HSED-YOLO:一种轻量化的带钢表面缺陷检测模型

摘要/Abstract

参考文献

相关文章 15

Metrics

doi:10.16088/j.issn.1001-6600.2024051502

摘要： 针对当前带钢表面缺陷检测算法计算复杂度高、检测精度较低、容易产生漏检和误检等问题,本文提出一种轻量化的带钢表面缺陷检测模型HSED-YOLO。首先,将原始YOLOv8n主干网络更换为改进后的HGNetV2,减少特征图计算冗余,从而降低模型的参数量。然后,为了进一步降低模型的复杂度,在模型颈部网络结构中引入Slim-Neck结构化设计;同时,在特征融合阶段引入EMA(efficient multi-scale attention module)注意力机制,提高模型的特征提取能力;为了进一步提高模型的检测精度,使用DIoU损失函数设计。最后,在带钢缺陷数据集上进行大量实验,得到改进后模型的参数量和计算量分别为2.1×10⁶和6.1×10⁹,仅为基准模型的70%和75.3%,并且平均精度相比于基准模型提升2个百分点,表明改进模型是有效的。

关键词: 缺陷检测, 带钢, YOLOv8, 注意力机制, 损失函数, 图像识别

Abstract: In response to the high computational complexity, low detection accuracy, and the issues of missed detection and false alarms in the current strip steel surface defect detection algorithm, a lightweight strip steel surface defect detection model, HSED-YOLO, is proposed. Initially, the original YOLOv8n backbone network is replaced with the improved HGNetV2, reducing the redundancy in feature map computation and thereby decreasing the number of the model’s parameters. Subsequently, to further reduce the model’s complexity, a Slim-Neck structured design is introduced into the model’s bottleneck network structure. Concurrently, an EMA attention mechanism is introduced during the feature fusion stage to enhance the model’s feature extraction capability. To further improve the model’s detection accuracy, a DIoU loss function is designed. Extensive experiments are conducted on the strip steel defect dataset. The number of improved model’s parameters and computational load are 2.1×10⁶ and 6.1×10⁹ FLOPs, respectively, which are only 70% and 75.3% of those of the baseline model. Moreover, the average accuracy is improved by 2% compared with the baseline model. These results demonstrate the effectiveness of the improved network.

Key words: defect detection, strip steel, YOLOv8, attention mechanism, loss function, image recognition

中图分类号: TP391.41

戴林华, 黎远松, 石睿, 何忠良, 李雷. HSED-YOLO:一种轻量化的带钢表面缺陷检测模型[J]. 广西师范大学学报（自然科学版）, 2025, 43(2): 95-106.

DAI Linhua, LI Yuansong, SHI Rui, HE Zhongliang, LI Lei. HSED-YOLO: A Lightweight Model for Detecting Surface Defects in Strip Steel[J]. Journal of Guangxi Normal University(Natural Science Edition), 2025, 43(2): 95-106.

[1] WANG Z P, WANG J, CHEN S. Fault location of strip steel surface quality defects on hot-rolling production line based on information fusion of historical cases and process data[J]. IEEE Access, 2020, 8: 171240-171251. DOI: 10.1109/ACCESS.2020.3024582.
[2] 冯毅雄, 赵彬, 郑浩, 等. 集成迁移学习的轴件表面缺陷实时检测[J]. 计算机集成制造系统, 2019, 25(12): 3199-3208. DOI: 10.13196/j.cims.2019.12.021.
[3] SONG K C, YAN Y H. A noise robust method based on completed local binary patterns for hot-rolled steel strip surface defects[J]. Applied Surface Science, 2013, 285(Part B): 858-864. DOI: 10.1016/j.apsusc.2013.09.002.
[4] 李少波, 杨静, 王铮, 等. 缺陷检测技术的发展与应用研究综述[J]. 自动化学报, 2020, 46(11): 2319-2336. DOI: 10.16383/j.aas.c180538.
[5] SI B L, YASENGJIANG M, WU H W. Deep learning-based defect detection for hot-rolled strip steel[J]. Journal of Physics: Conference Series, 2022, 2246(1): 012073. DOI: 10.1088/1742-6596/2246/1/012073.
[6] 翁玉尚, 肖金球, 夏禹. 改进Mask R-CNN算法的带钢表面缺陷检测[J]. 计算机工程与应用, 2021, 57(19): 235-242. DOI: 10.3778/j.issn.1002-8331.2010-0446.
[7] TANG M, LI Y Y, YAO W, et al. A strip steel surface defect detection method based on attention mechanism and multi-scale maxpooling[J]. Measurement Science and Technology, 2021, 32(11): 115401. DOI: 10.1088/1361-6501/ac0ca8.
[8] LIU W, ANGUELOV D, ERHAN D, et al.SSD: single shot multibox detector[C] // Computer Vision-ECCV 2016: LNCS Volume 9905. Cham: Springer, 2016: 21-37. DOI: 10.1007/978-3-319-46448-0_2.
[9] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C] // 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2016: 779-788. DOI: 10.1109/CVPR.2016.91.
[10] ZHAO H, WAN F, LEI G B, et al. LSD-YOLOv5: a steel strip surface defect detection algorithm based on lightweight network and enhanced feature fusion mode[J]. Sensors, 2023, 23(14): 6558. DOI: 10.3390/s23146558.
[11] 王春梅, 刘欢. YOLOv8-VSC:一种轻量级的带钢表面缺陷检测算法[J]. 计算机科学与探索, 2024, 18(1): 151-160. DOI: 10.3778/j.issn.1673-9418.2308060.
[12] LI Y S, XU S B, ZHU Z F, et al. EFC-YOLO: an efficient surface-defect-detection algorithm for steel strips[J]. Sensors, 2023, 23(17): 7619. DOI: 10.3390/s23177619.
[13] WANG G Q, ZHANG C Z, CHEN M S, et al. A high-accuracy and lightweight detector based on a graph convolution network for strip surface defect detection[J]. Advanced Engineering Informatics, 2024, 59: 102280. DOI: 10.1016/J.AEI.2023.102280.
[14] 张阳婷, 黄德启, 王东伟,等. 基于深度学习的目标检测算法研究与应用综述[J]. 计算机工程与应用, 2023, 59(18): 1-13. DOI: 10.3778/j.issn.1002-8331.2305-0310.
[15] ZHAO Y, LV W Y, XU S L, et al. DETRs beat YOLOs on real-time object detection[EB/OL]. (2024-04-03)[2024-05-15]. https://arxiv.org/abs/2304.08069. DOI: 10.48550/arXiv.2304.08069.
[16] HAN K,WANG Y H,TIAN Q,et al. GhostNet: more features from cheap operations[C] // 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2020: 1577-1586. DOI: 10.1109/CVPR42600.2020.00165.
[17] LI H L, LI J, WEI H B, et al. Slim-neck by GSConv: a lightweight-design for real-time detector architectures[J]. Journal of Real-Time Image Processing, 2024, 21: 62. DOI: 10.1007/s11554-024-01436-6.
[18] QIN X Y, LI N, WENG C, et al. Simple attention module based speaker verification with iterative noisy label detection[C] // ICASSP 2022: 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Piscataway, NJ: IEEE, 2022: 6722-6726. DOI: 10.1109/ICASSP43922.2022.9746294.
[19] LAU K W, PO L M, REHMAN Y A U. Large separable kernel attention:rethinking the large kernel attention design in CNN[J]. Expert Systems with Applications, 2024, 236: 121352. DOI: 10.1016/J.ESWA.2023.121352.
[20] OUYANG D L, HE S, ZHANG G Z, et al. Efficient multi-scale attention module with cross-spatial learning[C] // ICASSP 2023: 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Piscataway, NJ: IEEE, 2023: 1-5. DOI: 10.1109/ICASSP49357.2023.10096516.
[21] 王博韬. 基于YOLOv4的车辆目标检测改进算法[J]. 电脑编程技巧与维护, 2022(9): 31-33. DOI: 10.16184/j.cnki.comprg.2022.09.012.
[22] 田枫, 贾昊鹏, 刘芳. 改进YOLOv5的油田作业现场安全着装小目标检测[J]. 计算机系统应用, 2022, 31(3): 159-168. DOI: 10.15888/j.cnki.csa.008359.
[23] 杜孟新, 毕玉, 杜鹏昊. 基于卷积神经网络的带钢表面缺陷图像检测算法[J]. 火力与指挥控制, 2022, 47(8): 132-135. DOI: 10.3969/j.issn.1002-0640.2022.08.021.
[24] 胡海涛, 杜昊晨, 王素琴, 等. 改进YOLOX的药品泡罩铝箔表面缺陷检测方法[J]. 图学学报, 2022, 43(5): 803-814. DOI: 10.11996/JG.j.2095-302X.2022050803.
[25] 杨博. 基于深度学习的高铁接触网4C故障检测软件设计与实现[D]. 武汉: 华中科技大学, 2019. DOI: 10.27157/d.cnki.ghzku.2019.003062.
[26] ZHANG X R, WANG Y L, FANG H S. Steel surface defect detection algorithm based on ESI-YOLOv8[J]. Materials Research Express, 2024, 11(5): 056509. DOI: 10.1088/2053-1591/ad46ec.

[1]	郭翔羽, 石天怡, 陈燕楠, 南新元, 蔡鑫. 基于YOLO-CDBW模型的列车接触网异物检测研究[J]. 广西师范大学学报（自然科学版）, 2025, 43(2): 56-69.
[2]	苏春海, 夏海英. 抗噪声双约束网络的面部表情识别[J]. 广西师范大学学报（自然科学版）, 2025, 43(2): 70-82.
[3]	刘玉娜, 马双宝. 基于改进YOLOv8n的轻量化织物疵点检测算法[J]. 广西师范大学学报（自然科学版）, 2025, 43(2): 83-94.
[4]	刘俊杰, 马凯, 黄泽华, 田苗, 邱芹军, 陶留锋, 谢忠. 基于迁移学习及通道先验注意力机制的地质构造识别[J]. 广西师范大学学报（自然科学版）, 2025, 43(2): 107-120.
[5]	余快, 宋宝贵, 邵攀, 余翱. 基于层级尺度交互的U-Net遥感影像建筑物提取方法[J]. 广西师范大学学报（自然科学版）, 2025, 43(2): 121-132.
[6]	卢家辉, 陈庆锋, 王文广, 余谦, 何乃旭, 韩宗钊. 基于多尺度注意力的器官图像分割方法[J]. 广西师范大学学报（自然科学版）, 2024, 42(6): 138-148.
[7]	于雪松, 徐宝. 加权p、q对称损失函数下Burr分布参数的Bayes估计[J]. 广西师范大学学报（自然科学版）, 2024, 42(6): 186-193.
[8]	杜帅文, 靳婷. 基于用户行为特征的深度混合推荐算法[J]. 广西师范大学学报（自然科学版）, 2024, 42(5): 91-100.
[9]	田晟, 胡啸. 基于Transformer模型的车辆轨迹预测[J]. 广西师范大学学报（自然科学版）, 2024, 42(3): 47-58.
[10]	王天雨, 袁嘉伟, 齐芮, 李洋. 多类型知识增强的微博立场检测模型[J]. 广西师范大学学报（自然科学版）, 2024, 42(1): 79-90.
[11]	肖宇庭, 吕晓琪, 谷宇, 刘传强. 基于拆分残差网络的糖尿病视网膜病变分类[J]. 广西师范大学学报（自然科学版）, 2024, 42(1): 91-101.
[12]	席凌飞, 伊力哈木·亚尔买买提, 刘雅洁. 基于改进YOLOv5的铝型材表面缺陷检测方法[J]. 广西师范大学学报（自然科学版）, 2024, 42(1): 111-119.
[13]	宋冠武, 陈知明, 李建军. 基于ResNet-50的级联注意力遥感图像分类[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 80-91.
[14]	郭嘉梁, 靳婷. 基于语义增强的多模态情感分析[J]. 广西师范大学学报（自然科学版）, 2023, 41(5): 14-25.
[15]	吴正清, 曹晖, 刘宝锴. 基于注意力卷积神经网络的中文虚假评论检测[J]. 广西师范大学学报（自然科学版）, 2023, 41(5): 26-36.

Just accepted

Online first

Just accepted

Online first

Viewed

Full text

Abstract

Just accepted	Online first	Issue

0	0	4

From	Others	local

Times	3	1
Rate	75%	25%

Cited

Web of Science	Crossref	ScienceDirect	Search for Citations in Google Scholar >>


This page requires you have already subscribed to WoS.

Shared

Discussed