注意力引导梯形金字塔融合网络识别新冠肺炎X射线影像

新型冠状病毒肺炎(corona virus disease 2019,COVID-19)严重影响人类社会和经济的发展，威胁人类的健康。如何更准确、快速地排查感染病毒的患者，使用卷积神经网络(convolutional neural network, CNN)的方法识别COVID-19胸部X射线影像，完成计算机自动辅助诊断。但是，由于识别精度不高，难以准确判断是否感染了COVID-19。为了提高网络模型对COVID-19胸部X射线影像的识别性能，首先提出注意力引导梯形金字塔融合网络(attention steered trapezoid pyramid fusion network, ASTPNet),该网络可以附加在不同的CNN上，有效地利用模型中深层与浅层网络的特点；其次提出注意力引导块(attention steered block, AS Block),通过通道和空间注意力，强调通道和空间中的有效语义信息，弱化无效的干扰信息，高效地聚合加权信息。最终实验结果表明：将ASTPNet附加在VGG16/19、ResNet34/50和ResNeXt上，识别精度有了显著提升；应用于自建的C...

Attention steered trapezoid pyramid fusion network for COVID-19 X-ray image recognition

The corona virus disease 2019 (COVID-19) is severely affects the development of society and economy,and threatens human health.In order to solve the problem that how to identify patients infected with the virus more accurately and quickly,convolutional neutral network (CNN) methods are used to identify COVID-19 chest X-ray images.However,due to the low recognition accuracy of CNN,it is difficult to accurately determine whether a patient is infected with COVID-19.In order to improve the recognition performance of the network for COVID-19 chest X-ray images,firstly,the attention steered trapezoid pyramid fusion network (ASTPNet) is proposed.The ASTPNet can be attached to different CNNs.The characteristics of deep and shallow networks in the model are effectively utilized.Secondly,the attention steered block (AS Block) is proposed to aggregate the weighted information more efficiently to emphasize effective semantic information in channels and spaces,and weaken ineffective interference information through channel and spatial attention.The results show that the accuracy is significantly improved after attaching the ASTPNet to VGG16/19,ResNet34/50 and ResNeXt.When applied to the self-built COVID-19 dataset,and compared with other CNN methods,ASTP-ResNet34 has the better performance.The accuracy reaches 98.40% (two classes) and 97.10% (three classes).It can accurately determine whether the infection of COVID-19.