基于改进YOLOv5s的轻量化智能钻机管柱检测方法

牛柯; 彭斌; 杨小亮

doi:10.13700/j.bh.1001-5965.2024.0088

基于改进YOLOv5s的轻量化智能钻机管柱检测方法

doi: 10.13700/j.bh.1001-5965.2024.0088

牛柯¹,
彭斌^1, ,,
杨小亮^{1, 2}

1.
兰州理工大学机电工程学院，兰州 730050
2.
兰州兰石石油装备工程股份有限公司，兰州 730050

基金项目:

国家自然科学基金(51675254,51966009)；国家重点研发计划(SQ2020YFF0420989)；中央引导地方科技发展资金项目(23ZYQA312)

详细信息

通讯作者:
E-mail：pengb2000@lut.edu.cn

中图分类号: TP391.4
计量
- 文章访问数: 413
- HTML全文浏览量: 153
- PDF下载量: 25
- 被引次数: 0
出版历程
- 收稿日期: 2024-02-22
- 录用日期: 2024-04-19
- 网络出版日期: 2024-05-07
- 整期出版日期: 2026-04-30

Lightweight intelligent rig pipe column inspection method based on improved YOLOv5s

NIU Ke¹,
PENG Bin^{1
, ,},
YANG Xiaoliang^{1, 2}

1.
Faculty of Mechanical and Electrical Engineering，Lanzhou University of Technology，Lanzhou 730050，China
2.
Lanzhou Lanshi Petroleum Equipment Engineering Co., Lanzhou 730050，China

Funds:

National Natural Science Foundation of China (51675254,51966009); National Key Research and Development Program of China (SQ2020YFF0420989); Central Government Guides Local Funds for Science and Technology Development (23ZYQA312)

More Information

Corresponding author: E-mail：pengb2000@lut.edu.cn

摘要

摘要:
针对钻井平台管柱自动化转运及上卸扣作业水平较低的问题，提出一种改进的多目标检测算法以实现钻杆与管柱接头的精确识别。所提方法以轻量化 EfficientNetV2 为特征提取网络，引入 SPPF 模块降低参数量，并在 Backbone 中融合 CBAM 注意力机制以抑制背景干扰；采用 BiFPN 替代 PANet 提升多尺度特征融合能力，引入 CARAFE 算子增强上采样特征表达；在 Head 部分使用 GhostConv 减少计算复杂度，以 SIoU 损失函数优化边界回归精度，通过AdamW 优化器提升模型收敛性与泛化能力。基于自建数据集的实验结果表明：改进模型在复杂工况下对不同姿态管柱具有良好识别性能，检测准确率达到 90.6%，平均精度均值达到 94.6%，较原模型分别提升 3.9% 和 4.5%，验证了所提方法的有效性与鲁棒性。
- 计算机应用技术 /
- YOLOv5s算法 /
- 智能钻机 /
- 钻杆 /
- 管柱接头 /
- 目标检测 /
- 算法改进
Abstract:
To address the low level of automation in drill string handling and make-up/break-out operations on drilling platforms, an improved multi-object detection algorithm is proposed for accurate identification of drill pipes and tool joints. The model employs a lightweight EfficientNetV2 as the backbone, with an SPPF module to reduce parameters. A CBAM attention module is integrated to suppress background interference. BiFPN replaces PANet to enhance multi-scale feature fusion, while CARAFE is adopted to improve upsampling quality and feature representation. In the detection head, GhostConv is used to reduce computational cost, and the SIoU loss is introduced to improve bounding box regression. The AdamW optimizer is applied to accelerate convergence and enhance generalization. Experiments on a self-built dataset demonstrate that the proposed method achieves robust performance under complex conditions, accurately detecting drill strings with varying poses. The detection accuracy reaches 90.6% and mAP reaches 94.6%, representing improvements of 3.9% and 4.5% over the baseline, respectively, confirming its effectiveness and robustness.
- computer application technology /
- YOLOv5s algorithm /
- automated drilling rig /
- drill pipe /
- pipe column joint /
- target detection /
- algorithm improvement

HTML全文

图 1 管柱检测示意图

Figure 1. Detection schematic of pipe columns

下载: 全尺寸图片幻灯片

图 2 基于深度学习的管柱实时检测框架

Figure 2. Deep learning-based framework for real-time detection of pipe columns

下载: 全尺寸图片幻灯片

图 3 传统YOLOv5s网络模型

Figure 3. Traditional YOLOv5s model structure

下载: 全尺寸图片幻灯片

图 4 注意力机制模块

Figure 4. Convolutional block attention module

下载: 全尺寸图片幻灯片

图 5 MBConv和Fused-MBConv结构图

Figure 5. MBConv and Fused-MBConv structure diagrams

下载: 全尺寸图片幻灯片

图 6 CARAFE结构图

Figure 6. CARAFE module structure

下载: 全尺寸图片幻灯片

图 7 多尺度特征融合网络架构

Figure 7. Multi-scale feature fusion network architecture

下载: 全尺寸图片幻灯片

图 8 普通卷积和Ghost卷积结构图

Figure 8. Structural diagrams of ordinary convolution and Ghost convolution

下载: 全尺寸图片幻灯片

图 9 角度成本

Figure 9. Angle cost

下载: 全尺寸图片幻灯片

图 10 改进后的YOLOv5s网络结构

Figure 10. Improved YOLOv5s network structure

下载: 全尺寸图片幻灯片

图 11 部分管柱检测数据集

Figure 11. Partial sample image

下载: 全尺寸图片幻灯片

图 12 数据集类别数量分布图和图像标注锚框的分布图

Figure 12. Plot of the distribution of the number of categories in the dataset and the distribution of the image labeling anchor frames

下载: 全尺寸图片幻灯片

图 13 评价指标

Figure 13. Evaluation indices

下载: 全尺寸图片幻灯片

图 14 热力图分析

Figure 14. Heat map analysis

下载: 全尺寸图片幻灯片

图 15 不同损失函数曲线

Figure 15. Different loss function curves

下载: 全尺寸图片幻灯片

图 16 优化器性能对比图

Figure 16. Performance comparison diagram of the optimizer

下载: 全尺寸图片幻灯片

图 17 初始YOLOv5s和改进YOLOv5s混淆矩阵对比

Figure 17. Comparison of initial yolov5s and improved algorithm confusion matrices

下载: 全尺寸图片幻灯片

图 18 算法改进前后效果对比

Figure 18. Comparison of effects before and after algorithm improvement

下载: 全尺寸图片幻灯片

表 1 EfficientNetv2-B0 结构

Table 1. EfficientNetv2-B0 structure

网络阶段	结构	卷积核步距	通道数	层数
1	Stem3×3	2	32	1
2	Fused-MBConv1, k 3×3	1	16	1
3	Fused-MBConv8, k 3×3	2	32	2
4	Fused-MBConv4, k 3×3	2	48	2
5	MBConv8, k 3×3, SE0.25 MBConv6, k 3×3, SE0.25	2	96	3
6	MBConv8, k 3×3, SE0.25 MBConv6, k 3×3, SE0.25	1	112	5
7	MBConv8, k 3×3, SE0.25 MBConv6, k 3×3, SE0.25	2	192	8
8	Conv1×1&Pooling&FC		1280	1

下载: 导出CSV

表 2 模型训练参数

Table 2. Model training parameters

训练参数	参数值
迭代次数/轮	200
批量大小^[21]/B	16
初始学习速率	0.01
动量	0.937
图片尺寸/(像素×像素)	640×640

下载: 导出CSV

表 3 YOLOv5不同版本性能比较

Table 3. Comparison of the performance of different versions of YOLOv5

模型	浮点运算速度/ 10⁹ s⁻¹	参数量	G_mAP/%	检测速度/ (帧·s⁻¹)	权重大小/ MB
YOLOv5s	16.3	7.06×10⁶	91.6	73.167	14.4
YOLOv5m	47.9	21.6×10⁶	91.2	72.4	42.1
YOLOv5l	107.7	46.1×10⁶	91.7	86.7	92.8
YOLOv5x	204.7	86.2×10⁶	90.4	91.6	86.4

下载: 导出CSV

表 4 注意力机制性能对比

Table 4. Attention mechanism performance comparison

模型	参数量	G_mAP/%	检测速度/ (帧·s⁻¹)	权重大小/ MB
YOLOv5s	7.06×10⁶	91.6	73.167	14.4
YOLOv5s+CBAM	6.41×10⁶	92.0	101.919	13.2
YOLOv5s+SE	7.23×10⁶	89.8	97.313	14.8
YOLOv5s+CA	6.42×10⁶	91.6	101.542	13.2
YOLOv5s+ECA	7.20×10⁶	90.4	103.465	14.7

下载: 导出CSV

表 5 不同损失函数检测性能

Table 5. Detection performance of different loss functions

损失函数	G_mAP/%	精确率/%	召回率/%	训练时长/h
CIoU	86.2	86.6	82.6	8.724
GIoU	85.8	85.3	88.4	26.561
EIoU	83.9	88.0	77.3	8.355
SIoU	89.4	88.5	89.6	8.207

下载: 导出CSV

表 6 消融实验评价指标

Table 6. Results of ablation experiments

实验	准确率/%	召回率/%	G_mAP/%	参数量	浮点运算速度/10⁹ s⁻¹	推理时间/(帧·s⁻¹)
1	86.7	91.7	90.1	7.06×10⁶	15.8	10.4
2	88.1	90.4	89.8	5.42×10⁶	6.9	6.2
3	86.6	92.6	92	6.41×10⁶	16	11.7
4	86.4	91.2	89.6	7.17×10⁶	16.4	10.8
5	87.8	91.3	91.2	6.82×10⁶	15.9	11.2
6	88.5	89.6	90.4	7.06×10⁶	15.8	10.2
7	88.8	89.1	90.7	5.57×10⁶	12.1	8.2
8	87.4	88.6	89.8	5.61×10⁶	13.6	8.5
9	89.5	88.4	93.7	5.75×10⁶	13.8	8.9
10	90.3	91.4	94.2	5.75×10⁶	14.2	9.5
11	90.6	91.2	94.6	5.24×10⁶	13.4	8.4

下载: 导出CSV

表 7 YOLO系列主流模型对比实验

Table 7. Comparative experiments of mainstream models in the YOLO series

模型	AP/%			G_mAP@0.5%	参数量	权重大小/MB	浮点运算速度/10⁹ s⁻¹	检测速度/(帧·s⁻¹)
模型	钻杆	接头	工人	G_mAP@0.5%	参数量	权重大小/MB	浮点运算速度/10⁹ s⁻¹	检测速度/(帧·s⁻¹)
YOLOv3^[28]	81.3	87.7	84.9	88	8.67×10⁶	17.4	12.9	39.6
YOLOv5s	84.8	90.2	85.3	90.1	7.06×10⁶	14.4	16.3	42.6
YOLOv6^[29]	82.4	86.5	81.6	90	18.5×10⁶	38.7	45.2	104
YOLOv7-tiny^[30]	85.5	89.9	85.5	91.4	6.01×10⁶	11.6	13.0	84
YOLOv8^[31]	84.8	91.3	85.7	92	11.1×10⁶	22.5	28.4	108
YOLOX^[32]	88.6	85.4	86.2	90.4	8.94×10⁶	12.4	26.8	86
本文	93.9	92.9	87.6	94.6	5.24×10⁶	10.9	13.4	86.4

下载: 导出CSV

参考文献(32)

[1]	白尚懿, 杨小亮, 何生兵, 等. 自动化钻机管柱处理系统研究现状与发展趋势[J]. 机械研究与应用, 2020, 33(5): 203-207. BAI S Y, YANG X L, HE S B, et al. Research status and development trend of pipe handling system for automated drilling rig[J]. Mechanical Research & Application, 2020, 33(5): 203-207(in Chinese).
[2]	杨传书, 张好林, 肖莉. 自动化钻井关键技术进展与发展趋势[J]. 石油机械, 2017, 45(5): 10-17. YANG C S, ZHANG H L, XIAO L. Key technical progress and development trend of automated drilling[J]. China Petroleum Machinery, 2017, 45(5): 10-17(in Chinese).
[3]	康亮, 李亚伟, 黎善猛, 等. 石油钻机管柱处理系统研究现状和发展趋势分析[J]. 液压气动与密封, 2019, 39(5): 1-5. KANG L, LI Y W, LI S M, et al. Research present situation and development trend analysis of pipeline handling system for drill rig[J]. Hydraulics Pneumatics & Seals, 2019, 39(5): 1-5(in Chinese).
[4]	JOCHER G, STOKEN A, BOROVEC J, et al. Ultralytics/yolov5: v3.0 [EB/OL]. (2020-08-13)[2023-12-20]. https://doi.org/10.5281/zenodo.3983579.
[5]	LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 8759-8768.
[6]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the Computer Vision-ECCV 2018. Cham: Springer, 2018: 3-19.
[7]	WANG D C, CHEN X N, JIANG M Y, et al. ADS-Net: an attention-based deeply supervised network for remote sensing image change detection[J]. International Journal of Applied Earth Observation and Geoinformation, 2021, 101: 102348.
[8]	TAN M X, LE Q V. EfficientNetV2: smaller models and faster training[C]//Proceedings of the International Conference on Machine Learning. New York: PMLR, 2021: 10096-10106.
[9]	WANG J Q, CHEN K, XU R, et al. CARAFE: content-aware ReAssembly of FEatures[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE Press, 2019: 3007-3016.
[10]	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. Piscataway: IEEE Press, 2017: 936-944.
[11]	TAN M X, PANG R M, LE Q V. EfficientDet: scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 10778-10787.
[12]	ZHU L L, GENG X, LI Z, et al. Improving YOLOv5 with attention mechanism for detecting boulders from planetary images[J]. Remote Sensing, 2021, 13(18): 3776.
[13]	HAN K, WANG Y H, TIAN Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 1577-1586.
[14]	ZHOU D F, FANG J, SONG X B, et al. IoU loss for 2D/3D object detection[C]//Proceedings of the I International Conference on 3D Vision. Piscataway: IEEE Press, 2019: 85-94.
[15]	ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 12993-13000.
[16]	GEVORGYAN Z. SIoU loss: more powerful learning for bounding box regression [EB/OL]. (2022-05-25)[2023-12-20]. https://arxiv.org/abs/2107.08430.
[17]	WANG J Y, JOSHI G. Cooperative SGD: a unified framework for the design and analysis of local-update SGD algorithms[J]. Journal of Machine Learning Research, 2021, 22(1): 9709-9758.
[18]	LOSHCHILOV I, HUTTER F. Decoupled weight decay regularization[EB/OL]. (2017-11-14) [2023-12-20]. https://doi.org/10.48550/arxiv.1711.05101.
[19]	ZHOU P, FENG J, MA C, et al. Towards theoretically understanding why sgd generalizes better than adam in deep learning[J]. Advances in Neural Information Processing Systems, 2020, 33: 21285-21296.
[20]	杨波, 何金平, 张立娜. 基于改进SPP-x的YOLOv5神经网络水稻叶片病害识别检测[J]. 中国农机化学报, 2023, 44(9): 190-197. YANG B, HE J P, ZHANG L N. Identification and detection of rice leaf diseases by YOLOv5 neural network based on improved SPP-x[J]. Journal of Chinese Agricultural Mechanization, 2023, 44(9): 190-197(in Chinese).
[21]	XUE B, HE Y, JING F, et al. Robot target recognition using deep federated learning[J]. International Journal of Intelligent Systems, 2021, 36(12): 7754-7769.
[22]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks [C]// Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 7132-7141.
[23]	QIN Z Q, ZHANG P Y, WU F, et al. FcaNet: frequency channel attention networks[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE Press, 2021: 763-772.
[24]	WANG Q L, WU B G, ZHU P F, et al. ECA-net: efficient channel attention for deep convolutional neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 11531-11539.
[25]	SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 618-626.
[26]	WANG Y B, WANG H F, PENG Z H. Rice diseases detection and classification using attention based neural network and Bayesian optimization[J]. Expert Systems with Applications, 2021, 178: 114770.
[27]	PENG S D, JIANG W, PI H J, et al. Deep snake for real-time instance segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 8530-8539.
[28]	REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. (2021-10-18) [2023-12-20]. https://arxiv.org/abs/1804.02767.
[29]	DANG F Y, CHEN D, LU Y Z, et al. YOLOWeeds: a novel benchmark of YOLO object detectors for multi-class weed detection in cotton production systems[J]. Computers and Electronics in Agriculture, 2023, 205: 107655.
[30]	WANG C Y, BOCHKOVSKIY A, LIAO H M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2023: 7464-7475.
[31]	REIS D, KUPEC J, HONG J, et al. Real-time flying object detection with YOLOv8[EB/OL]. (2023-05-17)[2023-12-20]. https://arxiv.org/abs/2305.09972.
[32]	GE Z, LIU S T, WANG F, et al. YOLOX: exceeding YOLO series in 2021[EB/OL]. (2021-08-06)[2023-12-20]. https://arxiv.org/abs/2107.08430.