Research On Recognition Method Of Pneumonia X-Ray Image Based On CNN Algorithm

In the 2019 outbreak of New Coronavirus pneumonia,pneumonia entered the field of vision.The harm of pneumonia is great and the types are complex.It is difficult for doctors to judge the prevalence and types of pneumonia by lung X-rays,especially the high mortality and low differentiation of viral and bacterial pneumonia.At present,convolutional neural network algorithm is often used in the classification of lung X-ray images.Lung parenchyma segmentation is usually carried out before classification.UNet and VGG algorithms are important classification and segmentation methods,but UNET and VGG algorithms have some problems,such as gradient disappearance,too large model and too slow convergence speed.In view of these problems,this thesis improves on the basis of UNet and VGG-16 algorithms,This is of great significance to improve the recognition effect of pneumonia X-ray image.Firstly,the UNet algorithm is improved,and the depth separable convolution operation with residual edge is used to replace the original convolution operation,which can reduce the network size and alleviate the disappearance of gradient;The pool operation is changed into convolution operation to improve the segmentation accuracy of the model;In addition,the normalization function is added to the improved segmentation model to accelerate the convergence speed of the network.Secondly,the VGG-16 algorithm is improved.The depth separable convolution operation with serial expansion convolution is used to replace the serial convolution operation,which can reduce the amount of network parameters under the condition of ensuring the same receptive field;The global average pooling module is used to remove the full connection layer and form a full convolution network structure to reduce the network size;Similarly,the convergence speed of the network is accelerated by adding normalization function to the improved recognition model.This thesis uses Kaggle data set to test the improved algorithm.The segmentation accuracy and Io U value of the improved pneumonia X-ray image segmentation algorithm based on convolutional neural network are the highest in the comparison algorithm,which are 98.17% and 98.05% respectively;The parameter quantity of the improved pneumonia X-ray image recognition algorithm is only 53%of that of the small recognition network Mobile Net V1,but in the comparison algorithm,the recognition accuracy in the three states is the highest,91%,88% and88% respectively.Through the improvement of this thesis,the problems of too large algorithm model and slow convergence speed are effectively solved,which is helpful to assist radiologists to provide accurate diagnosis.