基于YOLOv3的轻量化口罩检测算法研究
Research on Lightweight Mask Detection Algorithm Based on YOLOv3
DOI: 10.12677/CSA.2022.126170, PDF,   
作者: 张寿明：昆明理工大学信息工程与自动化学院，云南 昆明；刘 凯：昆明理工大学信息工程与自动化学院，云南 昆明;昆明理工大学云南省人工智能重点实验室，云南 昆明
关键词: 深度学习；目标检测；轻量化；YOLOv3；通道注意力机制；CIOU；Deep Learning； Object Detection； Lightweight； YOLOv3； Channel Attention Mechanism； CIOU
摘要: 针对当前基于深度学习的口罩检测算法在实时性与检测精度上不能同时具有良好的表现性能，本文提出一种基于YOLOv3的轻量化口罩检测算法，通过EfficientNet-B1网络替换掉原有的网络参数量大，网络结构复杂的骨干网络Darknet-53，为进一步提升网络性能，实验引入ECA通道注意力机制与特征金字塔结构相结合，最后采用CIOU对原有的边界框损失进行优化。实验结果表明，该网络结构模型与YOLOv3相比，检测精度仅降低1.73%，但模型参数量降低了79%，且单张图片检测速度也提升了3.93倍，一定程度上体现了本文算法的良好性能。
Abstract: In view of the fact that the current deep learning-based mask detection algorithm cannot have good performance in real-time and detection accuracy at the same time, this thesis proposes a light-weight mask detection algorithm based on YOLOv3, which replaces the original backbone network Darknet-53 which has a large number of network parameters and a complex network structure through the EfficientNet-B1 network. In order to further improve the network performance, the experiment introduces the ECA channel attention mechanism combined with the feature pyramid structure, and finally uses CIOU to optimize the original bounding box loss. The experimental results show that the network structure model is compared with YOLOv3. The detection accuracy is only reduced by 1.73%, but the amount of model parameters is reduced by 79%, and the detection speed of a single image is also increased by 3.93 times, which reflects the good performance of the algorithm in this paper to a certain extent.
文章引用：张寿明, 刘凯. 基于YOLOv3的轻量化口罩检测算法研究[J]. 计算机科学与应用, 2022, 12(6): 1700-1709. https://doi.org/10.12677/CSA.2022.126170

参考文献

[1] Chavez, S., Long, B., Koyfman, A. and Stephen, Y. (2020) Coronavirus Disease (COVID-19): A Primer for Emergency Physicians. American Journal of Emergency Medicine, 44, 220-229. [Google Scholar] [CrossRef] [PubMed]
[2] Tian, Z., Shen, C. and Chen, H. (2020) FCOS: Fully Convolution-al One-Stage Object Detection. 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, 27-28 October 2019, 9627-9636. [Google Scholar] [CrossRef] 
[3] Chen, P., Li, Y. and Zhou, H. (2020) Detection of Small Ship Ob-jects Using Anchor Boxes Cluster and Feature Pyramid Network Model for SAR Imagery. Journal of Marine Science and Engineering, 8, 112. [Google Scholar] [CrossRef] 
[4] 张路达, 邓超. 多尺度融合的YOLOv3人群口罩佩戴检测方法[J]. 计算机工程与应用, 2021, 57(16): 283-290.
[5] Redmon, J. and Farhadi, A. (2018) YOLOv3: An Incremental Im-provement.
[6] 李雨阳, 沈记全, 翟海霞, 冯伟华. 基于改进SSD的口罩佩戴检测算法[J/OL]. 计算机工程, 1-9. 2022-06-29.[CrossRef] 
[7] Liu, W., Anguelov, D. and Erhan, D. (2016) SSD: Single Shot MultiBox Detector. Springer, Cham. [Google Scholar] [CrossRef] 
[8] 王艺皓, 丁洪伟, 李波, 杨志军, 杨俊东. 复杂场景下基于改进YOLOv3的口罩佩戴检测算法[J]. 计算机工程, 2020, 46(11): 11.
[9] He, K., Zhang, X. and Ren, S. (2016) Deep Residual Learning for Image Recognition. [Google Scholar] [CrossRef] 
[10] Gong, H., Li, H. and Xu, K. (2019) Object Detection Based on Im-proved YOLOv3-Tiny. 2019 Chinese Automation Congress (CAC), Hangzhou, 22-24 November 2019, 3240-3245. [Google Scholar] [CrossRef] 
[11] Chollet, F. (2017) Xception: Deep Learning with Depth-wise Separable Convolutions. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, 21-26 July 2017, 1251-1258. [Google Scholar] [CrossRef] 
[12] Mahendran, A. and Vedaldi, A. (2016) Visualizing Deep Convolu-tional Neural Networks Using Natural Pre-Images. International Journal of Computer Vision, 120, 233-255. [Google Scholar] [CrossRef] 
[13] Lin, M., Chen, Q. and Yan, S. (2013) Network in Network. Computer Science.
[14] Han, S., Pool, J., Tran, J. and William, J.D. (2015) Learning Both Weights and Connections for Efficient Neural Networks. Proceedings of the 28th International Conference on Neural Information Processing Systems, Volume 1, 1135-1143.
[15] Geoffrey, H., Oriol, V. and Jeff, D. (2015) Distilling the Knowledge in a Neural Network. Computer Science, 14, 38-39.
[16] Mingxing, T. and Quoc, V.L. (2019) EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks.
[17] Hu, J., Shen, L., Sun, G., Albanie, S. and Wu, E.H. (2017) Squeeze-and-Excitation Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42, 2011-2023.
[18] Wang, Q.L., et al. (2019) ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Net-works. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2020, Seattle, June 2020, 13-19.