material defects.Aiming to address the challenges associlated with extracting surface defect features from aerospace materials,such as low target distinctiveness and scale variation,a Global Channel and Spatial Attention(GCSA)module is designed and integrated into YOLOv8.The GCSA module combines channel attention,channel shuffle,and spatial attention mechanisms to strengthen the model’s capability in modeling global contextual dependencies,thereby effectively capturing subtle defects and irregular-shaped damages.Experimental results on a self-built aerospace defect dataset demonstrate that the improved algorithm achieves a 1.5%increase in mAP@0.5 and a 1.1%improvement in mAP@0.5:0.95 compared to the original YOLOv8 model.The proposed algorithm maintains an optimal balance between detection accuracy and speed,significantly enhancing the capability to identify defects in critical aerospace engine components.This work provides a more reliable solution for defect detection tasks in complex industrial scenarios.

关 键 词：YOLOv8 表面缺陷检测 航空材料 注意力通道 

分 类 号：V241.07[航空宇航科学与技术—飞行器设计]