摘 要：为了能够快速准确地诊断出新冠肺炎患者，文章参考 MobileNetV2 架构并结合注意力网络，改进损失优化函数，依据 CNN 网络设计准则搭建新型轻量网络 MHNet。在 COVIDx CXR-2 公开数据集上进行的实验表明，该网络在准确率、召回率、特异性、精准率、F1 分数、模型大小、CPU 单张图推理耗时、GPU 单张图推理耗时上的指标分别为 92%、99%、85%、86.84%、92.52%、3.91 MB、59.51 ms、17.66 ms。相较于其他传统网络，该网络对新冠肺炎感染者的诊断率较高、诊断效果较好。
关键词：新冠肺炎；ECA-Net；FocalLoss；高效 CNN 网络设计准则
DOI:10.19850/j.cnki.2096-4706.2022.013.020
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2022）13-0082-05
Study on the Identification of COVID-19 by Light Weight Network MHNet
HOU Linshuo, WANG Yin, LONG Qihang, LI Yuxiang, MA Shukang
(College of Mechatronics and Information Engineering, China University of Mining and Technology-Beijing, Beijing 100083, China)
Abstract: In order to diagnose COVID-19 patients more accurately and quickly, this paper refers to MobileNetV2 architecture and combines attention network, improves loss optimization function, and builds a new lightweight network MHNet according to CNN network design criteria. Experiments on a public COVIDx CXR-2 dataset show that the indicators of the network in accuracy, recall, specificity, accuracy, F1 score, model size, CPU single graph reasoning time, GPU single graph reasoning time are 92%, 99%, 85%, 86.84%, 92.52%, 3.91 MB, 59.51 ms, 17.66 ms respectively. Compared with other traditional networks, this network has higher diagnostic rate and better diagnostic effect for patients infected with covid-19 infection.
Keywords: COVID-19; ECA-Net; FocalLoss; Guidelines for efficient CNN network design
参考文献：
[1] 高静文，蔡永香，何宗宜 .TP-FER：基于优化卷积神经网络的三通道人脸表情识别方法 [J].计算机应用研究，2021，38（7）：2213-2219.
[2] 顾德英，王娜，李文超，等 .基于模型融合的低照度环境下车道线检测方法 [J]. 东北大学学报（自然科学版），2021，42（3）：305-309.
[3] 李沿宏，江茜，邹可，等 .融合注意力机制的多流卷积肌电手势识别网络 [J]. 计算机应用研究，2021，38（11）：3258-3263.
[4] KRIZHEVSKY A，SUTSKEVER I，HINTON G E. ImageNet Classification with Deep Convolutional Neural Networks [J]. Communications of the ACM，2017，60（6）：84-90.
[5] KAREN S，ANDREW Z. Very Deep Convolutional Networks for Large-Scale Image Recognition [EB/OL].[2022-04-13].https:// ui.adsabs.harvard.edu/abs/2014arXiv1409.1556S/abstract.
[6] HE K M，ZHANG X Y，REN S Q，et al. Deep Residual Learning for Image Recognition [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas：IEEE， 2016：770-778.
[7] SZEGEDY C，LIU W，JIA Y Q，et al. Going Deeper with Convolutions [C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Boston：IEEE，2015：1-9.
[8] IANDOLA F N，HAN S，MOSKEWICZ M W， et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size [J/OL].arXiv:1602.07360v4 [cs.CV].[2022-04-12]. https://arxiv.org/pdf/1602.07360.pdf.
[9] HOWARD A G，ZHU M，CHEN B，et al. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications [J/0L].arXiv:1704.04861 [cs.CV].[2022-04-12].https://arxiv.org/ abs/1704.04861.
[10] CO A，OA A，MM B . CVDNet: A novel deep learning architecture for detection of coronavirus (Covid-19) from chest x-ray images [J]. Chaos，Solitons & Fractals，2020，140：110245.
[11] BAYOUDH K，HAMDAOUI F，MTIBAA A. HybridCOVID: a novel hybrid 2D/3D CNN based on cross-domain adaptation approach for COVID-19 screening from chest X-ray images [J].Physical and Engineering Sciences in Medicine，2020，43（4）：1415-1431.
[12] 易三莉，王天伟，杨雪莲，等 .ARS-CNN 算法在新冠肺炎识别中的研究 [J]. 液晶与显示，2021，36（11）：1565-1572.
[13] CAO Y，XU J，LIN S，et al. GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond [EB/OL].[2022-04- 10].https://www.researchgate.net/publication/332669752_GCNet_Nonlocal_Networks_Meet_Squeeze-Excitation_Networks_and_Beyond.
[14] WANG Q L，WU B G，ZHU P F，et al. ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks [C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle：IEEE，2020：11531-11539.
[15] MA N N，ZHANG X Y，ZHENG H T，et al. ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design [J/ OL]. arXiv:1807.11164v1 [cs.CV].[2022-04-05].https://arxiv.org/ pdf/1807.11164.pdf.
[16] LIN T Y，GOYAL P，Girshick R，et al. Focal Loss for Dense Object Detection [J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2020，42（2）：318-327.
[17] HU J，LI S，ALBANIE S，et al. Squeeze-and-Excitation Networks [J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2020，42（8）：2011-2023.
作者简介：侯麟朔（1998—），男，汉族，河南南阳人，硕士研究生在读，研究方向：图像处理；王寅（1998—），男，汉族，江苏连云港人，硕士研究生在读，研究方向：图像处理；龙启航（1996—），男，汉族，湖北襄阳人，硕士研究生在读，研究方向：图像处理；李宇翔（1998—），男，汉族，湖南株洲人，硕士研究生在读，研究方向：图像处理；马淑康（1998—），男，汉族，山东临沂人，硕士研究生在读，研究方向：图像处理。