In response to the problems that current deep learning methods often fail to achieve sufficient accuracy in detecting multi-scale features and have poor performance in identifying minor defects when detecting surface defects of hot-rolled steel strips, a lightweight defect detection algorithm for hot-rolled steel strips, HGFD-YOLOv8, was proposed. The innovation of this algorithm lies in: integrating HGnetv2 and lightweight convolutional LightConv to reconstruct the feature extraction network, reducing the number of parameters while optimizing the feature extraction ability; designing a gradient-weighted feature fusion pyramid network (GWFFPN), leveraging the progressive fusion and dynamic weighting strategies to effectively alleviate the semantic loss problem in the process of feature interaction between non-adjacent layers; constructing a dynamic multi-scale feature extraction module (DMF), embedding the simultaneous attention mechanism (SimAM) to achieve dynamic allocation of feature channels, precisely focusing on the core area of defects, and significantly improving the performance of identifying minor defects. Experimental results show that compared with the baseline model YOLOv8n, the HGFD-YOLOv8 algorithm reduces the number of parameters by 31.2% on the NEU-DET dataset for hot-rolled steel strip surface defects, increases the mean accuracy by 5.1%, and improves the precision and recall rates by 8.6% and 5.4% respectively. Compared with detection algorithms such as YOLOv13 and DEIM, the proposed defect detection algorithm has better detection performance and industrial application value.
JU Qian-kun, HOU Hong-ling, LI Xiang-yao, BIAN Yan-yan, YANG Ao, JIN Jian-qiang.
Hot Rolling Strip Steel Defect Detection Algorithm Based on Feature Fusion and Lightweight Network[J].
Science Technology and Engineering, 2026, 26(29): 12714-12725 DOI:10.12404/j.issn.1671-1815.2506411
实验中选用6种指标评判模型算法的性能,分别为平均精度值 (mean average precision,mAP)、召回率(recall,R)、准确率 (precision,P)、刷新率 (frames per second,FPS)、计算量(FLOPs)和参数量(parameters)。准确率指所有预测正确的样本占所有样本的比例,表示模型检测的准确程度;召回率表示检测正确样本占应该被正确检测到的样本的比例;mAP表示检测网络的综合性能,其中mAP0.5表示交并比阈值为0.5时的平均精度均值;FPS用来反映网络检测速度;Parameters表示模型的参数量,单位为M,即106;FLOPs表示在模型运行时所涉及的浮点运算量,它用来反映模型复杂程度,单位为G,即109。为验证改进模型在精度、mAP0.5和召回率指标上的提升并非随机因素导致,每组实验采用5次平行实验结果进行配对t检验(paired t-test),对改进前后数据进行显著性分析,设定显著性水平α=0.05(严格标准)。
LiuG H, MaQ W. Strip steel surface defect detecting method combined with a multi-layer attention mechanism network[J]. Measurement Science and Technology, 2023, 34(5): 055403.
[2]
杨琛. 超声检测技术研究进展[J]. 流体测量与控制, 2022, 3(3): 8-10.
[3]
YangChen. Research progress in ultrasonic detection technology[J]. Fluid Measurement and Control, 2022, 3(3): 8-10.
[4]
KimD, YuD, SawantA, et al. Remote detection of radioactive material using high-power pulsed electromagnetic radiation[J]. Nature Communications, 2017, 8: 15394.
WangXiaolei, JiZhigang, GuoXiangqian, et al. Current status and development trends of non-destructive testing technology for maintenance and remanufacturing of coal mining machinery equipment[J]. Science Technology and Engineering, 2021, 21(2): 423-433.
[7]
LiW H, ZhangH O, WangG L, et al. Deep learning based online metallic surface defect detection method for wire and arc additive manufacturing[J]. Robotics and Computer-Integrated Manufacturing, 2023, 80: 102470.
ZhouMengran, WangHaonan, GaoLipeng, et al. Steel surface defect detection based on YOLOv5s-FCS[J]. Science Technology and Engineering, 2024, 24(14): 5901-5910.
[14]
YangG, LeiJ, ZhuZ, et al. AFPN: asymptotic feature pyramid network for object detection[C]// 2023 IEEE International Conference on Systems, Man, and Cybernetics (SMC). Honolulu: IEEE, 2023: 2184-2189.
[15]
DingT, LiG, LiuZ, et al. Cross-scale edge purification network for salient object detection of steel defect images[J]. Measurement, 2022, 199: 111429.
[16]
WenX, LiS, YuH, et al. Multi-scale context feature and cross-attention network-enabled system and software-based for pavement crack detection[J]. Engineering Applications of Artificial Intelligence, 2024, 127: 107328.
JiZhongping, LinChenwei, HeZhiwei, et al. Steel strip surface defect detection integrating global and local perception and hierarchical feature fusion[J]. Computer Engineering and Applications, 2026, 62(4): 373-381.
[19]
LuM, ShengW, ZouY, et al. WSS-YOLO: an improved industrial defect detection network for steel surface defects[J]. Measurement, 2024, 236: 115060.
[20]
SohanM, SaiR T, RamiR C V. A review on YOLOv8 and its advancements[C]// International Conference on Data Intelligence and Cognitive Informatics. Singapore: Springer Singapore, 2024: 529-545.
[21]
PanboonyuenT, ThongbaiS, WongweeranimitW, et al. Object detection of road assets using transformer-based YOLOX with feature pyramid decoder on thai highway panorama[J]. Information, 2021, 13(1): 5.
[22]
AhmedK, TorresaniL. Star-caps: capsule networks with straight-through attentive routing[J]. Advances in Neural Information Processing Systems, 2019, 32: 9098-9107.
HuMingqi, ChenHuiming, XuWei, et al. Lightweight YOLOv8 steel surface defect detection with MobileNetv3 integration[J]. Science Technology and Engineering, 2025, 25(16): 6831-6840.
[25]
WangC Y, BochkovskiyA, LiaoH Y. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver: IEEE, 2023: 7464-7475.
[26]
LiT, ZhangY, LiQ, et al. AB-DLM: an improved deep learning model based on attention mechanism and BiFPN for driver distraction behavior detection[J]. IEEE Access, 2022, 10: 83138-83151.
[27]
KangM, TingC M, TingF F, et al. ASF-YOLO: a novel YOLO model with attentional scale sequence fusion for cell instance segmentation[J]. Image and Vision Computing, 2024, 147: 105057.
[28]
XuX, JiangY, ChenW, et al. DAMO-YOLO: a report on real-time object detection design[PP/OL]. arXiv, (2022-11-28)[2026-07-28]. https://arxiv.org/abs/2211.15444.arXiv:2211.15444.
[29]
WangC, HeW, NieY, et al. Gold-YOLO: efficient object detector via gather-and-distribute mechanism[J]. Advances in Neural Information Processing Systems, 2023, 36: 51094-51112.
[30]
WangR, ZhangY, PengL, et al. An efficient and precise fire detection model based on improved RT-DETR[C]// 2024 International Conference on Image Processing, Computer Vision and Machine Learning (ICICML). Shenzhen: IEEE, 2024: 1563-1567.
[31]
ChenY, ZhangC, ChenB, et al. Accurate leukocyte detection based on deformable-DETR and multi-level feature fusion for aiding diagnosis of blood diseases[J]. Computers in Biology and Medicine, 2024, 170: 107917.
[32]
WangY. An improved model based on RT-DETR for remote sensing object detection[C]//2024 6th International Conference on Frontier Technologies of Information and Computer (ICFTIC). Qingdao: IEEE, 2024: 836-840.
LiangLiming, LongPengwei, FengYao, et al. Improved lightweight VTG-YOLOv7-tiny for steel surface defect detection[J]. Optical Precision Engineering, 2024, 32(8): 1227-1240.
[35]
HuangS, LuZ, CunX, et al. DEIM: DETR with improved matching for fast convergence[PP/OL]. arXiv, (2024-12-05)[2026-07-28]. https://arxiv.org/abs/2412.04234.arXiv:2412.04234.
ZhuKangkai, HouHongling, YangAo, et al. Research on defect detection methods for interference-fit riveting[J]. Journal of Plastic Engineering, 2025, 32(2): 96-104.