基于增强特征融合网络的安全帽佩戴检测

doi:10.16180/j.cnki.issn1007-7820.2023.04.006

摘要/Abstract

摘要：

佩戴安全帽是保证工人施工安全的重要方式之一。现有的安全帽检测器的检测精度与速度都有待提高,这使得这些检测器难以大规模应用于实际的生产活动中。针对这些问题,文中推出了基于EfficientDet的安全帽检测器,并在此基础上从特征融合的角度对其进行了改进。该模型通过使用特征补充的方式减少了特征融合过程中的信息损失,并利用改进的特征金字塔及自适应空间融合模块提升了融合的效率,最终达到提升性能的目的。实验表明,文中改进的模型在安全帽佩戴数据集上的精确率达到83.03%,相较于未改进的模型有所提升,且模型大小没有明显增加。该模型在PASCAL VOC 2007上的精确率则达到了82.76%。

关键词: 安全帽佩戴检测, 特征融合, 特征金字塔, 目标检测, EfficientDet, 空间融合, 深度学习, 卷积神经网络

Abstract:

Wearing safety helmet is one of the important ways to ensure the safety of workers in production activities. The detection accuracy and speed of the existing helmet detector need to be improved, which makes it difficult for the existing detectors to be applied in real production activities on a large scale. To solve these problems, the helmet detector based on EfficientDet is introduced and improved from the perspective of feature fusion. Specifically, the model reduces the information loss in the process of feature fusion using feature supplement module, and improves the efficiency of feature fusion using improved feature pyramid and adaptive spatial fusion module, and finally achieves the goal of improving performance. Experimental results show that the accuracy of the improved model on the helmet wearing data set is 83.03%, and the model size does not increase significantly. The accuracy of the model on PASCAL VOC 2007 data set is 82.76%.

Key words: helmet wearing detection, feature fusion, feature pyramid, object detection, EfficientDet, spatial fusion, deep learning, convolutional neural network

中图分类号:

TP391

崔卓栋,陈玮,尹钟. 基于增强特征融合网络的安全帽佩戴检测[J]. 电子科技, 2023, 36(4): 44-51.

CUI Zhuodong,CHEN Wei,YIN Zhong. Helmet Wearing Detection Based on Enhanced Feature Fusion Network[J]. Electronic Science and Technology, 2023, 36(4): 44-51.

图/表 16

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

图11

表1

图12

表2

表3

图13

参考文献 22

[1]	Du S, Shehata M, Badawy W. Hard hat detection in video sequences based on face features, motion and color information[C]. Shanghai: Proceedings of the Third International Conference on Computer Research and Development, 2011.
[2]	Fang Q, Li H, Luo X C. Detection non-hardhat-use by a deep learning method from far field surveillance videos[J]. Automation in Construction, 2018, 85(1):1-9. doi: 10.1016/j.autcon.2017.09.018
[3]	Ren S Q, He K M, Girshick R, et al. Faster R-CNN:To-wards real-time object detection with region proposal networks[J]. Advances in Neural Information Processing Systems, 2015, 28(2):91-99.
[4]	方明, 孙腾腾, 邵桢. 基于改进YOLOv2的快速安全帽佩戴检测情况检测[J]. 光学精密工程, 2019, 27(5):1196-1205. doi: 10.3788/OPE.
	Fang Ming, Sun Tengteng, Shao Zhen. Fast helmet-wearing condition detection based on improved YOLOv2[J]. Optics and Precision Engineering, 2019, 27(5):1196-1205. doi: 10.3788/OPE.
[5]	Redmon J, Farhadi A. YOLO9000:Better,faster,stronger[C]. Honolulu: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017.
[6]	Howard A G, Zhu M L, Chen B, et al. MobileNets:Efficient convolutional neural networks for mobile vision applications[C]. Honolulu: Proceedings of the International Conference on Computer Vision and Pattern Recognition, 2017.
[7]	徐守坤, 王雅如, 顾玉宛, 等. 基于改进Faster RCNN的安全帽佩戴检测研究[J]. 计算机应用研究, 2020, 37(3):901-905.
	Xu Shoukun, Wang Yaru, Gu Yuwan, et al. Safety helmet wearing detection study based on improved Faster RCNN[J]. Application Research of Computers, 2020, 37(3):901-905.
[8]	仝泽友, 冯仕民, 侯晓晴, 等. 基于安全帽佩戴检测的矿山人员违规行为研究[J]. 电子科技, 2019, 32(9):26-31.
	Tong Zeyou, Feng Shimin, Hou Xiaoqing, et al. Recognition of underground miners’ rule-violated behavior based on safety helmet detection[J]. Electronic Science and Technology[J]. 2019, 32(9):26-31.
[9]	Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[J]. Advances in Neural Information Processing Systems, 2012, 25(12):1097-1105.
[10]	Simonyan K, Zisserman A. Very deep convolutional networks for largescale image recognition[C]. Banff: Proceedings of the International Conference on Learning Representations, 2014.
[11]	Liu W, Anguelov D, Erhan D, et al. SSD:Single shot multibox detector[C]. Amsterdam: Proceedings of the European Conference on Computer Vision, 2016.
[12]	Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C]. Washington,D.C.:Proce-edings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017.
[13]	Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]. Venice: Proceedings of the IEEE International Conference on Computer Vision, 2017.
[14]	Bochkovskly A, Wang C Y, Liao H Y M. YOLOv4:Optimal speed and accuracy of object detection[C]. Seattle: Proceedings of the International Conference on Computer Vision and Pattern Recognition, 2020.
[15]	Tan M, Pang R, Le Q V. EfficientDet:Scalable and effi-cient object detection[C]. Washington,D.C.:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020.
[16]	Liu S, Qi L, Qin H, et al. Path aggregation network for instance segmentation[C]. Washington,D.C.:Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018.
[17]	Guo C X, Fan B, Zhang Q, et al. AugFPN:Impeoving multi-scale feature learning for object detection[C]. Seattle: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020.
[18]	Liu S T, Huang D, Wang Y H. Learning spatial fusion for single-shot object detection[J]. Computing Research Repository, 2019,1-10.
[19]	王平, 江雨泽, 赵光辉. 目标检测的多尺度定位提升算法[J]. beoplay体育提现学报, 2021, 48(3):85-90.
	Wang Ping, Jiang Yuze, Zhao Guanghui. Object detectionbased on the multiscale location enhancement network[J]. Journal of Xidian University, 2021, 48(3):85-90.
[20]	袁帅, 王康, 单义, 等. 基于多分支并行空洞卷积的多尺度目标检测算法[J]. 计算机辅助设计与图形学学报, 2021, 33(6):864-872.
	Yuan Shuai, Wang Kang, Shan Yi, et al. Multi-scale obj-ect detection method based on multi-branch parallel dilated convolution[J]. Journal of Computer-Aided Design & Computer Graphics, 2021, 33(6):864-872.
[21]	Liu S T, Huang D. Receptive filed block net for accurate and fast object detection[C]. Munich: Proceedings of the European Conference on Computer Vision, 2018.
[22]	刘晋, 邓洪敏, 徐泽林, 等. 面向目标识别的轻量化混合卷积神经网络[J]. 计算机应用, 2021, 41(10):1-8
	Liu Jin, Deng Hongmin, Xu Zelin, et al. Lightweight hybrid convolutional neural network for object recognition[J]. Journal of Computer Applications, 2021, 41(10):1-8.

模型	特征补充	改进 BiFPN	ASF	AP/%	FPS	权重文件大小/MB
SSD				78.20	20.6	100.3
YOLOv3				80.40	29.1	235.0
Faster-RCNN(VGG)				77.10	3.5	522.0
Faster-RCNN(ResNet)				78.30	7.2	108.0
CenterNet				81.80	18.1	125.0
RetinaNet				82.40	21.8	140.0
EfficientDet				81.19	25.3	15.0
EfficientDet	√			82.08	25.2	15.1
EfficientDet	√	√		82.32	25.0	15.8
EfficientDet	√	√	√	83.03	22.4	16.3

模型	主干网络	AP/%	FPS	权重文件大小/MB
RefineDet^[19]	VGG	81.80	34.1	78.1
YOLOv3^[19]	DarkNet	82.20	19.6	236.0
Faster-RCNN^[20]	ResNet	76.40	2.4	109.0
SDD^[20]	VGG	78.60	26.6	101.0
RFB512^[21]	VGG	82.20	38.0	190.0
改进的YOLOv4^[22]	MobileNet	82.50	29.7	20.6
EfficientDet	EfficientNet	81.61	26.9	15.0
Ours	EfficientNet	82.76	24.2	16.3

类别	AP/%
Horse	98
Sheep	96
TV monitor	96
Train	95
Motorbike	95
Cat	94
Car	92
Aeroplane	91
Bicycle	91
Bus	88
Person	87
Dog	86
Cow	84
Bird	76
Bottle	72
Sofa	70
Chair	62
Boat	61
Potted plant	61
Dining table	59