Steel Surface Defect Detection Model Based on GFIF-YOLO
DU Yang
WANG Qian
HE Yongjiao
ZHANG Xiaoshuang
LI Gexian
WANG Lin
Abstract：To address the challenges of inconspicuous features and significant scale variations in steel surface defects that make detection difficult,a global feature interaction fusion-you only look once(GFIF-YOLO)model was proposed for steel surface defect detection.Firstly,the feature pyramid network(FPN)in the YOLO version 8 nano(YOLOv8n)was replaced with a gather-and-distribute(GD)mechanism,which enhanced feature fusion capability while mitigating information loss during cross-layer transmission.Secondly,a global attention mechanism(GAM)was embedded into the final layer of the backbone network to strengthen the ability of global feature representation.Finally,an efficient multi-scale convolution head(EMSCHead)module was introduced to reduce model parameters and computational costs while improving detection accuracy.The results demonstrated that the GFIF-YOLO model achieved mean average precision of 82.0%and 66.7%on the Northeastern University detection(NEU-DET)and the generic component 10-class detection(GC10-DET)dataset,respectively,outperforming the baseline YOLOv8n by 3.0 and 3.4 percentage points.The GFIF-YOLO model showed good performance and could effectively complete the detection task of steel surface defects.
Keywords：defect detectionGD mechanismattention mechanismEMSCHeadglobal featuresmodel lightweight
Publication Date：2025-09-20
Online Publishing Date：2025-09-24(First online date of this platform, not the publication date of the document)
Pages：8( 388-395 )