Research On Classification Method Of Chest X-ray Image Based On Neural Network And Feature Fusion

Pneumonia is usually caused by pathogenic microbial infection,physical and chemical irritation,impaired immune function,allergy and drug factors.Among these factors,bacterial and viral pneumonia are the most common types,posing a serious health threat to children and the elderly.Since December 2019,the novel coronavirus(COVID-19)caused by the SARSCo V-2 virus strain has spread widely around the world and has also become a serious global public health concern.As a painless,non-invasive,suitable for a wide range of people,and relatively inexpensive test,pulmonary X-ray is one of the most commonly used radiological screening and diagnosis methods for lung diseases.Previous studies have shown that X-ray lung images have specific differences in imaging findings between common pneumonia and COVID-19.Therefore,the classification method of lung X-ray images was studied based on neural network and feature fusion.The main research work is divided into the following parts:(1)Chest X-ray classification based on convolutional and long-and long-duration memory networks.Using Xception neural network combined with long term short term memory(LSTM)deep learning technology,this method can realize automatic diagnosis of patients with pneumonia in X-ray images.The traditional cross entropy loss function cannot balance the unmatched class distribution in the training set samples.In order to solve this problem,we can learn from Pearson’s idea of feature selection,fuse the correlation of two loss functions together,and optimize the problem.(2)X-ray image classification based on Dense Net and VGG feature fusion.This method proposes a chest X-ray image classification based on the fusion of Dense Net and VGG features.Based on the model feature fusion,two attention mechanisms are added to extract depth features.Resnet network was used to segment effective image information and remove useless background information to form a new data set.The proposed combination model has achieved good results in this experiment.Therefore,the application of deep learning and feature fusion technology in chest X-ray image classification will also become an auxiliary tool for clinicians and radiologists.(3)Pneumonia X-ray image classification based on truncated lightweight convolutional neural network.This method uses a truncated lightweight convolutional network and a lightweight network model based on the initial Inception-Resnet-V2 network.This work freezes a pre-trained convolutional neural network,the Inception-Resnet-V2 model.The loss function combining Jaccard and traditional cross entropy is selected to calculate the difference between the real value and the predicted value and identify the similarity and difference between the sample sets,which can help the model to judge the error information effectively.By combining the model with the optimized loss function,a faster training time is obtained after experimental evaluation,while a rich selection of relevant features is maintained.By improving the running time efficiency of the model,the high precision of pneumonia image is obtained.