基于改进Inception-v3网络的肺炎检测方法  

Pneumonia detection method based on improved Inception-v3 network

作　　者：蒲秋梅[1] 田景龙 邢容畅 赵丽娜 PU Qiu-mei;TIAN Jing-long;XING Rong-chang;ZHAO Li-na(School of Information Engineering,Minzu University of China,Beijing 100081,China;Multi-Disciplinary Research Division,Institute of High Energy Physics,The Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中央民族大学信息工程学院,北京100081 [2]中国科学院高能物理研究所多学科研究中心,北京100049

出　　处：《东北师大学报(自然科学版)》2023年第4期67-76,共10页Journal of Northeast Normal University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(31971311).

摘　　要：在Inception-v3网络的基础上提出了一种有效的改进方法以提升X光胸片肺炎诊断的准确度.在原始Inception-v3网络结构中引入了残差连接以减轻网络加深带来的梯度消失问题.使用空间注意力与通道注意力机制提升模型的特征提取能力,同时通过通道混洗(Channel Shuffle)促进不同通道特征图之间的信息融合.最终改进模型的分类准确率为94.64%,较原始Inception-v3网络提升了4.33%,且对于新型冠状病毒感染具有更高的召回率,达到99.72%.实验结果表明,引入注意力机制的Inception-v3网络在四类肺炎检测分类任务(正常、普通病毒性肺炎、新型冠状病毒感染、其他肺部感染)中具有更高泛化能力与鲁棒性.Based on the Perception-v3 network,this paper proposes an effective method to improve the accuracy of X-ray chest diagnosis of pneumonia.The introduction of spatial attention and channel attention mechanism in the original Inception-v3 network structure improves the feature extraction ability of the model,and promotes the information fusion between different channel feature maps through channel shuffle.The classification accuracy of the improved model is 94.6%,which is 4.3%higher than that of the original Inception-v3 network and has a higher recall rate for COVID-19,reaching 99.72%.The experimental results show that Inception-v3 network with attention mechanism has higher generalization ability and robustness in four types of pneumonia detection and classification tasks(normal,common viral pneumonia,COVID-19,other lung infections).

关 键 词：肺炎检测 X光胸片 Inception-v3 注意力机制模块 通道混洗 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]