期刊名称：International Journal of Advances in Engineering and Management
电子版ISSN：2395-5252
出版年度：2021
卷号：3
期号：9
页码：1255-1262
DOI：10.35629/5252-030910761083
语种：English
出版社：IJAEM JOURNAL
摘要：At present, the application of convolutional neural network (CNN) in image classification has become one of the hotspots in the field of intelligent vision, and it has excellent performance in image processing. Deep learning often requires a lot of time and computing resources for training, which is also a major reason for bothering the development of deep learning algorithms. Although CNN has achieved good results in the field of image classification, its algorithm has a great impact on the effect and efficiency of image classification. Only the optimization algorithm that requires fewer computational resources and faster model convergence can fundamentally accelerate the learning speed and effect of the machine. Prevention and diagnosis of COVID-19 is a hot issue during this pandemic. Identifying COVID-19 from lung CT images can assist diagnosis of patients, which is of certain significance to improve the accuracy of COVID-19 diagnosis. In this paper, three classical CNN model are used to analyze the lung CT image data, which are divided into normal conditions, common pneumonia and COVID-19, so as to realize the recognition of COVID-19. In this scheme, image preprocessing is firstly carried out. Due to the small amount of open data set on the Internet, data expansion method is adopted to expand the data, and then training set and test set are divided into 4:1. Finally, AlexNet, VGG and ResNet CNN are used to extract the features of lung CT images respectively. The detection accuracy of three kinds of networks for pneumonia was obtained. The final test results show that VGG network has a better recognition effect on pneumonia detection. At the same time, VGG network can converge faster in training with calibrated data set. Tests on the data set show that VGG CNN can successfully classify CT images as COVID-19, normal and viral pneumonia with 94.47% accuracy. It can be proved that the model is feasible in pneumonia recognition and classification.
