YOLOv5目标检测的轻量化研究

doi:10.3778/j.issn.1002-8331.2206-0231

摘要/Abstract

摘要： 现有目标检测算法通常存在体积较大、结构复杂等问题，致使室内机器人作业过程中识别速率与精度较差。针对这一问题，以室内目标检测为基础，提出了一种改进的YOLOv5s轻量化检测方法。该方法主要是在YOLOv5s网络的基础上引入ShuffleNet v2特征提取机制来实现网络的轻量化，同时采用加权双向特征金字塔BiFPN和边框回归损失EIOU获取特征信息更为丰富的特征图，来提升目标检测精度，从而得到一种新的室内目标检测模型。研究结果表明，改进后的模型参数量明显减少，模型复杂度减少了46%，平均精确率均值mAP提升到63.9%，实现了轻量化和检测准确率的平衡，该研究为目标轻量化研究提供了参考。

关键词: 目标检测, YOLOv5s, ShuffleNetv2, 轻量化, 边框回归损失EIOU

Abstract: The existing target detection algorithms usually have problems such as large size and complex structure, which leads to poor recognition speed and accuracy in the process of indoor robot operation. To solve this problem, based on indoor target detection, a lightweight detection method of improved YOLOv5s is proposed. This method mainly introduces the ShuffleNet v2 feature extraction mechanism based on the YOLOv5s network to realize the lightweight of the network. At the same time, the weighted bidirectional feature pyramid BiFPN and the frame regression loss EIOU are used to obtain the feature map with richer feature information to improve the accuracy of target detection, so as to obtain a new indoor target detection model. The experimental results show that the parameters of the improved model are significantly reduced, the complexity of the model is reduced by 46%, and the average accuracy rate is increased to 63.9%. The balance between lightweight and detection accuracy is achieved, which provides a reference for target lightweight research.

Key words: target detection, YOLOv5s, ShuffleNetv2, lightweight, frame regression loss EIOU

何雨, 田军委, 张震, 王沁, 赵鹏. YOLOv5目标检测的轻量化研究[J]. 计算机工程与应用, 2023, 59(1): 92-99.

HE Yu, TIAN Junwei, ZHANG Zhen, WANG Qin, ZHAO Peng. Lightweight Research of YOLOv5 Target Detection[J]. Computer Engineering and Applications, 2023, 59(1): 92-99.

参考文献

[1] LóPEZ-CIFUENTES A，ESCUDERO-VI?OLO M，BESCóS J，et al.Semantic-aware scene recognition[J].Pattern Recognition，2020，102：107256.
[2] SONG X，JIANG S，HERRANZ L，et al.Learning effective RGB-D representations for scene recognition[J].IEEE Transactions on Image Processing，2018，28（2）：980-993.
[3] DALAL N，TRIGGS B.Histograms of oriented gradients for human detection[C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition，2005：886-893.
[4] WANG P，SHEN C H，BARNES N，et al.Fast and robust object detection using asymmetric totally corrective boosting[J].IEEE Transactions on Neural Networks and Learning Systems，2012，23（1）：33-46.
[5] HEDAU V，HOIEM D，FORSYTH D.Recovering the spatial layout of cluttered rooms[C]//Proceedings of the IEEE 12th International Conference on Computer Vision，2009.
[5] HEDAU V，HOIEM D，FORSYTH D.Recovering the spatial layout of cluttered rooms[C]//Proceedings of the IEEE 12th International Conference on Computer Vision，2009：1849-1856.
[6] TORRALBA A，SINHA P.Recognizing indoor scenes[C]//Proceedings of the Conference on Computer Vision & Pattern Recognition，2009.
[6] QUATTONI A，TORRALBA A.Recognizing indoor scenes[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2009：413-420.
[7] CHENG X J，LU J W，FENG J J，et al.Scene recognition with objectness[J].Pattern Recognition，2018，74：474-487.
[8] GIRSHICK R，DONAHUE J，DARRELL T，et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2014：580-587.
[9] LI J N，LIANG X D，SHEN S M，et al.Scale-aware fast R-CNN for pedestrian detection[J].IEEE Transactions on Multimedia，2018，20（4）：985-996.
[10] GIRSHICK R.Scale-aware fast R-CNN for pedestrian detection[J].Computer Science，2015（1）.
[11] REDMON J，DIVVALA S，GIRSHICK R，et al.You only look once：Unified，real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2016：779-788.
[12] LIU W，ANGUELOV D，ERHAN D，et al.SSD：Single shot MultiBox detector[C]//Proceedings of the IEEE Conference on Computer Vision，2016.
[13] 李仁鹰，钱慧芳，郭佳豪.基于M-YOLOV4模型的轻量化目标检测算法[J].国外电子测量技术，2022，41（4）：15-21.
LI R Y，QIAN H F，GUO J H.Lightweight target detection algorithm based on M-YOLOV4 model[J].Foreign Electronic Measurement Technology，2022，41（4）：15-21.
[14] 王晨，袁庆霓，白欢，等.面向仓储货物的轻量化目标检测算法[J/OL].激光与光电子学进展：1-10（2021-11-10）[2022-05-31].https：//kns.cnki.net/kcms/detail/31.1690.tn20211110.
1211.024.html.
WANG C，YUAN Q N，BAI H，et al.Lightweight object detection algorithm for storage goods[J/OL].Laser & Optoelectronics Progress：1-10（2021-11-10）[2022-05-31].https：//kns.cnki.net/kcms/detail/31.1690.tn20211110.1211.
024.html.
[15] 秦伟伟，宋泰年，刘洁瑜，等.基于轻量化YOLOv3的遥感军事目标检测算法[J].计算机工程与应用，2021，57（21）：263-269.
QIN W W，SONG T N，LIU J Y，et al.Remote sensing military target detection algorithm based on lightweight YOLOv3[J].Computer Engineering and Applications，2021，57（21）：263-269.
[16] 李雨诗，张才裕，赵杨珂，等.基于模型压缩的轻量化障碍物检测模型研究[J].激光杂志，2022，43（9）：38-43.
LI Y S，ZHANG C Y，ZHAO Y K，et al.Research on lightweight obstacle detection model based on model compression[J].Laser Journal，2022，43（9）：38-43.
[17] 范良辰，夏颖慧，李雨诗，等.基于网络剪枝的轻量化道路损坏检测模型研究[J].信息技术，2022（3）：96-102.
FAN L C，XIA Y H，LI Y S，et al.Research on lightweight road damage detection model based on network pruning[J].Information Technology，2022（3）：96-102.
[18] TAN L，LV X Y，LIAN X F，et al.YOLOv4_Drone：UAV image target detection based on an improved YOLOv4 algorithm[J].Computers & Electrical Engineering，2021，93：107261.
[19] MA N N，ZHANG X Y，ZHENG H T，et al.ShuffleNet V2：Practical guidelines for efficient CNN architecture design[C]//Proceedings of the European Conference on Computer Vision，2018.
[20] ZHANG Y F，REN W Q，ZHANG Z，et al.Focal and efficient IOU loss for accurate bounding box regression[J].Neurocomputing，2022，506：146-157.
[21] BRITZ D，GOLDIE A，LUONG M T，et al.Massive exploration of neural machine translation architectures[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing，2017.