融合残差与CIFAR10-quick结构的改进胶囊网络肺炎识别研究

打开文本图片集

中图分类号：TP391.4；TP181 文献标识码：A 文章编号：1006-8228（2025）08-34-07

Abstract：Pneumoniaisacommondiseaseoftherespiratorysystem.Inrecentyears，despitetheappicationofpowerful antibactrialdrugs，thefatalityrateofpeumoniainsomespecificpopulationshasnotbeenfurtherreducedTorealizethetimely diagnosisofpneumonia，thispaperproposesanoveldeeplearningmethodwithanimprovedcapsulenetworkforpneumonia recognitionbasedonchestX-rays.Theproposed methodusespartoftheCIFARi0-Quick network toextractfeaturesand inputs themintothemaincapsulelayerinvectorform.Thissolvestheproblemof insuficientinvariantfeatureextractionfromthe originalcapsulenetworkandefectivelyimprovesthemodel'sresistancetogradientvanishingthroughskipconnectionsacross residualmodules.TwocomparativeexperimentsontheChestXRay20l7datasetdemonstratethattheproposedmethodoutperforms convolutionalneuralnetworksinresistingoverfitingandhandingsmal-sampledatasets.Italsoachieveshigherclasification acuracycomparedtothreeothercapsulenetworkvariantsInthetestingphase，theprecision，specifityrecall，andF-scoreof the proposed method are 0.966，0.985，0.731，and 0.832，respectively，indicating itssuperiority.

KeyWords：CapsuleNetwork；ResidualBlock；PneumoniaRecognition;DeepLearning;Convolutional Neural Network

0引言

肺炎（Pneumonia）是一种呼吸系统常见病，一般指终末气道、肺泡和肺间质的炎症，因此，实现肺炎的早期诊断，对患者的治疗和痊愈就显得尤为重要。（剩余7947字）