基于自注意力卷积的新冠肺炎CXR图像识别  

COVID-19 CXR Image Recognition Based on Self Attention Convolution

作　　者：吕嫄 LYU Yuan(Basic Teaching Department,Wuhu Institute of Technology,Wuhu 241000,China)

机构地区：[1]芜湖职业技术学院基础教学部,安徽芜湖241000

出　　处：《安徽师范大学学报（自然科学版）》2023年第3期217-221,共5页Journal of Anhui Normal University(Natural Science)

基　　金：安徽省教育厅科技项目(KJ2020A0914);安徽省教育厅项目“技术技能型大师工作室”(2020dsgzs40)。

摘　　要：2019冠状病毒病(COVID-19)是近年来对世界经济发展影响最大的流行病。早期发现是治疗COVID-19患者的关键,而胸片作为一种快速有效的辅助诊断方法被广泛用于实际的医疗案例中。基于深度学习的图像识别方法能更快、更准确地诊断CXR图像,可以取得较好的效果。然而,常见的深度学习模型在对数据进行特征提取时没有针对性。对此,本文提出基于卷积注意力的新冠肺炎图像识别网络,提升对COVID-19阳性样本的敏感性和特异性,并且增加的模型参数量和训练时间可以忽略不计。本文结合VGG16、MobileNet、InceptionV3、ResNet50等经典深度学习网络搭建了卷积注意力模型,并在COVIDRD公开数据库上进行了验证。实验结果表明本文提出的网络架构有效的提升了对新冠肺炎识别的准确性、敏感性和特异性。COVID-19 is the epidemic that has the greatest impact on the world economic development in recent years.Early detection is the key to the treatment of COVID-19 patients,and chest radiograph as a fast and effective auxiliary diagnostic method is widely used in practical medical cases.The image recognition method based on depth learning can diagnose CXR images faster and more accurately,and can achieve good results.However,common deep learning models are not targeted when extracting features from data.In this regard,this paper proposes a new coronal pneumonia image recognition network based on convolutional attention to improve the sensitivity and specificity of COVID-19 positive samples,and the increased model parameters and training time can be ignored.This paper builds a convolutional attention model based on VGG16,MobileNet,InceptionV3,ResNet50 and other classic deep learning networks,and verifies it on the COVIDRD public database.The experimental results show that the network architecture proposed in this paper effectively improves the accuracy,sensitivity and specificity of new coronal pneumonia recognition.

关 键 词：新冠肺炎 CXR图像 深度学习 卷积注意力 

分 类 号：P315.69[天文地球—地震学]