基于智能影像技术的新冠肺炎图像识别方法研究详细信息

基于智能影像技术的新冠肺炎图像识别方法研究
Research on New Coronary Pneumonia Image Recognition Method Based on Intelligent Imaging Technology

公开范围：延期公开 (1年) 所属项目课题相关成果

编制时间： Wednesday, March 03, 2021 报告类型：最终报告

报告作者：

[1]纪东升(兰州理工大学)

中文摘要：

COVID-19是一种新型冠状病毒感染引起的肺炎，能够在多种途径下传播，而且能够长时间存活在冷链中。在这项研究中，利用胸部x光片对COVID-19肺炎患者和健康人进行鉴别，发展一种有效的诊断COVID-19疾病的方法。本文首先对所有胸部X光图像进行了分析和预处理后，利用迁移学习的知识，通过特征融合的方法提高诊断性能。实验使用了多个预训练的CNN模型，选用Xception，ResNet152，DenseNet201，VGG19和InceptionResNetV2卷积网络，通过特征融合的方法提高诊断性能，并和不同CNN对比分析。实验收集了Kaggle上的4099张胸部X射线图片，其中3278张用于训练，821张用于验证。在实验结果分析中，与其他模型相比，使用五个级联CNN模型验证准确率最高（即95.12%），在测试图片上能够达到99%的准确率。

英文摘要：

COVID-19 is a pneumonia caused by New Coronavirus infection, which can spread in many ways and can survive in cold chain for a long time. In this study, chest X-ray was used to identify patients with cowid-19 pneumonia and healthy people, and to develop an effective method for the diagnosis of cowid-19 disease. In this paper, after analyzing and preprocessing all the chest X-ray images, we use the knowledge of transfer learning to improve the diagnostic performance by feature fusion. Several pre trained CNN models are used in the experiment, and xception, resnet152, densenet201, vgg19 and inception resnetv2 convolution networks are selected to improve the diagnosis performance by feature fusion method, and compared with different CNN. 4099 chest X-ray images on kaggle were collected, of which 3278 were used for training and 821 for verification. In the analysis of experimental results, compared with other models, the five cascaded CNN models have the highest accuracy (95.12%), and can achieve 99% accuracy on the test images.

中文关键词：

COVID-19;迁移学习;特征融合;深度学习;深度卷积神经网络

英文关键词：

COVID-19;Transfer learning;Feature fusion;Deep learning; Deep convolution neural network;

全文页数：

17 ,图12,表3

馆藏号：

用户名		真实姓名
工作单位
付费下载此篇报告共0页共元。我是账户：，本人承诺将获取的科技报告仅用于个人学习、研究使用，保证不将科技报告全文传递到互联网上或用于任何商业目的。如违反相关规定，所造成的后果和损失将由本人承担。