基于改进YOLOv4算法的矿用设备关键部件识别研究

doi:10.11799/ce202409025

摘要/Abstract

摘要：

为了使煤矿设备检测平台能够精准快速识别出复杂矿用设备的关键部件，提出一种基于YOLOv4的轻量化改进模型GhostNet-YOLOv4，模型引入GhostNet作为YOLOv4的CSPDarkNet53，减少模型参数量与冗余计算量的同时，降低模型训练时间；GhostNet当中的shortcut 结构减少网络退化现象，提高模型特征提取能力。实验结果证明，改进模型参数量与YOLOv4相比减少83%，与MobileNet3-YOLOv4相比减少6%，识别平均准确率达到92.67%，图片检测速度（FPS）可达到33.75帧/s，相较于YOLOv4提升了43%。在复杂矿用设备的关键部件检测任务中有效地解决了检测速度慢、模型体积大的问题。

关键词: 复杂矿用设备, 关键部件, YOLOv4, GhostNet

Abstract:

To enable the coal mine equipment detection platform to accurately and rapidly identify critical components of complex mining equipment, a lightweight improved model named Ghost-YOLOv4 based on YOLOv4 is proposed. This model introduces GhostNet as the backbone of YOLOv4's CSPdarknet-53, which not only reduces the number of model parameters and redundant computations but also shortens the model training time. The shortcut structure in GhostNet mitigates network degradation and enhances the model's feature extraction capability. Experimental results demonstrate that the number of parameters in GhostNet is reduced by 83% compared to YOLOv4 and by 6% compared to MobileNetv3-YOLOv4, with an average recognition accuracy of 92.51%. The image detection speed (FPS) can reach 33.75 frames per second, representing a 43% improvement over YOLOv4. Ghost-YOLOv4 effectively addresses the issues of slow detection speed and large model size in the task of detecting critical components of complex mining equipment.

中图分类号:

TD679

吴钰晶. 基于改进YOLOv4算法的矿用设备关键部件识别研究[J]. 煤炭工程, 2024, 56(9): 165-171.

参考文献

[1]Alexey Bochkovskiy,Chien-Yao Wang,Hong-Yuan Mark Liao.YOLOv4:Optimal Speed and Accuracy of Object Detection[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington: IEEE Press, 2020: 102-123
[2]Kai Han,Yunhe Wang, Qi Tian et al.GhostNet: More Features from Cheap Operations[C].In CVPR, 2020.
[3]Andrew Howard, Mark Sandler, Grace Chu et al. Searching for mobilenetv3[C]. In ICCV, 2019.
[4]Redmon J, Divvala S, Girshick R , et al. You only look Once: unified, real-time object detection[EB/OL].[2016-05-09], https:// arxiv .org /pdf/1506.02640.pdf
[5]周志华.机器学习[M].北京：清华大学出版社，2016:97-115，2022-11 重印
[6]Kaiming He,Xiangyu Zhang,Shaoqing Ren, Jian Sun.Deep Residual Learning for Image Recognition[C].Dec, 2015.URL:https://doi.org/10.48550/arXiv.1512.03385
[7]Zheng Z , Wang P , Liu W ,et al.Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression[J].arXiv, 2019.DOI:10.1609/aaai.v34i07.6999.
[8]Misra D .Mish: A Self Regularized Non-Monotonic Neural Activation Function[J]. 2019.DOI:10.48550/arXiv.1908.08681.
[9]Howard A , Sandler M , Chu G ,et al.Searching for MobileNetV3[J]. 2019.DOI:10.48550/arXiv.1905.02244.
[10]Wang C Y, Liao H Y M, Wu Y H, et al. CSPNet: A new backbone that can enhance learning capability of CNN[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Seattle: IEEE, 2020: 1571-1580.
[11]Li H, Li C, Li G, et al. A real-time table grape detection method based on improved YOLOv4-tiny network in complex background[J]. Biosystems Engineering, 2021, 212: 347-359
[12]Karen Simonyan and Andrew Zisserman.Very Deep Convolutional Networks for Large-Scale Image Recognition[C].Sep,2014.
[13]王国法,赵国瑞,任怀伟.智慧煤矿与智能化开采关键核心技术分析[J].煤炭学报, 2019(1):8.DOI:CNKI:SUN:MTXB.0.2019-01-004.
[14]雷世威,肖兴美,张明.基于改进YOLOv3的煤矸识别方法研究[J].矿业安全与环保, 2021, 48(3):6.DOI:10.19835/j.issn.1008-4495.2021.03.010.
[15]蔡秀凡,谢金辰.YOLOv4煤矸石检测方法研究[J].煤炭工程, 2022, 54(8):6.
[16]卢万杰,付华,赵洪瑞.基于深度学习算法的矿用巡检机器人设备识别[J].工程设计学报, 2019, 26(5):7.DOI:CNKI:SUN:GCSJ.0.2019-05-007.
[17]张玉涛,张梦凡,史学强,等.基于深度学习的井下运动目标跟踪算法研究[J].煤炭工程, 2022, 54(10):5.
[18]蔡舒平，孙仲鸣，刘慧，等. 基于改进型YOLOv4的果园障碍物实时检测方法[J]. 农业工程学报，2021，37(2)：36-43.
[19]周桂红，马帅，梁芳芳. 基于改进YOLOv4模型的全景图像苹果识别[J]. 农业工程学报，2022，38(21)：159-168.
[20]周兆宇.煤矿无轨胶轮车智能检测系统浅析[J].机电信息,2021(21):59-61.DOI:10.19514/j.cnki.cn32-1628/tm.2021.21.024.
[21]Andrew Howard, Mark Sandler, Grace Chu et al. Searching for mobilenetv3[C]. In ICCV, 2019.
[22]Redmon J, Divvala S, Girshick R , et al. You only look Once: unified, real-time
[23]object detection[EB/OL].[2016-05-09], https:// arxiv .org /pdf/1506.02640.pdf
[24]Kai Han, Yunhe Wang, Qi Tian et al.GhostNet: More Features from Cheap Operations[C].In CVPR, 2020.
[25]Alexey Bochkovskiy,Chien-Yao Wang,Hong-Yuan Mark Liao.YOLOv4:Optimal Speed and Accuracy of Object Detection[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington: IEEE Press, 2020: 102-123
[26]周志华.机器学习[M].北京：清华大学出版社，2016:97-115，2022-11 重印
[27]Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.Deep Residual Learning for Image Recognition[C].Dec, 2015.URL:https://doi.org/10.48550/arXiv.1512.03385
[28]Zheng Z , Wang P , Liu W ,et al.Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression[J].arXiv, 2019.DOI:10.1609/aaai.v34i07.6999.
[29]Misra D .Mish: A Self Regularized Non-Monotonic Neural Activation Function[J].
[30]2019.DOI:10.48550/arXiv.1908.08681.
[31]雷世威,肖兴美,张明.基于改进YOLOv3的煤矸识别方法研究[J].矿业安全与环保, 2021, 48(3):6.DOI:10.19835/j.issn.1008-4495.2021.03.010.