[1]林远达,陈海坤,叶钦,等.融合跨阶段动态多尺度注意力的带钢缺陷检测方法[J].华侨大学学报(自然科学版),2025,46(5):561-568.[doi:10.11830/ISSN.1000-5013.202508040]
　LIN Yuanda,CHEN Haikun,YE Qin,et al.Strip Steel Defect Detection Method of Integrating Cross-Stage Dynamic Multi-Scale Attention[J].Journal of Huaqiao University(Natural Science),2025,46(5):561-568.[doi:10.11830/ISSN.1000-5013.202508040]
点击复制

融合跨阶段动态多尺度注意力的带钢缺陷检测方法()
分享到：

《华侨大学学报（自然科学版）》[ISSN:1000-5013/CN:35-1079/N]

卷:
第46卷
期数:
2025年第5期
页码:
561-568
栏目:
出版日期:
2025-09-20

文章信息/Info

Title:
Strip Steel Defect Detection Method of Integrating Cross-Stage Dynamic Multi-Scale Attention
文章编号:
1000-5013(2025)05-0561-08
作者:
林远达1;  陈海坤1;  叶钦1;  潘书万1; 2;  郑力新1; 2
1. 华侨大学 工学院, 福建 泉州 362021;2. 华侨大学 物联网产业学院, 福建 泉州 362021
Author(s):
LIN Yuanda1;  CHEN Haikun1;  YE Qin1; PAN Shuwan1; 2;  ZHENG Lixin1; 2
1. College of Engineering, Huaqiao University, Quanzhou 362021, China; 2. College of Internet of Things Industry, Huaqiao University, Quanzhou 362021, China
关键词:
表面缺陷检测;  多尺度;  注意力;  带钢表面缺陷
Keywords:
surface defect detection;  multi-scale;  attention;  surface defect of steel strip
分类号:
TP317.4
DOI:
10.11830/ISSN.1000-5013.202508040
文献标志码:
A
摘要:
针对带钢表面缺陷检测在工业质量控制中面临的多尺度、形态复杂和尺寸微小等技术挑战,提出一种高效准确的检测方法。跨阶段动态多尺度注意力检测(CDMA-DET)模型集成跨阶段动态双流多尺度变压器模块、多尺度三元注意力池化模块和内容感知特征重组上采样模块,实现多尺度缺陷特征的自适应提取和融合。结果表明:在GC10-DET数据集上,CDMA-DET模型的mAP@0.50达到69.3%,mAP@0.50:0.95达到34.7%,相比基线模型分别提升5.5%和2.1%,模型参数量仅为2.5×106,浮点运算次数为6.7×109,推理速度达到234.51 帧·s-1;CDMA-DET模型有效平衡了检测精度、模型复杂度和推理效率。
Abstract:
To address the technical challenges of multi-scale characteristics, complex morphologies, and tiny defect sizes in steel surface detection for industrial quality control, a high-efficiency and high-accuracy detection method is proposed. The cross-stage dynamic multi-scale attention detection(CDMA-DET)model integrates a cross-stage dynamic dual-stream multi-scale transformer module,a multi-scale ternary attention pooling module, and a content-aware feature reorganization upsampling module, enabling adaptive extraction and fusion of multi-scale defect features. Experimental results on the GC10-DET dataset show that the proposed CDMA-DET achieves an mAP@0.50 of 69.3% and an mAP@0.50: 0.95 of 34.7%, outperforming the baseine model by 5.5% and 2.1%, respectively. The model contains only 2.5×106 parameters and 6.7×109 FLOPs, achieving an inference speed of 234.51 frames per second. These results demonstrate that CDMA-DET effectively balance detection accuracy, model complexity, and computational efficiency.

参考文献/References:

[1] TANG Ming,LI Yuanyuan,YAO Wei,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.
[2] LI Mengjiao,WANG Hao,WAN Zhibo.Surface defect detection of steel strips based on improved YOLOv4[J].Computers and Electrical Engineering,2022,102:108208.DOI:10.1016/j.compeleceng.2022.108208.
[3] DU Yongzhao,CHEN Haixin,FU Yuqing,et al.AFF-Net: A strip steel surface defect detection network via adaptive focusing features[J].IEEE Transactions on Instrumentation and Measurement,2024,73:1-14.DOI:10.1109/TIM.2024.3398131.
[4] HUANG Xiaohua,ZHU Jiahao,HUO Ying.SSA-YOLO: An improved YOLO for hot-rolled strip steel surface defect detection[J].IEEE Transactions on Instrumentation and Measurement,2024,73:1-17.DOI:10.1109/TIM.2024.3488136.
[5] ZHOU Qiqi,WANG Haichao.CABF-YOLO: A precise and efficient deep learning method for defect detection on strip steel surface[J].Pattern Analysis and Applications,2024,27(2):36.DOI:10.1007/s10044-024-01252-5.
[6] LU Jiaobo,ZHU Mingrui,MA Xiaoya,et al.Steel strip surface defect detection method based on improved YOLOv5s[J].Biomimetics,2024,9(1):28.DOI:10.3390/biomimetics9010028.
[7] SHEN Kunye,ZHOU Xiaofei,LIU Zhi.MINet: Multiscale interactive network for real-time salient object detection of strip steel surface defects[J].IEEE Transactions on Industrial Informatics,2024,20(5):7842-7852.DOI:10.1109/TII.2024.3366221.
[8] WANG C Y,LIAO H M,WU Y H,et al.A new backbone that can enhance learning capability of CNN.2020 IEEE[C]//CVF Conference on Computer Vision and Pattern Recognition Workshops.Seattle:IEEE Press,2020:390-391.DOI:10.1109/CVPRW50498.2020.00203.
[9] VASWANI A,SHAZEER N,PARMAR N,et al.Attention is all you need[J].Advances in Neural Information Processing Systems,2017,30:5998-6008.DOI:10.5555/3295222.3295349.
[10] MISRA D,NALAMADA T,ARASANIPLLAI A U,et al.Rotate to attend: Convolutional triplet attention module[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision.Waikoloa:IEEE Press,2021:3139-3148.DOI:10.1109/WACV48630.2021.00318.
[11] WANG Jiaqi,CHEN Kai,XU Rui,et al.CARAFE: Content-aware reassembly of features[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.Waikoloa:IEEE Press,2019:3007-3016.DOI:10.1109/ICCV.2019.00310.
[12] Lü Xiaoming,DUAN Fajie,JIANG Jiajia,et al.Deep metallic surface defect detection: The new benchmark and detection network[J].Sensors,2020,20(6):1562.DOI:10.3390/s20061562.
[13] GUI Zili,GENG Jianping.YOLO-ADS: An improved YOLOv8 algorithm for metal surface defect detection[J].Electronics,2024,13(16):3129.DOI:10.3390/electronics13163129.
[14] WANG Ao,CHEN Hui,LIU Lihao,et al.YOLOV10: Real-time end-to-end object detection[J].Advances in Neural Information Processing Systems,2024,37:107984-108011.DOI:10.5555/3737916.3741345.
[15] ZHAO Yian,Lü Wenyu,XU Shangliang,et al.Detrs beat yolos on real-time object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Seattle:IEEE Press,2024:16965-16974.DOI:10.1109/CVPR52733.2024.01605.
[16] GUO Shijing,LI Bohan,ZHANG Jinjing,et al.SDS-YOLOv8n: A lightweight detection method for flames and smoke[C]//Journal of Physics: Conference Series.Changchun: Institute of Physics Publishing,2024:012020.DOI:10.1088/1742-6596/2858/1/012020.
[17] LI Chengfei,WEN Zhikai,HUANG Haijian,et al.An efficient lightweight detection model for steel surface defects with dynamic deformable head[J].Engineering Research Express,2025,7(1):015282.DOI:10.1088/2631-8695/adbab4.
[18] HUANG Shihua,LU Zhichao,CUN Xiaodong,et al.DEIM: DETR with improved matching for fast convergence[C]//Proceedings of the Computer Vision and Pattern Recognition Conference.Nashville: IEEE Press,2025:15162-15171.DOI:10.1109/CVPR52734.2025.01412.
[19] HUANG Weibo,WEI Peng,ZHANG Manhua,et al.HRIPCB: A challenging dataset for PCB defects detection and classification[J].The Journal of Engineering,2020,2020(13):303-309.DOI:10.1049/joe.2019.1183.

相似文献/References:

[1]刘韶涛,姚灿荣.结合PSO的改进压缩跟踪方法[J].华侨大学学报(自然科学版),2017,38(1):121.[doi:10.11830/ISSN.1000-5013.201701024]
　LIU Shaotao,YAO Canrong.Improved Compress Tracking Algorithm Based on PSO[J].Journal of Huaqiao University(Natural Science),2017,38(5):121.[doi:10.11830/ISSN.1000-5013.201701024]
[2]吴琼,陈锻生.多尺度卷积循环神经网络的情感分类技术[J].华侨大学学报(自然科学版),2017,38(6):875.[doi:10.11830/ISSN.1000-5013.201606077]
　WU Qiong,CHEN Duansheng.Sentiment Classification With Multiscale Convolutional Recurrent Neural Network[J].Journal of Huaqiao University(Natural Science),2017,38(5):875.[doi:10.11830/ISSN.1000-5013.201606077]
[3]林文华,房怀英,范璐璐,等.采用双相机多尺度方法的机制砂级配测量及空隙率预测[J].华侨大学学报(自然科学版),2022,43(3):285.[doi:10.11830/ISSN.1000-5013.202112021]
　LIN Wenhua,FANG Huaiying,FAN Lulu,et al.Gradation Measurement and Void Content Prediction of Manufactured Sand Using Dual Camera Multi-Scale Method[J].Journal of Huaqiao University(Natural Science),2022,43(5):285.[doi:10.11830/ISSN.1000-5013.202112021]

备注/Memo

备注/Memo:
收稿日期: 2025-08-04
通信作者: 潘书万(1982-),男,副教授,博士,主要从事工业制造缺陷智能检测的研究。E-mail:shuwanpan@hqu.edu.cn。
基金项目: 福建省高校产学合作项目(2022H6013); 福建省泉州市科技局高层次人才创新创业项目(2021C047R)
更新日期/Last Update: 2025-09-20