Research On X-ray Image Classification Model Of COVID-19 Based On Tensor Network Machine Learning

COVID-19 is a global pandemic with the widest impact on human beings in the past 100 years,and human life safety and health are facing major threats.In the context of the global epidemic,limited medical personnel often need to diagnose a large number of patients.Diagnosing the chest X-ray images of these large numbers of suspected new crown patients also requires the assistance of doctors from multiple departments.However,during the epidemic,medical staff are busier than usual,and hospitals are even more crowded,which may lead to failure to isolate confirmed patients in the first place,or medical staff to make mistakes in judging the patient’s condition.Building an accurate and fast auxiliary system can effectively save medical resources and provide doctors with diagnostic reference opinions.At present,the processing of chest X-ray images mainly uses neural network models,but the use of deep learning models such as neural networks has problems such as high model complexity,long model training time,and strict requirements on hardware computing capabilities.In response to the above problems,this paper is devoted to mining the application of tensor network method in machine learning,and constructing a new coronary pneumonia X-ray image classification model and classification system based on tensor network machine learning.The main research work of this paper focuses on the following aspects:(1)Research the structure composition and model characteristics of the locally orderless tensor network model.Aiming at the traditional method for doctors to manually distinguish chest X-ray images,there are problems of low classification efficiency and high requirements for doctors’ basic quality when solving the classification task of X-ray images of new coronary pneumonia.Based on the above problems,this paper studies the structural composition and model characteristics of the locally orderless tensor network model.By using image flattening and setting up a multi-layer tensor network,the model can retain the spatial structure information of the image and learn the deeper features of the image during training.At the same time,the model also has the advantages of short training time and low model structure complexity.(2)Optimize the locally orderless tensor network model and build a hybrid locally orderless tensor network model.Locally orderless tensor network models have limitations in architectural scalability and general approximation capabilities.This paper will start from the direction of model structure and model performance,and optimize the locally orderless tensor network model.In terms of model structure,by introducing a convolutional neural network and adding an activation function inside the model,the problems of the limited scalability of the tensor network in machine learning and the lack of general approximation capabilities of the model are solved.In terms of model performance,the activation function with the best model classification effect(Se LU activation function)is selected through experiments,and the Dropout method is added to avoid model overfitting.Finally,an optimized model is constructed,that is,the hybrid locally orderless tensor network model.(3)The effect of the hybrid locally orderless tensor network model constructed in this paper is verified by using the public COVID-19 X-ray image dataset.Including the self-structure experiment of the hybrid locally orderless tensor network model,the comparative experiment of the hybrid locally orderless tensor network model and the traditional deep learning model,and the comparative experiment of the hybrid locally orderless tensor network and the locally orderless tensor network.The experimental results show that the number of hidden layers in the convolutional neural network part of the hybrid locally orderless tensor network model is 2(two convolutional layers and a pooling layer),and the model classification effect is the best when the activation function selects the Se LU function.The experimental index data of the hybrid locally orderless tensor network model and the common deep learning model are almost the same,but the model volume and model parameters are smaller than the comparison model.In addition,the hybrid locally orderless tensor network model has a certain improvement in the classification effect compared with the locally orderless tensor network model.Based on the above experimental results,the hybrid locally orderless tensor network model constructed in this paper is effective,and the given optimization techniques are effective.(4)Implement a X-ray image classification system for COVID-19 based on the Flask web framework,the underlying model adopts the hybrid locally orderless tensor network model constructed in this paper,and its main functions are tested and recorded.The new coronary pneumonia X-ray image classification web system built by frameworks such as Flask web can realize the functions of three classifications of new coronary pneumonia X-ray images and recording test image information.The system tested 500 images with an accuracy rate of 98.2%.The system can provide auxiliary diagnosis and treatment methods for primary clinicians to diagnose new coronary pneumonia X-ray images.