Research On COVID-19 Detection Technology Based On Deep Learning And Biogeography-Based Optimization

In recent years,with the deepening research and application of deep learning technology,it has been widely used in the detection of coronavirus disease 2019(COVID-19),and has achieved good detection results.However,these detection methods still need to be improved in two aspects.One is the difficulty of selecting hyperparameter values in convolutional neural networks(CNN);the other is the weak robustness and generalization ability of the model.In order to solve the above problems,this paper introduces biogeography-based optimization(BBO)to optimize the hyperparameter values in the convolutional model.The innovation of this paper is reflected in the following aspects.1.Aiming at the difficulty of optimizing the hyperparameters value of convolutional neural network,the BBO algorithm is introduced,and the convolutional neural network model architecture combining CNN and BBO is proposed.At the same time,based on the VGG-16 model,a new network model Expert VGG-16(EVGG)was designed and built to detect COVID-19.In order to prove the effectiveness and reliability of BBO for CNN hyperparameters value optimization,this paper selects three classic CNN models(Le Net-5,VGG-16,Res Net-18)based on EVGG for experimental verification.The experimental results demonstrate the effectiveness of the BBO optimization method for hyperparameter selection,and at the same time demonstrate that the BBO-optimized models have stronger robustness and better generalization ability than the original models.2.Aiming at the slow convergence speed of BBO,the Momentum Factor Biogeography-based Optimization(MF-BBO)method is proposed by introducing the momentum factor(MF).Different from BBO,in the migration phase of MF-BBO,after determining the immigration variable and the emigration variable,a numerical comparison operation is added,and the value calculation is standardized and constrained by MF,thereby enhancing the optimization ability of the algorithm.The experimental results show that MF-BBO has stronger convergence ability and better optimization effect than BBO.3.Aiming at the problem that MF-BBO has a high probability of falling into the local optimal solution and loses the ability of subsequent optimization,the Iteration Momentum Factor Biogeography-based Optimization(IMF-BBO)is proposed by introducing the iteration momentum factor(IMF).Different from MF-BBO,IMF-BBO divides the total number of iterations into multiple sub-iteration parts,and performs different migration operations on different sub-iteration parts,so as to improve the optimization ability while ensuring the time complexity of algorithms are same.Moreover,for the problem of long model training time,depthwise separable convolution(DSC)and dilated convolution(DC)are introduced to reduce the number of parameters of the models and reduce the training time of the models.The experimental results show that IMF-BBO has more balanced optimization ability and better convergence effect than MF-BBO,and the introduction of DSC and DC also effectively reduces the training time of the models.There are 44 figures,42 tables,and 118 references in this paper.