融合注意力机制与DenseNet的胸部X光片肺炎检测算法  

Pneumonia detection algorithm for chest X⁃ray films by fusing attention mechanism and DenseNet

作　　者：盛承光 Sheng Chengguang(School of Foreign Languages,Shenzhen Institute of Information Technology,Shenzhen 518172,China)

机构地区：[1]深圳信息职业技术学院应用外语学院,深圳518172

出　　处：《现代计算机》2023年第13期65-68,共4页Modern Computer

摘　　要：针对肺炎胸部X光片自动检测准确率较低的问题,提出了一种融合注意力机制与DenseNet的胸部X光片肺炎检测算法。该算法以DenseNet121为基础框架,利用其强大的表征能力自动学习X光片的成像特征;同时,引入注意力机制,在通道和空间两个维度上序列化地产生注意力特征图,构建通道之间的相互依赖关系与获取空间特征位置信息,以提升网络的特征提取与学习能力,使其更能关注到图像中的具有辨识性的病变区域。在公开的肺炎X光片数据集上的实验结果表明,所提出算法的准确率、召回率、精确率和F1⁃score值分别为94.40%、95.09%、95.42%和95.23%,相对其他模型具有更高的识别精度。Aiming at the problem of low accuracy in the automatic detection of pneumonia using chest X‑ray images,a pneumonia detection algorithm is proposed by integrating the attention mechanism and DenseNet.A DenseNet121 is firstly used as the basic framework to automatically learn imaging features from chest X‑ray images by using its powerful representation ability.Then,the attention module is introduced,which can serially generate attention feature maps in the channel and space dimensions,and construct the interdependence relationships between channels and capture position information of spatial features,so as to improve the feature learning ability of the network,so that the network can pay more attention to discriminative lesion regions in the image.The experimental results on the public chest X‑ray dataset show that the accuracy,recall,precision,and F1-score value of the proposed method are 94.40%,95.09%,95.42%,and 95.23%,respectively,which have higher recognition performance than other models.

关 键 词：肺炎检测 注意力机制 DenseNet 胸部X光片 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] R816.4[自动化与计算机技术—计算机科学与技术]