基于改进YOLOv11-EB的芯片表面缺陷检测方法

Chip Surface Defect Detection Method Based on Improved YOLOv11-EB

摘要: 针对芯片缺陷检测过程中缺陷特征相似、缺陷目标小、缺陷尺度差异大造成检测精度低等问题，提出一种基于改进YOLOv11的芯片表面缺陷检测模型YOLOv11-EB。首先采用EfficientNetV2替代原YOLOv11骨干网络，通过其更强的特征提取能力增强细粒度信息表达，从而有效缓解缺陷特征相似、小目标易漏检以及复杂背景下误检的问题。其次将特征融合层(Neck)替换为加权双向特征金字塔网络(BiFPN)，以提升多尺度特征融合效果。实验结果表明，该方法在复杂背景和多尺度缺陷条件下，能够显著降低误检率与漏检率，并有提高模型的检测精度与召回率。与原模型相比，改进模型在精确率、召回率和平均精度均值分别提升了5.3%、11.9%和7.3%。

Abstract: To address the issues of low detection accuracy in chip surface defect inspection caused by highly similar defect features, small target sizes, and large scale variations, this paper proposes an improved YOLOv11-based model, termed YOLOv11-EB.First, EfficientNetV2 is adopted to replace the original backbone network of YOLOv11, leveraging its superior feature extraction capability to enhance fine-grained feature representation. This effectively alleviates the challenges of feature similarity, missed detections of small defects, and false detections in complex backgrounds. Second, the original feature fusion network (Neck) is replaced with a weighted Bidirectional Feature Pyramid Network (BiFPN) to improve multi-scale feature fusion. Experimental results demonstrate that the proposed method significantly reduces both false positive and false negative rates under complex background and multi-scale defect conditions, while improving overall detection performance. Compared with the original model, the proposed model achieves improvements of 5.3%, 11.9%, and 7.3% in precision, recall, and mean Average Precision, respectively.