安全帽佩戴检测网络模型的轻量化设计

doi:10.19678/j.issn.1000-3428.0064219

摘要/Abstract

摘要： 现有的安全帽佩戴检测网络模型存在准确率低、推理速度慢、部署到边缘计算设备时精度和实时性均达不到应用要求等问题。提出一种轻量化设计的DT-YOLO模型，对YOLOv4-Tiny目标检测模型进行改进，通过增加一个检测层提高模型在密集场景下对小目标的检测能力，并引入空间金字塔池化模块，提高模型对不同尺寸目标的检测能力。使用局部稀疏因子衰减算法进行稀疏化训练，从而使经过稀疏化训练后模型的平均精度均值（mAP）得到提高。根据缩放系数判断通道的重要性，并进行模型的通道剪枝，压缩模型的大小和计算量。使用TensorRT推理加速引擎进行网络层水平和垂直融合，消除拼接层操作，并将参数压缩成16位浮点型，提高模型的推理速度，最后在Jeston Nano边缘计算设备上实现模型部署。实验结果表明，与YOLOv4-Tiny模型相比，DT-YOLO模型的mAP提高了3.6个百分点，模型大小减少了83.5%，帧率提高137.7%，能够满足安全帽佩戴检测的要求。

关键词: 安全帽佩戴检测, YOLOv4-Tiny模型改进, 局部稀疏因子衰减, 模型压缩, TensorRT推理加速引擎, Jeston Nano边缘计算设备

Abstract: The existing safety helmet wearing detection network model has such problems as low accuracy, slow reasoning speed, and the accuracy and real-time performance when deployed to edge computing devices can not meet the application requirements.Accordingly, a lightweight DT-YOLO model is proposed, the YOLOv4-Tiny target detection model features two significant improvements.Specifically, a detection layer is added to improve the detection of small targets in dense scenes, and a Spatial Pyramid Pooling(SPP) module is used to improve the detection of variable-sized targets.Secondly, the local sparsity factor attenuation algorithm is used during sparsity training to improve the model's mean Average Precision(mAP).Subsequently, the importance of the channel is judged according to the scaling factor, and channel pruning is carried out to compress the size and calculation burden of the model.Thirdly, the TensorRT reasoning acceleration engine is used to fuse the horizontal and vertical network layers, eliminate the splicing layer operation, and compress the parameters into 16 bit floating-point type, thereby accelerating the model's reasoning.Finally, the model is deployed on the edge computing device Jeston Nano.Experimental results indicate that compared with that of the YOLOv4-Tiny model, the mAP of DT-YOLO is increased by 3.6 percentage points, the model volume is reduced by 83.5%, and the frame rate is increased by 137.7%, and the proposed model satisfies the requirements of helmet wearing detection.

Key words: safety helmet wearing detection, YOLOv4-Tiny model improvement, local sparsity factor attenuation, model compression, TensorRT reasoning acceleration engine, Jeston Nano edge computing device

中图分类号:

TP391.41

郭奕裕, 周箩鱼. 安全帽佩戴检测网络模型的轻量化设计[J]. 计算机工程, 2023, 49(4): 312-320.

GUO Yiyu, ZHOU Luoyu. Lightweight Design of Safety Helmet Wearing Detection Network Model[J]. Computer Engineering, 2023, 49(4): 312-320.

https://www.ecice06.com/CN/Y2023/V49/I4/312

图/表 19

20230417190806

20230417190809

20230417190812

20230417190815

20230417190819

20230417191029

20230417191032

20230417191036

20230417191039

20230417191042

20230417191045

20230417191048

20230417191051

20230417191054

20230417191059

20230417191321

20230417191325

20230417191329

20230417191332

参考文献

[1] 边星, 晋良海, 陈雁高, 等.施工作业人员佩戴安全帽行为意向研究[J].中国安全科学学报, 2016, 26(11):43-48. BIAN X, JIN L H, CHEN Y G, et al.Research on workers' behavioral intention of wearing safety helmet[J].China Safety Science Journal, 2016, 26(11):43-48.(in Chinese)
[2] CHEN S B, TANG W H, JI T Y, et al.Detection of safety helmet wearing based on improved faster R-CNN[C]//Proceedings of International Joint Conference on Neural Networks.Washington D.C., USA:IEEE Press, 2020:1-7.
[3] 许凯, 邓超.基于改进YOLOv3的安全帽佩戴识别算法[J].激光与光电子学进展, 2021, 58(6):300-307. XU K, DENG C.Research on helmet wear identification based on improved YOLOv3[J].Laser & Optoelectronics Progress, 2021, 58(6):300-307.(in Chinese)
[4] 郭继峰, 孙文博, 庞志奇, 等.一种改进YOLOv4的交通标志识别算法[J].小型微型计算机系统, 2022, 43(7):1471-1476. GUO J F, SUN W B, PANG Z Q, et al.Improved traffic sign recognition algorithm for Yolov4[J].Journal of Chinese Mini-Micro Computer Systems, 2022, 43(7):1471-1476.(in Chinese)
[5] 赵睿, 刘辉, 刘沛霖, 等.基于改进YOLOv5s的安全帽检测算法[J/OL].北京航空航天大学学报:1-16[2021-11-23].https://doi-org-s.z.library.sh.cn/10.13700/j.bh.1001-5965.2021.0595. ZHAO R, LIU H, LIU P L, et al.Research on safety helmet detection algorithm based on improved YOLOv5s[J/OL].Journal of Beijing University of Aeronautics and Astronautics:1-16[2021-11-23].https://doi-org-s.z.library.sh.cn/10.13700/j.bh.1001-5965.2021.0595.(in Chinese)
[6] 徐先峰, 赵万福, 邹浩泉, 等.基于MobileNet-SSD的安全帽佩戴检测算法[J].计算机工程, 2021, 47(10):298-305, 313. XU X F, ZHAO W F, ZOU H Q, et al.Detection algorithm of safety helmet wear based on MobileNet-SSD[J].Computer Engineering, 2021, 47(10):298-305, 313.(in Chinese)
[7] CHENG R, HE X W, ZHENG Z L, et al.Multi-scale safety helmet detection based on SAS-YOLOv3-tiny[J].Applied Sciences, 2021, 11(8):3652-3662.
[8] WANG H X, GE H Y, LI M X.PFG-YOLO:a safety helmet detection based on YOLOv4[C]//Proceedings of the 5th Information Technology, Networking, Electronic and Automation Control Conference.Washington D.C., USA:IEEE Press, 2021:1242-1246.
[9] HE K M, ZHANG X Y, REN S Q, et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9):1904-1916.
[10] 张有波, 郭威, 周悦, 等.基于多粒度剪枝的水下遗迹实时目标检测[J].激光与光电子学进展, 2021, 58(14):286-295. ZHANG Y B, GUO W, ZHOU Y, et al.Real-time target detection of underwater relics based on multigranularity pruning[J].Laser & Optoelectronics Progress, 2021, 58(14):286-295.(in Chinese)
[11] 韦越, 陈世超, 朱凤华, 等.基于稀疏正则化的卷积神经网络模型剪枝方法[J].计算机工程, 2021, 47(10):61-66. WEI Y, CHEN S C, ZHU F H, et al.Pruning method for convolutional neural network models based on sparse regularization[J].Computer Engineering, 2021, 47(10):61-66.(in Chinese)
[12] 杨民杰, 梁亚玲, 杜明辉.基于参数子空间和缩放因子的YOLO剪枝算法[J].计算机工程, 2021, 47(2):111-117. YANG M J, LIANG Y L, DU M H.YOLO pruning algorithm based on parameter subspace and scaling factor[J].Computer Engineering, 2021, 47(2):111-117.(in Chinese)
[13] YOO H B, PARK M S, KIM S H.Real time face detection method using TensorRT and SSD[J].KIPS Transactions on Software and Data Engineering, 2020, 9(10):323-328.
[14] 周立君, 刘宇, 白璐, 等.使用TensorRT进行深度学习推理[J].应用光学, 2020, 41(2):337-341. ZHOU L J, LIU Y, BAI L, et al.Using TensorRT for deep learning and inference applications[J].Journal of Applied Optics, 2020, 41(2):337-341.(in Chinese)
[15] 张倩宇, 贾维, 彭博.基于深度学习的三叉神经区域自动检测及TensorRT加速[J].太赫兹科学与电子信息学报, 2021, 19(6):1065-1069. ZHANG Q Y, JIA W, PENG B.Automatic detection of trigeminal neural region based on deep learning and TensorRT acceleration[J].Journal of Terahertz Science and Electronic Information Technology, 2021, 19(6):1065-1069.(in Chinese)
[16] DEBAUCHE O, SAD M, MAHMOUDI S A, et al.Edge computing and artificial intelligence for real-time poultry monitoring[J].Procedia Computer Science, 2020, 175(8):534-541.
[17] YU S, HE Y, SHAO M.Real-time target detection based on jetson nano and YOLOv3-Tiny[J].International Core Journal of Engineering, 2021, 7(2):378-383.
[18] BOCHKOVSKIY A, WANG C Y, LIAO H Y M.YOLOv4:optimal speed and accuracy of object detection[EB/OL].[2022-02-15].https://arxiv.org/abs/2004.10934.
[19] 胡嘉沛, 李震, 黄河清, 等.采用改进YOLOv4-Tiny模型的柑橘木虱识别[J].农业工程学报, 2021, 37(17):197-203. HU J P, LI Z, HUANG H Q, et al.Citrus psyllid detection based on improved YOLOv4-Tiny model[J].Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(17):197-203.(in Chinese)
[20] LI X, PAN J, XIE F, et al.Fast and accurate green pepper detection in complex backgrounds via an improved YOLOv4-Tiny model[J].Computers and Electronics in Agriculture, 2021, 191(11):503-513.
[21] 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, 22(8):347-359.
[22] GUO C, LV X L, ZHANG Y, et al.Improved YOLOv3-Tiny network for real-time electronic component detection[J].Scientific Reports, 2021, 11(9):44-56.
[23] YANG Q F, ZHANG Z Q, YAN L, et al.Lightweight bird's nest location recognition method based on YOLOv3-Tiny[C]//Proceedings of IEEE International Conference on Electrical Engineering and Mechatronics Technology.Washington D.C., IEEE Press, 2021:402-405.
[24] IOFFE S, SZEGEDY C.Batch normalization:accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on Machine Learning.New York, USA:ACM Press, 2015:448-456.
[25] 张杰妹, 杨词慧.基于RV-FCN的CT肝脏影像自动分割算法[J].计算机工程, 2019, 45(7):258-263. ZHANG J M, YANG C H.Automatic segmentation algorithm of CT liver image based on RV-FCN[J].Computer Engineering, 2019, 45(7):258-263.(in Chinese)
[26] HE Y, DONG X Y, KANG G L, et al.Asymptotic soft filter pruning for deep convolutional neural networks[J].IEEE Transactions on Cybernetics, 2020, 50(8):3594-3604.
[27] LIU Z, LI J G, SHEN Z Q, et al.Learning efficient convolutional networks through network slimming[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2017:2755-2763.

选择文件类型/文献管理软件名称

选择包含的内容