【摘要】 为解决当下肺炎胸部X光自动检测准确率低的问题,提出一种基于胸部X光分类网络(chest X-r ayclassification network,Ch XNet)的肺炎检测方法,该方法能够自动对胸部X光进行检测诊断。利用预处理操作来丰富特征的多样性,将高效通道注意力模块(efficientchannelattentionmodule,ECA)以密集连接的方式加入密集连接网络中,增强有用信息的传递同时抑制无用信息的传递,使用基于Dropout方法构建的多层过渡分类结构,增强对相似特征的描述能力,减少冗余特征。通过大量实验,Ch XNet在三分类(正常、非新冠病毒肺炎、新冠肺炎)和四分类(正常、细菌性肺炎、普通病毒性肺炎、新冠肺炎)检测的最高准确率分别为99.845%和97.842%,表明该方法准确率高,检测速度快,可在肺炎的检测中作为辅助诊断方法。更多还原

【Abstract】 To solve the problem of low accuracy of chest X-ray automatic detection of pneumonia,a pneumonia detection method based on chest X-ray classification network(ChXNet) was proposed,which automatically detected and diagnosed chest X-ray.Preprocessing was used to enrich the diversity of features.The efficient channel attention module(ECA) was added to the dense connection network to enhance the transmission of useful information and inhibit the transmission of useless information.The multi-layer transition classification structure based on dropout method was used to enhance the description ability of similar features and reduce redundant features.Through a large number of experiments,the highest accuracy of ChXNet in the detection of three categories(normal,non-novel corona virus pneumonia,COVID-19) and four categories(normal,bacterial pneumonia,common viral pneumonia,COVID-19) is 99.845% and 97.842%,respectively.It shows that this method has high accuracy and detection speed,and can be used as an auxiliary diagnostic method in the detection of pneumonia.更多还原