Research And Implementation Of COVID-19 Detection Based On Transfer Learning

In the past two years,Novel coronavirus pneumonia has caused a world pandemic.The virus was first reported in Wuhan at the end of 2019 and then spread worldwide.Nowadays,there are even more variants of the virus,which have appeared in many countries one after another,posing a major threat to human health worldwide.Therefore,the research related to the detection of 2019 novel coronavirus has far-reaching and imminent practical significance.It can not only make judgments efficiently and quickly,but also provide guidance information for medical staff,and at the same time,it also guarantees the safety of individuals in a sense.For this reason,it is necessary to propose a detection system to quickly assist in diagnosis,so as to help control the further spread of COVID-19.In this paper,the transfer learning method is used to study the chest X-Ray image detection task.The main innovations are as follows:(1)Collect novel coronavirus pneumonia,other pneumonia and normal chest X-ray images to construct COVID-19 X-Ray data set,and perform denoising processing on the unique text noise of the training set samples to further expand and standardize the data set.In order to solve the problem of limited data set samples,it is difficult to directly train the model and improve the training speed,the knowledge learned by the neural network model pre-trained on the ImageNet data set is transferred to the task of COVID-19 detection,a research method to obtain more feature information is proposed,a ResNet50-cc network model based on transfer learning is constructed.The experimental results show that the models trained on the AlexNet,ResNet,DenseNet and ResNet50-cc networks based on transfer learning have better expressive power.(2)A detection model based on quantum neural network is proposed,and the quantum neural network based on transfer learning is applied to the detection task of novel coronavirus pneumonia.Combining the classic neural network model based on transfer learning with the advantages of quantum information,a new quantum convolutional layer is constructed.This method uses the structure of the classical network model in the feature extraction part.For the classifier,the number of features required by the quantum convolution layer is obtained through the fully connected layer,and then through the quantum convolution layer,it encodes the classical data information into quantum information.Quantum calculations are performed through random quantum circuits,and then qubits are decoded into classical data,thereby adding a quantum convolutional layer on the basis of the classical network model to form a new quantum convolutional neural network based on transfer learning.A new fusion network model(TQFNet)is further proposed,which selectively merges the neural network model based on transfer learning and the quantum neural network model,and obtains the final detection result through voting,which improves the detection accuracy.Experimental results show that the detection accuracy of this method has been improved,and the final classification accuracy can reach 95.71%.