CN 41-1243/TG ISSN 1006-852X

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于CBAM-ResNet50的金刚石颗粒净度检测方法

费文倩 赵凤霞 杜全斌 王庆海

费文倩, 赵凤霞, 杜全斌, 王庆海. 基于CBAM-ResNet50的金刚石颗粒净度检测方法[J]. 金刚石与磨料磨具工程, 2024, 44(5): 588-598. doi: 10.13394/j.cnki.jgszz.2023.0153
引用本文: 费文倩, 赵凤霞, 杜全斌, 王庆海. 基于CBAM-ResNet50的金刚石颗粒净度检测方法[J]. 金刚石与磨料磨具工程, 2024, 44(5): 588-598. doi: 10.13394/j.cnki.jgszz.2023.0153
FEI Wenqian, ZHAO Fengxia, DU Quanbin, WANG Qinghai. Diamond particle clarity detection method based on CBAM-ResNet50[J]. Diamond & Abrasives Engineering, 2024, 44(5): 588-598. doi: 10.13394/j.cnki.jgszz.2023.0153
Citation: FEI Wenqian, ZHAO Fengxia, DU Quanbin, WANG Qinghai. Diamond particle clarity detection method based on CBAM-ResNet50[J]. Diamond & Abrasives Engineering, 2024, 44(5): 588-598. doi: 10.13394/j.cnki.jgszz.2023.0153

基于CBAM-ResNet50的金刚石颗粒净度检测方法

doi: 10.13394/j.cnki.jgszz.2023.0153
基金项目: 河南省超硬材料智能制造装备集成重点实验室开放课题(JDKJ2022-03)。
详细信息
    通讯作者:

    赵凤霞,女,1971年生,博士、教授。主要研究方向:现代精密测量技术及应用。E-mail:zfxmail@163.com

  • 中图分类号: TQ164; TP391.4

Diamond particle clarity detection method based on CBAM-ResNet50

  • 摘要: 针对金刚石颗粒净度传统检测方法效率低、准确率差的问题,提出了一种基于迁移学习和改进ResNet50的金刚石颗粒净度检测算法CBAM-ResNet50。该算法通过在ResNet50主干网络的每层中增加CBAM,以提升模型特征的提取能力;且在主干网络的Layer3和Layer4中融入FPN结构,对提取的特征进行部分特征聚合,来解决采样过程中小目标特征易丢失的问题;同时引入迁移学习方法,用交叉熵损失函数优化模型的初始参数,提升模型的泛化能力。结果表明:在学习率设置为0.000 1时,提出的CBAM-ResNet50模型训练准确率达到99.2%;根据混淆矩阵计算得到模型的精确度在99.20%以上,特异性在99.70以上%,F1分数在99.20%,分类召回率在98.70%以上,优于其他主流分类网络的结果,有效提高了金刚石颗粒净度检测的识别能力。

     

  • 图  1  CBAM-ResNet50的模型结构

    Figure  1.  Model structure of CBAM-ResNet50

    图  2  残差单元的结构

    Figure  2.  Structure of residual units

    图  3  改进后的残差结构

    Figure  3.  Improved residual structurs

    图  4  金刚石图像及4层Layer提取的特征

    Figure  4.  Diamond image and features extracted from four layers

    图  5  金刚石图像及4层CBAM提取的特征

    Figure  5.  Diamond image and features extracted from four CBAM

    图  6  CBAM结构

    Figure  6.  CBAM structure

    图  7  特征融合流程示意图

    Figure  7.  Schematic diagram of feature fusion process

    图  8  预处理后的金刚石图像

    Figure  8.  Processed diamond images

    图  9  金刚石净度检测装置

    Figure  9.  Diamond clarity testing device

    图  10  迁移学习对模型训练结果的影响

    Figure  10.  Influence of transfer learning on model training results

    图  11  学习率对模型训练结果的影响

    Figure  11.  Influence of Learning rate on model training results

    表  1  不同模型训练结果

    Table  1.   Training results of different models

    模型验证准
    确率
    A1 / %
    训练准
    确率
    A2 / %
    平均训
    练时间
    t1 / s
    平均检
    测时间
    t2 / s
    VGG1687.895.833.4560.02435
    AlexNet87.692.520.9310.00826
    ResNet5098.698.823.8390.01588
    CBAM-ResNet5098.799.224.1430.01629
    下载: 导出CSV

    表  2  金刚石颗粒测试集混淆矩阵

    Table  2.   Confusion matrix of diamond particle test set

    预测类别
    ABCDE
    真实标签A4000000
    B0766130
    C0262260
    D02011480
    E000050
    下载: 导出CSV

    表  3  CBAM-ResNet50模型金刚石颗粒分类结果

    Table  3.   CBAM-ResNet50 model diamond particle classification results

    金刚石
    等级
    精确度
    P / %
    召回率
    R / %
    特异性
    S / %
    F1 / %
    A100.00100.00100.00100.00
    B99.5099.5099.8099.50
    C99.8098.7099.7099.20
    D99.2099.8099.9099.50
    E100.00100.00100.00100.00
    下载: 导出CSV

    表  4  不同模型的金刚石颗粒分类结果

    Table  4.   Classification results of diamond particles of different models

    金刚石等级模型精确度
    P / %
    召回率
    R / %
    特异性
    S / %
    F1 / %
    AResNet5096.15100.00100.0098.04
    ResNet50 + FPN99.01100.00100.0099.50
    ResNet50 + CBAM100.00100.00100.00100.00
    CBAM-ResNet50100.00100.00100.00100.00
    BResNet5097.9294.0098.5195.92
    ResNet50 + FPN98.0098.0099.5098.00
    ResNet50 + CBAM98.9998.0099.5098.49
    CBAM-ResNet5099.5099.5099.8099.50
    CResNet5098.9897.0099.2597.98
    ResNet50 + FPN99.4998.5099.6298.99
    ResNet50 + CBAM99.4998.5099.6298.99
    CBAM-ResNet5099.8098.7099.7099.20
    DResNet5096.0497.0099.2596.52
    ResNet50 + FPN97.2497.5099.2597.38
    ResNet50 + CBAM98.0198.5099.6298.25
    CBAM-ResNet5099.2099.8099.9099.50
    EResNet5099.01100.00100.0099.50
    ResNet50 + FPN99.01100.00100.0099.50
    ResNet50 + CBAM100.00100.00100.00100.00
    CBAM-ResNet50100.00100.00100.00100.00
    下载: 导出CSV
  • [1] 周青超, 沈锡田. 从专利角度分析人造金刚石技术的发展 [J]. 超硬材料工程,2021,33(5):29-36. doi: 10.3969/j.issn.1673-1433.2021.05.007

    ZHOU Qingchao, SHEN Xitian. Analysis of the development of synthetic diamond technology from the perspective of patents [J]. Superhard Material Engineering,2021,33(5):29-36. doi: 10.3969/j.issn.1673-1433.2021.05.007
    [2] 熊辉. 基于模板匹配的金刚石颗粒图像识别 [D]. 成都: 四川大学, 2005.

    XIONG Hui. The image recognition of the diamond particles based on the template matching [D]. Chengdou: Sichuan University, 2005.
    [3] 郭树青. 金刚石颗粒形貌检测系统关键技术研究 [D]. 郑州: 郑州大学, 2016.

    GUO Shuqing. Research on the key technology of diamond particle morphology detection system [D]. Zhengzhou: Zhengzhou University, 2005.
    [4] WANG W, CAI L. Inclusion extraction from diamond clarity images based on the analysis of diamond optical properties [J]. Optics Express,2019,27(19):27242. doi: 10.1364/OE.27.027242
    [5] WANG W, CAI L. On the development of an effective image acquisition system for diamond quality grading [J]. Applied Optics,2018,57(33):9887-9897. doi: 10.1364/AO.57.009887
    [6] 石广丰, 王雪, 王淑坤, 等. 基于机器视觉的金刚石原石检测系统 [J]. 金刚石与磨料磨具工程,2019,39(6):7-12. doi: 10.13394/j.cnki.jgszz.2019.6.0002

    SHI Guangfeng, WANG Xue, WANG Shukun, et al. Diamond raw detection system based on machine vision [J]. Diamond & Abrasives Engineering,2019,39(6):7-12. doi: 10.13394/j.cnki.jgszz.2019.6.0002
    [7] 狄超雄. 钻石净度检测系统研究 [D]. 武汉: 华中科技大学, 2021.

    DI Chaoxiong. Research on diamond clarity detection system [D]. Wuhan: Huazhong University of Science & Technology, 2021.
    [8] 季长清, 高志勇, 秦静, 等. 基于卷积神经网络的图像分类算法综述 [J]. 计算机应用,2022,42(4):1044-1049. doi: 10.11772/j.issn.1001-9081.2021071273

    JI Changqing, GAO Zhiyong, QIN Jing, et al. Review of image classification algorithms based on convolutional neural network [J]. Journal of Computer Applications,2022,42(4):1044-1049. doi: 10.11772/j.issn.1001-9081.2021071273
    [9] 邢延动, 李远. BP神经网络在金刚石锯片磨粒识别中的应用 [J]. 超硬材料过程,2014(1):1-4. doi: 10.3969/j.issn.1673-1433.2014.01.001

    XING Yandong, LI Yuan. Application of BP neural network in diamond saw blade abrasive particle recognition [J]. Superhard Material Engineering,2014(1):1-4. doi: 10.3969/j.issn.1673-1433.2014.01.001
    [10] 潘秉锁, 潘文超, 刘子玉. 基于空洞卷积神经网络的金刚石图像语义分割 [J]. 金刚石与磨料磨具工程,2019,39(6):20-24. doi: 10.13394/j.cnki.jgszz.2019.6.0004

    PAN Bingsuo, PAN Wenchao, LIU Ziyu. Semantic segmentation of diamond image using dilated convolutional neural network [J]. Diamond & Abrasives Engineering,2019,39(6):20-24. doi: 10.13394/j.cnki.jgszz.2019.6.0004
    [11] 林振坤. 基于深度学习的金刚石品质检测技术与实现 [D]. 郑州: 郑州航空工业管理学院, 2020.

    LIN Zhenkun. Diamond quality detection technology and implementation based on deep learning [D]. Zhengzhou: Zhengzhou University of Aeronautics, 2020.
    [12] 杨建新, 兰小平, 赵振, 等. 基于改进郊狼算法与极限学习机的工业金刚石检测 [J]. 计算机集成制造系统,2023,29(2):449-459. doi: 10.13196/j.cims.2023.02.008

    YANG Jianxin, LAN Xiaoping, ZHAO Zhen, et al. Industrial diamond detection method based on improved coyote optimization algorithm and extreme learning machine [J]. Computer Integrated Manufacturing Systems,2023,29(2):449-459. doi: 10.13196/j.cims.2023.02.008
    [13] SANDLER M, HOWARD A, ZHU M, et al. MobileNetV2: Iinverted residuals and linear bottlenecks: 2018 IEEE conference on computer vision and pattern recognition (CVPR) [C]. Salt Lake: IEEE, 2018.
    [14] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition: 2016 IEEE conference on computer vision and pattern recognition (CVPR) [C]. Las Vegas: IEEE, 2016.
    [15] WOO S, PARK J, LEE J, et al. CBAM: Convolutional block attention module: European conference on computer vision (ECCV) [C]. Munich: CVPR, 2018.
    [16] LIN T, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection: 2017 IEEE conference on computer vision and pattern recognition (CVPR) [C]. Honolulu: IEEE, 2017.
    [17] 沈微微, 李颖, 杨志豪, 等. 防止过拟合的属性约简 [J]. 计算机应用研究,2020,37(9):2665-2668. doi: 10.19734/j.issn.1001-3695.2019.05.0116

    SHEN Weiwei, LI Ying, YANG Zhihao, et al. Attribute reduction with avoiding overfitting [J]. Application Research of Computers,2020,37(9):2665-2668. doi: 10.19734/j.issn.1001-3695.2019.05.0116
    [18] 邓建国, 张素兰, 张继福, 等. 监督学习中的损失函数及应用研究 [J]. 大数据,2020,6(1):60-80. doi: 10.11959/j.issn.2096-0271.2020006

    DENG Jianguo, ZHANG Sulan, ZHANG Jifu, et al. Loss function and application research in supervised learning [J]. Big Data Research,2020,6(1):60-80. doi: 10.11959/j.issn.2096-0271.2020006
    [19] PAN S, YANG Q. A survey on transfer learning [J]. IEEE Transactions on Knowledge & Data Engineering,2010,22(10):1345-1359. doi: 10.1109/TKDE.2009.191
    [20] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition: 3rd international conference on learning representations (ICLR 2015) [C]. San Diego: arXiv, 2015.
    [21] KRIZHEVSKY A, SUTSKEVER I, HINTON G. ImageNet classification with deep convolutional neural networks: 25th international conference on neural Information processing systems (NeurIPS 2012) [C]. New York: Curran Associates Inc., 2012.
  • 加载中
图(11) / 表(4)
计量
  • 文章访问数:  105
  • HTML全文浏览量:  54
  • PDF下载量:  12
  • 被引次数: 0
出版历程
  • 收稿日期:  2023-07-30
  • 修回日期:  2023-12-04
  • 录用日期:  2023-12-12
  • 网络出版日期:  2024-11-20
  • 刊出日期:  2024-10-01

目录

    /

    返回文章
    返回