基于改进YOLOv5s的外脚手架隐患图像识别技术

doi:10.16265/j.cnki.issn1003-3033.2023.12.2011

摘要/Abstract

摘要：

为提高外脚手架安全管理的质量和效率,基于图像识别技术提出一种改进YOLOv5s的外脚手架隐患识别方法。首先,为解决背景信息过多造成的识别精度下降问题,在主干网络嵌入设计卷积注意模块(CBAM),获取隐患的各种细节特征;其次,改进原算法颈部特征融合模块为加权双向特征金字塔网络(BiFPN)结构,有效处理外脚手架隐患目标尺寸分布不均衡造成的多尺度特征不平衡问题;然后,使用边界框损失函数斯库拉交并比(SIoU) Loss替换原损失函数;最后,通过消融试验分析改进模块对模型性能的影响,并与其他算法进行对比分析,验证隐患识别效果。结果表明:改进后的网络实现均值平均精度(mAP@0.5:0.95)评分提升5.13%,召回率提升3.45%,对多尺度、多目标及复杂背景下的外脚手架隐患具有良好的识别效果。

关键词: YOLOv5s, 外脚手架隐患, 图像识别, 多尺度特征, 均值平均精度(mAP), 加权双向特征金字塔网络(BiFPN)

Abstract:

In order to improve the quality and efficiency of external scaffold safety management, an improved YOLOv5 s external scaffold hidden danger identification method is proposed based on image recognition technology.Firstly, in order to solve the problem that the recognition precision decreases due to more background information, a convolutional block attention module(CBAM) was embedded in the algorithm backbone network to obtain various detailed features of hidden danger. Secondly, the neck feature fusion module was improved into BiFPN, which can effectively deal with the problem of multi-scale feature imbalance caused by the uneven size distribution of scaffolding hidden targets. Thirdly, the original frame loss function was replaced by Scylla intersection over union(SIoU) Loss. Finally, the influence of the improved module on the performance of the model was analyzed by ablation test and compared with other algorithms to verify the recognition effect. The results show that the score of mAP@0.5:0.95 is increased by 5.13%, and the recall rate is increased by 3.45 %. The algorithm has a good recognition effect in multi-scale, the multi-target and complex background of construction sites. It provides technical support for the further application of image recognition technology in the field of external scaffold safety management in a construction site.

Key words: YOLOv5s, external scaffold hidden danger, image recognition, multi-scale features, mean average precision(mAP), bi-directional feature pyramid network(BiFPN)

赵江平, 刘星星, 张想卓. 基于改进YOLOv5s的外脚手架隐患图像识别技术[J]. 中国安全科学学报, 2023, 33(12): 60-66.

ZHAO Jiangping, LIU Xingxing, ZHANG Xiangzhuo. Research on image recognition technology for external scaffold hidden danger based on improved YOLOv5s[J]. China Safety Science Journal, 2023, 33(12): 60-66.

图/表 10

图1

图2

表1

表2

图3

图4

图5

表3

表4

图6

参考文献 14

[1]	王凯全, 徐显维, 缪志国, 等. 移动脚手架升降作业的FMECA分析[J]. 中国安全科学学报, 2008, 18(3):138-142.
	WANG Kaiquan, XU Xianwei, MIAO Zhiguo, et al. FMECA analysis of the work safety on attached self-lifting scaffolds[J]. China Safety Science Journal, 2008, 18(3): 138-142.
[2]	REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6):1137-1149. doi: 10.1109/TPAMI.2016.2577031 pmid: 27295650
[3]	REDMON J, DIVVALA S K, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2016: 779-788.
[4]	LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]. European Conference on Computer Vision, 2016: 21-37.
[5]	赵江平, 徐恒, 党悦悦. 基于改进Faster R-CNN的铁路客车螺栓检测研究[J]. 中国安全科学学报, 2021, 31(7):82-89. doi: 10.16265/j.cnki.issn 1003-3033.2021.07.012
	ZHAO Jiangping, XU Heng, DANG Yueyue. Research on bolt detection of railway passenger cars based on improved Faster R-CNN[J]. China Safety Science Journal, 2021, 31(7):82-89. doi: 10.16265/j.cnki.issn 1003-3033.2021.07.012
[6]	杨剑锋, 秦钟, 庞小龙, 等. 基于深度学习网络的输电线路异物入侵监测和识别方法[J]. 电力系统保护与控制, 2021, 49(4):37-44.
	YANG Jianfeng, QIN Zhong, PANG Xiaolong, et al. Foreign body intrusion monitoring and recognition method based on Dense-YOLOv3 deep learning network[J]. Power System Protection and Control, 2021, 49 (4): 37-44.
[7]	李瑞生, 张彦龙, 翟登辉, 等. 基于改进SSD的输电线路销钉缺陷检测[J]. 高电压技术, 2021, 47(11):3795-3802.
	LI Ruisheng, ZHANG Yanlong, ZHAI Denghui, et al. Pin defect detection of transmission line based on improved SSD[J]. High Voltage Engineering, 2021, 47 (11): 3795-3802.
[8]	李华, 王岩彬, 益朋, 等. 基于深度学习的复杂作业场景下安全帽识别研究[J]. 中国安全生产科学技术, 2021, 17(1):175-181.
	LI Hua, WANG Yanbin, YI Peng, et al. Research on recognition of safety helmets under complex operation scenes based on deep learning[J]. Journal of Safety Science and Technology, 2021, 17(1):175-181.
[9]	ZHANG Hongyang, JIN Yinlong, LIU Quan, et al. Intelligent monitoring method for tamping times during dynamic compaction construction using machine vision and pattern recognition[J]. Measurement, 2022,193:DOI:10.1016/j.meansurement.2022.110835.
[10]	张明媛, 曹志颖, 赵雪峰, 等. 基于深度学习的建筑工人安全帽佩戴识别研究[J]. 安全与环境学报, 2019, 19(2):535-541.
	ZHANG Mingyuan, CAO Zhiying, ZHAO Xuefeng, et al. On the identification of the safety helmet wearing manners for the construction company workers based on the deep learning theory[J]. Journal of Safety and Environment, 2019, 19(2):535-541.
[11]	杨永波, 李栋. 改进YOLOv5的轻量级安全帽佩戴检测算法[J]. 计算机工程与应用, 2022, 58 (9): 201-207. doi: 10.3778/j.issn.1002-8331.2111-0346
	YANG Yongbo, LI Dong. Lightweight helmet wearing detection algorithm of improved YOLOv5[J]. Computer Engineering and Applications, 2022, 58 (9): 201-207. doi: 10.3778/j.issn.1002-8331.2111-0346
[12]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]. Proceedings of the European Conference on Computer Vision(ECCV), 2018: 3-19.
[13]	TAN Mingxing, PANG Ruoming, LE Q V, et al. EfficientDet: scalable and efficient object detection[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020:10781-10 790.
[14]	GEVORGYAN Z. SIoU Loss: more powerful learning for bounding box regression[EB/OL]. (2018-12-12). https://export.arxiv.org/ftp/arxiv/papers/2205/2205.12740.pdf.

标签类别	标签数量	备注
密目网	782	防护网未密封类隐患
剪刀撑	544	剪刀撑不规范搭设类隐患
异物	1 297	架体集中荷载有异物类隐患

参数	数值	参数	数值
初始学习率	0.001	权重衰减系数	0.000 5
迭代批次	16	训练轮次	100
非极大抑制阈值	0.3	置信度阈值	0.5

模型	注意力模块		SIoU Loss	BiFPN	P/%	R/%	mAP/%		参数量/ 10⁶
模型	CBAM	SE	SIoU Loss	BiFPN	P/%	R/%	mAP@.5	mAP@.5:.95	参数量/ 10⁶
YOLOv5s	—	—	—	—	91.00	89.43	92.66	55.72	7.06
A	√	—	—	—	93.45	89.67	94.05	58.55	7.11
B	—	√	—	—	90.10	89.0	92.14	55.22	7.59
C	—	—	√	—	93.46	91.11	94.81	58.93	7.06
D	—	—	—	√	92.17	89.81	93.54	56.13	8.13
E	√		√	—	92.10	92.52	93.20	58.98	7.11
F	√	—	—	√	90.97	92.14	94.42	58.18	8.18
G	—	—	√	√	90.05	92.82	93.88	56.55	8.13
YOLOv5_SW	√	—	√	√	92.90 (+1.9%)	92.88 (+3.45%)	94.83 (+2.17%)	60.85 (+5.13%)	8.18 (+1.12%)

网络模型	P/%	R/%	mAP@.5/%	FPS/(帧·s^-1)	参数量/10⁶	浮点运算数/10⁹
YOLOv5s	91.00	89.43	92.66	97.0	7.06	16.500
YOLOv4	90.80	89.62	89.61	47.9	63.95	29.983
YOLOv3	91.60	85.67	88.45	71.7	61.54	32.806
SSD-VGG16	88.31	79.14	88.31	143.0	26.29	30.477
Faster R-CNN-VGG16	69.99	92.62	90.37	29.5	136.73	184.884
YOLOv5_SW	92.90	92.88	94.83	83.3	8.18	17.900