基于响应选择策略的连铸坯表面缺陷检测模型迭代方法

doi:10.3901/JME.260492

机械工程学报 ›› 2026, Vol. 62 ›› Issue (10): 73-82.doi: 10.3901/JME.260492

• 仪器科学与技术 • 上一篇

扫码分享

基于响应选择策略的连铸坯表面缺陷检测模型迭代方法

曹翔宇¹, 徐科¹, 张含鑫¹, 黄磊^2,3, 黄学忠^2,3, 韦毅^4,5

1. 北京科技大学钢铁共性技术协同创新中心北京 100083;
2. 广西北港新材料有限公司北海 536000;
3. 广西特钢新材料重点实验室北海 536000;
4. 广西产研院新型功能材料研究所有限公司南宁 530233;
5. 重庆大学化学化工学院重庆 401331

收稿日期:2025-05-13 修回日期:2025-12-08 发布日期:2026-07-29
作者简介:曹翔宇,男,1992年出生,博士研究生。主要研究方向为机器视觉和智能检测。E-mail:18810610485@163.com;徐科(通信作者),男,1972年出生,博士,研究员,博士研究生导师。主要研究方向为机器视觉、深度学习、智能检测。E-mail:xuke@ustb.edu.cn
基金资助:
国家重点研发计划课题资助项目(2021YFB3202403)。

Model Iteration Method for Surface Defect Detection of Continuous Casting Slabs Based on Response Selection Strategy

CAO Xiangyu¹, XU Ke¹, ZHANG Hanxin¹, HUANG Lei^2,3, HUANG Xuezhong^2,3, WEI Yi^4,5

1. Collaborative Innovation Center of Steel Technology, University of Science and Technology Beijing, Beijing 100083;
2. China Guangxi Beigang New Materials Co., Ltd., Beihai 536000;
3. Guangxi Key Laboratory of New Materials for Special Steel, Beihai 536000;
4. Institute of Novel Functional Materials, Guangxi Institute of Industry and Research, Nanning 530233;
5. College of Chemistry and Chemical Engineering, Chongqing University, Chongqing 401331

Received:2025-05-13 Revised:2025-12-08 Published:2026-07-29

摘要/Abstract

摘要： 在连铸板坯的生产过程中，表面缺陷样本数据稀少，使得检测模型需要不断的迭代更新。传统的迭代训练方式不仅耗时长、计算资源消耗大，且容易丢失先前学习到的关键特征知识。为解决上述问题，提出一种基于响应选择策略的增量学习方法，用于训练YOLOv8检测器。该方法基于知识蒸馏框架，设计一种依据统计规律过滤无效知识的响应选择策略，使学生模型能够更精准地继承教师模型的有效特征，从而增强蒸馏效果。通过在MS COCO 2017数据集与实际采集的连铸坯表面缺陷数据集上进行验证试验，评估该方法的有效性。试验结果表明，在仅使用新增样本进行微调的条件下，该方法能显著缓解模型遗忘问题，使检测器在铸坯缺陷数据集上的平均精度(Average precision，AP)下降幅度减少12.9。相比于传统全量数据重训练的耗时迭代方式，该策略虽然仅存在3.3 AP的精度差距，但极大地降低了计算资源开销和时间成本。研究表明，该迭代策略能够在保证一定检测精度的前提下，实现连铸坯表面缺陷检测模型的快速、低成本迭代，具有良好的工程应用价值。

关键词: 缺陷检测, 增量学习, 知识蒸馏, 连铸坯, 模型迭代

Abstract: In the production of continuous casting slabs, the scarcity of surface defect sample data necessitates frequent iterative updates of detection models. Traditional iterative training methods are time-consuming, computationally expensive, and easily lead to the loss of previously learned critical features. To address these issues, a model iteration method based on a response selection strategy is proposed to train the YOLOv8 detector via incremental learning. Based on the knowledge distillation framework, a response selection strategy utilizing statistical principles is designed to filter out ineffective knowledge, enabling the student model to inherit effective features from the teacher model more accurately and enhancing the distillation effect. The proposed method is validated through experiments on the MS COCO 2017 dataset and a collected surface defect dataset of continuous casting slabs. Experimental results indicate that under the condition of fine-tuning only with new samples, the degree of forgetting is significantly alleviated, reducing the drop in average precision(AP) by 12.9 on the casting defect dataset. Compared to traditional time-consuming iterative methods involving retraining on full data, the proposed strategy exhibits a gap of only 3.3 AP while drastically reducing computational resource consumption and time costs. It is concluded that this iterative strategy achieves rapid and low-cost iteration of the surface defect detection model for continuous casting slabs while maintaining acceptable detection accuracy, demonstrating significant engineering application value.

Key words: defect detection, incremental learning, knowledge distillation, continuous casting slabs, modal iteration

中图分类号:

TG142

曹翔宇, 徐科, 张含鑫, 黄磊, 黄学忠, 韦毅. 基于响应选择策略的连铸坯表面缺陷检测模型迭代方法[J]. 机械工程学报, 2026, 62(10): 73-82.

CAO Xiangyu, XU Ke, ZHANG Hanxin, HUANG Lei, HUANG Xuezhong, WEI Yi. Model Iteration Method for Surface Defect Detection of Continuous Casting Slabs Based on Response Selection Strategy[J]. Journal of Mechanical Engineering, 2026, 62(10): 73-82.

导出引用管理器 EndNote|Reference Manager|ProCite|BibTeX|RefWorks

链接本文: http://www.cjmenet.com.cn/CN/10.3901/JME.260492

http://www.cjmenet.com.cn/CN/Y2026/V62/I10/73

参考文献

[1] 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. LI Shaobo，YANG Jing，WANG Zheng，et al. Review of development and application of defect detection technology[J]. Acta Automatica Sinica，2020，46(11)：2319-2336.
[2] 秦飞红，袁懋诞，祝端阳，等. 基于多层特征融合目标检测网络的超声全聚焦钢板焊缝缺陷自动检测研究[J]. 机械工程学报，2025，61(4)：55-66. QIN Feihong，YUAN Maodan，ZHU Duanyang，et al. Automatic weld defect detection in steel plates based on ultrasonic total focusing method and multi-layer feature fusion target detection network[J]. Journal of Mechanical Engineering，2025，61(4)：55-66.
[3] 马智华，曾凯，陈波，等. 基于改进YOLOv7的连铸坯表面缺陷检测算法[J]. 中国冶金，2024，34(7)：101-112. MA Zhihua，ZENG Kai，CHEN Bo，et al. Surface defect detection algorithm for continuous casting billets based on improved YOLOv7[J]. China Metallurgy，2024，34(7)：101-112.
[4] Belouadah E，Popescu A，Kanellos I. A comprehensive study of class incremental learning algorithms for visual tasks[J]. Neural Networks，2021，135：38-54.
[5] RAJASEGARAN J，KHAN S，HAYAT M，et al. iTAML：An incremental task-agnostic meta-learning approach[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York：IEEE，2020：13588-13597.
[6] Shieh J L，Haq Q M，Haq M A，et al. Continual learning strategy in one-stage object detection framework based on experience replay for autonomous driving vehicle[J]. Sensors，2020，20(23)：6777.
[7] Zhao X，Wang L，Zhang Y，et al. A review of convolutional neural networks in computer vision[J]. Artificial Intelligence Review，2024，57(4)：99.
[8] LI D，TASCI S，GHOSH S，et al. RILOD：Near real-time incremental learning for object detection at the edge[C]//Proceedings of the 4th ACM/IEEE Symposium on Edge Computing (SEC). New York：ACM，2019：113-126.
[9] KANG M，PARK J，HAN B. Class-incremental learning by knowledge distillation with adaptive feature consolidation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York：IEEE，2022：16071-16080.
[10] Li Z，Hoiem D. Learning without forgetting[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence，2017，40(12)：2935-2947.
[11] Ma Y，Ma X，Hao T，et al. Knowledge distillation via hierarchical matching for small object detection[J]. Journal of Computer Science and Technology，2024，39(4)：798-810.
[12] Guo Z，Zhang P，Liang P. Shared knowledge distillation network for object detection[J]. Electronics，2024，13(8)：1595.
[13] TAO X，HONG X，CHANG X，et al. Few-shot class-incremental learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York：IEEE，2020：12183-12192.
[14] Tian S，Li L，Li W，et al. A survey on few-shot class-incremental learning[J]. Neural Networks，2024，169：307-324.
[15] JOCHER G，CHAURASIA A，QIU J. Ultralytics YOLOv8[EB/OL]. (2023-01-10)[2024-01-18]. https:// github.com/ultralytics/ultralytics.
[16] SOHAN M，SAI RAM T，RAMI REDDY C V. A review on YOLOv8 and its advancements[C]//Proceedings of International Conference on Data Intelligence and Cognitive Informatics (ICDICI). Singapore：Springer，2024：529-545.
[17] JOCHER G. YOLOv5 by Ultralytics[R/OL]. Version 7.0. Zenodo，2020[2024-01-16]. https://doi.org/10.5281/ zenodo.3908559.
[18] He K，Zhang X，Ren S，et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence，2015，37(9)：1904-1916.
[19] LIN T Y，DOLLÁR P，GIRSHICK R，et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York：IEEE，2017：2117-2125.
[20] ZHANG H，WANG Y，DAYOUB F，et al. VarifocalNet：An IoU-aware dense object detector[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York：IEEE，2021：8514-8523.
[21] Gou J，Yu B，Maybank S J，et al. Knowledge distillation：A survey[J]. International Journal of Computer Vision，2021，129(6)：1789-1819.
[22] Tian Y，Pei S，Zhang X，et al. Knowledge distillation on graphs：A survey[J]. ACM Computing Surveys，2025，57(8)：1-16.
[23] FENG T，WANG M，YUAN H. Overcoming catastrophic forgetting in incremental object detection via elastic response distillation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York：IEEE，2022：9427-9436.
[24] Liu S，Yao Y，Jia J，et al. Rethinking machine unlearning for large language models[J]. Nature Machine Intelligence，2025，7：181-194.
[25] Peng C，Zhao K，Maksoud S，et al. Sid：Incremental learning for anchor-free object detection via selective and inter-related distillation[J]. Computer Vision and Image Understanding，2021，210：103229.

基于响应选择策略的连铸坯表面缺陷检测模型迭代方法

Model Iteration Method for Surface Defect Detection of Continuous Casting Slabs Based on Response Selection Strategy

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	何超江, 李鹏, 何存富, 王钰珏, 刘秀成. 基于Terfenol-D与光纤光栅的高灵敏度磁场传感器[J]. 机械工程学报, 2025, 61(8): 1-8.
[2]	石杰, 吴昆鹏, 杨朝霖, 邓能辉, 王少聪, 苏成. 融合灰度和深度特征的钢管表面缺陷检测方法[J]. 机械工程学报, 2025, 61(4): 32-43.
[3]	曾治霖, 瞿昊, 杜正春. 基于深度学习和生成对抗网络的发动机缸体表面缺陷检测方法[J]. 机械工程学报, 2025, 61(2): 46-55.
[4]	余智祺, 徐洋, 王元飞, 盛晓伟. 基于多尺度标准化流模型的布匹缺陷检测[J]. 机械工程学报, 2025, 61(10): 241-249.
[5]	李可, 于成昊, 明雪飞, 顾杰斐, 宿磊, MICHAEL Pecht. 基于VMD-共振频带-能量熵的倒装焊芯片缺陷检测方法[J]. 机械工程学报, 2025, 61(10): 250-262.
[6]	韦逍遥, 刘文渊, 王龄裕, 丁斌杰, 曹衍龙, 杨将新, 黄芳. 基于光度立体视觉的钢轨焊缝打磨表面缺陷检测[J]. 机械工程学报, 2024, 60(24): 11-24.
[7]	傅杨, 张跃, 毛颖, 唐小华, 陈祖高, 徐和武, 杨雨沛, 高斌, 田贵云. 基于Feature Boosting的管道电磁内检测多传感信号缺陷解析算法[J]. 机械工程学报, 2024, 60(20): 51-67.
[8]	程进杰, 石文泽, 卢超, 魏运飞, 陈尧, 陈果. 脉冲压缩技术在高温连铸电磁超声测厚应用研究[J]. 机械工程学报, 2023, 59(8): 20-31.
[9]	郎宁, 王德成, 程鹏. 基于集成自适应欠采样的铝管表面缺陷检测方法研究[J]. 机械工程学报, 2023, 59(6): 18-31.
[10]	陆向宁, 刘凡, 何贞志, 廖广兰, 史铁林. 稀疏重构SAM芯片焊点检测方法研究[J]. 机械工程学报, 2023, 59(6): 95-102.
[11]	周健, 郑联语, 樊伟, 张学鑫, 曹彦生. 工业机器人定位误差在线自适应补偿[J]. 机械工程学报, 2023, 59(5): 53-66.
[12]	宿磊, 王立建, 祁阳, 张思雨, 顾杰斐, 李可. 基于IADSA深度迁移网络的金属表面缺陷检测[J]. 机械工程学报, 2023, 59(24): 46-55.
[13]	刘增华, 吴育衡, 王可心, 满润昕, 何存富, 吴斌. 基于太赫兹时域光谱技术的陶瓷基复合材料缺陷检测成像研究[J]. 机械工程学报, 2023, 59(14): 33-42.
[14]	吴强威, 文龙. 基于联合对抗训练的深度学习表面缺陷检测方法[J]. 机械工程学报, 2023, 59(12): 173-182.
[15]	张仕军, 金振林. 基于多模型级联的双目视觉铸件缺陷检测方法[J]. 机械工程学报, 2022, 58(5): 34-43.