连铸是钢铁生产过程中的一个关键环节,也是连接炼钢和轧钢的中间过程。为了降低连铸板坯号检测模型在实际部署时对计算机算力的要求,在保证连铸板坯号检测模型识别准确度的基础上,对板坯号检测模型进行了轻量化研究。首先,基于AD-PAN特征融合结构的检测算法,引入了MobileNetV3轻量级骨干网络来提取板坯号特征,旨在保持模型轻量化的同时提高图像分类性能。其次,对板坯号检测模型进行了协同互学习(Collaborative Mutual Learning, CML)蒸馏,旨在保证板坯号检测精度。最后,进行了试验对比以评估轻量化模型的性能,结果表明,通过模型轻量化研究牺牲了少量的模型精度,但大幅缩小了模型的参数量,并提高了模型的检测速度。
Abstract
Continuous casting is a key process in steel production, serving as an intermediate step between steelmaking and rolling. In order to reduce the computational power requirements of the slab identification model in practical deployment, the research on lightweighting the slab identification model while ensuring its accuracy was conducted. Initially, a detection algorithm based on the AD-PAN feature fusion structure incorporates the lightweight MobileNetV3 backbone network to extract features of the slab numbers, with the goal of enhancing image classification performance while maintaining the model's lightweight characteristic. Subsequently, the model underwent Collaborative Mutual Learning (CML) distillation to ensure the precision of slab number detection. Ultimately, experimental comparisons were conducted to assess the performance of the lightweight model. The outcomes demonstrate that although there was a modest trade-off in model accuracy due to the lightweight research, there was a significant reduction in the model's parameter volume and a marked improvement in the model's detection speed.
关键词
板坯号检测 /
模型部署 /
轻量化模型 /
CML蒸馏 /
MobileNet
{{custom_keyword}} /
Key words
slab number detection /
model deployment /
lightweight model /
CML distillation /
MobileNet
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 黄军,王宝峰,张雪元,等. 高温铸坯表面缺陷在线检测技术开发及应用[J]. 连铸,2022(6):61.
[2] 吴传开,黄锋,李斌,等. 基于改进深度可分离卷积和混合注意力机制的DMS-YOLOv8板坯检测算法[J]. 冶金自动化,2024,48(6):31.
[3] 陈云朋,周诗洋,陈勇勇,等. 人工智能技术在钢材带状碳化物检测中的应用[J]. 河北冶金,2025(1):76.
[4] 成彬,黄诺金,雷华,等. 融合深度学习与模型优化的连铸坯壳漏钢预报[J]. 钢铁研究学报,2024,36(7):885.
[5] CHOI S,YUN J P,KOO K,et al. A text localization algorithm in images for automatic recognition of slab identification numbers[C]//2009 European Control Conference (ECC). Budapest:IEEE,2009:4487.
[6] SHIVAKUMARA P,PHAN T Q,TAN C L. A laplacian approach to multi-oriented text detection in video[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2011,33(2):412.
[7] HUANG W L,LIN Z,YANG J C,et al. Text localization in natural images using stroke feature transform and text covariance descriptors[C]//2013 IEEE International Conference on Computer Vision. Sydney:IEEE,2013:1241.
[8] YI C C,TIAN Y L. Text string detection from natural scenes by structure-based partition and grouping[J]. IEEE Transactions on Image Processing:A Publication of the IEEE Signal Processing Society,2011,20(9):2594.
[9] LIAO M H,SHI B G,BAI X,et al. Textboxes:A fast text detector with a single deep neural network[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2017,31(1):4161.
[10] NIU S Z,LI X X,WANG M S,et al. A modified method for scene text detection by ResNet[J]. Computers,Materials and Continua,2020,65(3):2233.
[11] 魏伟,苏津磷,李帆,等. 基于改进PSENet与CRNN网络的智能电能表文本识别技术研究[J]. 电测与仪表,2023,60(12):176.
[12] WANG W H,XIE E Z,SONG X G,et al. Efficient and accurate Arbitrary-Shaped text detection with pixel aggregation network[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Seoul:IEEE,2019:8439.
[13] LEE S J,KWON W,KOO G,et al. Recognition of slab identification numbers using a fully convolutional network[J]. ISIJ International,2018,58(4):696.
[14] LEE S J,KIM S W. Localization of the slab information in factory scenes using deep convolutional neural networks[J]. Expert Systems with Applications,2017,77:34.
[15] 韩占光,周干水,谢长川. 基于机器学习的连铸坯低倍缺陷检测[J]. 连铸,2022(6):38.
[16] 盖晨阳,赵德颖,李海滨. PP-LCyolov5 s:一种轻量化的钢板目标表面检测算法[J/OL]. 机械科学与技术:1(2024-06-20). https://doi.org/10.13433/j.cnki.1003-8728.20240091.
[17] 田勇,王丙兴,张田,等. 基于机器视觉的钢板物料跟踪系统研究与应用[J]. 轧钢,2024,39(6):84.
[18] 刘乐,张晓松,黄锋,等. 基于改进DBNet和SVTR算法的连铸板坯号检测与识别[J]. 电子测量与仪器学报,2024,38(2):67.
[19] HOWARD A G,ZHU M L,CHEN B,et al. MobileNets:efficient convolutional neuralnetworks for mobile vision applications[EB/OL]. (2017-04-17). https://arxiv.org/abs/1704.04861.
[20] 任治洲,梁琨,王泽宇,等. 结合改进CBAM和MobileNetV2算法对小麦病斑粒分类[J]. 南京农业大学学报,2024,47(3):583.
[21] 崔金荣,魏文钊,赵敏. 基于改进MobileNetV3的水稻病害识别模型[J]. 农业机械学报,2023,54(11):217.
[22] DU Y N,LI C X,GUO R Y,et al. PP-OCRv2:Bag of tricks for ultra lightweight ocr system[EB/OL]. (2021-10-12). https://arxiv.org/abs/2109.03144.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
秦皇岛市科学技术研究与发展计划资助项目(202302B048);河钢集团重点科技资助项目(HG2021304);省级重点实验室绩效补助经费资助项目(22567619H)
{{custom_fund}}
