Classification Model Of Pneumonia Image Based On Convolution Neural Network

COVID-19 is a new type of acute respiratory infectious disease with strong infectivity and variability.Its emergence has posed a serious threat to human life safety.Nowadays,COVID-19 is still spreading worldwide,all walks of life are seriously affected,and human life is facing a severe test.In the process of fighting against the epidemic of COVID-19,the diagnosis and prevention of COVID-19 is particularly significant.At present,the main detection method is nucleic acid detection,which has problems such as too long detection time,insufficient medical staff,insufficient detection reagents and so on.As an important supplement,pulmonary CT imaging examination has increased the workload of doctors and requires high professional knowledge.Therefore,people urgently need new methods and technologies for its auxiliary diagnosis.Based on the above situation,in order to reduce the pressure of medical workers and improve the diagnostic efficiency and accuracy of COVID-19,this article combines big data technology and convolutional neural network technology to establish a classification statistical model of lung CT images to assist in the detection and diagnosis of COVID-19.The work of this article is as follows: First,the improved Inception Res Net v2 model is used to classify the lung CT images,and the batch normalization and attention mechanism are added to the original model to speed up the convergence and improve the accuracy of the model classification.Using 2019 n Co VR data set for training,the model realizes the three classifications of COVID-19 CT images,ordinary pneumonia CT images and normal lung CT images,and compares with other three classic network models to verify the effectiveness and superiority of the model in this article.Secondly,when the data set is too small,use data to enhance and expand the data set to improve the generalization ability and robustness of the model.Finally,in order to improve the readability of lung CT images,so that doctors can more effectively observe and diagnose the lung lesions,RA-UNet model is used to segment the lesion area image and mark the lesion area.RA-UNet model integrates residual module and attention mechanism on the original U-Net model,which enhances the ability of image feature extraction and improves the effect of lung lesion region segmentation.Compared with other three models,RA-UNet model is optimal in similarity coefficient,accuracy rate and recall rate.