摘要
【目的】为了解决红外热成像缺陷检测中缺陷信号微弱和边界模糊导致的检测准确性与鲁棒性不足的问题,提出一种基于增强图像引导的改进YOLOv11检测模型(improved YOLOv11 detection model based on enhanced image guidance,IEIG-YOLOv11)。【方法】首先,在原始YOLOv11模型的输入端引入预处理后的增强热图像,丰富输入特征表达;然后,在主干网络引入高效多尺度注意力机制(efficient multi-scale attention,EMA),结合残差连接提高浅层特征表达能力,增强对微弱缺陷的检测能力;最后,在颈部网络中引入内容感知特征重组(content-aware reassembly of features,CARAFE)上采样模块,以提高浅层特征重建质量,强化缺陷细节特征。【结果】经消融试验验证,上述改进均对模型性能有所提升。针对缺陷目标检测,IEIG-YOLOv11模型的精确率、召回率及F 1分数分别达到0.967、0.978和0.972。IEIG-YOLOv11模型与主流目标检测模型的对比试验结果表明,IEIG-YOLOv11模型的各项性能均优于其他模型,显著减少了漏检现象,有效增强了对边缘模糊及微弱缺陷的识别能力,并能兼顾检测精度与效率。【结论】IEIG-YOLOv11模型为实现微弱信号情况下的红外缺陷目标检测提供了解决方案,进而为红外热成像技术的实际应用奠定了基础。
[Objective]To address the limitations of accuracy and robustness in infrared thermography defect detection caused by weak defect signals and blurred boundaries,an improved YOLOv11 detection model based on enhanced image guidance(IEIG-YOLOv11)was proposed.[Method]First,the enhanced thermal images obtained through preprocessing were introduced at the input of the original YOLOv11 model to enrich feature representation;then,an efficient multi-scale attention(EMA)mechanism was incorporated into the backbone network,combined with residual connections,to improve the shallow feature representation and enhance the detection capability for weak defects;finally,a content-aware reassembly of features(CARAFE)upsampling module was embedded in the neck network to improve the shallow feature reconstruction quality and strengthen defect detail features.[Result]Ablation experiments verify that each of the above improvements contributes to performance enhancement.For defect target detection,the IEIG-YOLOv11 model achieves a precision of 0.967,a recall of 0.978,and an F 1-score of 0.972.Comparative experiments with mainstream detection models demonstrate that the IEIG-YOLOv11 model outperforms others across multiple metrics,significantly reducing missed detections,effectively enhancing the recognition capacity of blurred edges and weak defects,and achieving a balance between detection accuracy and efficiency.[Conclusion]The IEIG-YOLOv11 model provides a feasible solution for infrared defect detection under weak signal conditions,laying the groundwork for practical applications of infrared thermography technology.
作者
李宏伟
周乐
刘薇
LI Hongwei;ZHOU Le;LIU Wei(School of Automation and Electrical Engineering,Zhejiang University of Science and Technology,Hangzhou 310023,Zhejiang,China)
出处
《浙江科技大学学报》
2025年第6期571-582,共12页
Journal of Zhejiang University of Science and Technology
基金
国家自然科学基金项目(62173306)。
关键词
红外热成像
缺陷检测
YOLOv11
增强图像引导
高效多尺度注意力机制
内容感知特征重组
infrared thermal imaging
defect detection
YOLOv11
enhanced image guidance
efficient multi-scale attention mechanism
content-aware reassembly of features
