Research On Deep Learning Method Of Dual-channel Convolution Neural Networks

Convolution neural networks,as a branch of artificial intelligence,have achieved great success in many fields.The depth of neural networks is of central importance for its success.However,deeper neural networks are more difficult to be trained,so this thesis will focus on how to design and train a deep convolution neural networks model.The main contents are as follows:Firstly,based on the characteristics of traditional convolution neural networks,Single-channel Convolutional Neural Networks(SCNN)model is proposed,and the implementation and training process of the model are introduced in detail.A Dropout algorithm is applied into SCNN model so as to improve the generalization ability and prevent over-fitting of SCNN.A Batch Normalization(BN)algorithm applied into SCNN model so as to accelerate the training speed of the model,and the activations of convolutional layers are normalized.Secondly,in order to solve the problem that the deep convolution neural networks are difficult to be trained due to vanishing gradients,a fast and efficient Dual-channel Convolution Neural Networks(DCNN)model is put forward,which consists of a straight channel and a convolution channel.Straight channel is responsible for ensuring the patency of the deep neural networks,convolution channel is responsible for learning of the deep neural networks.The training of deep networks is prone to exhibit instability.To this end,the convolution attenuation factor is proposed,which can scale down the convolution channel's responses.In order to ensure the consistency of data dimension on each channel,a new pooling method called dual-pool layer is proposed to down-sample on the same feature map.On three image recognition datasets CIFAR-10,CIFAR-100 and MNIST,the classification accuracies of DCNN can archive 94.53%,73.40% and 99.74% respectively,compared with existing deep convolution neural networks models,the depth,stability and accuracy of DCNN are significantly increased.Finally,the proposed DCNN model is applied to predict flight delays.The DCNN model can archive 92.08% for predicting flight delay in the real flight operation data provided by the US Bureau of Transportation Statistics,the results can provide services and guidance to the airport and passengers,it has a very important practical significance.