Improved Model Based on GoogLeNet and Residual Neural Network ResNet

To improve the accuracy of image classification, a kind improved model is proposed. The shortcut added to GoogLeNet inception v1 and several other ways are given, they GRSN1_2, GRSN1_3, GRSN1_4. Among them, information input layer directly output each subsequent in form shortcut. new has advantages multi-size small convolution kernel same network reduce loss. Meanwhile, as number blocks increases, channels increased deepen extraction information. GRSN, GRSN1_4, GoogLeNet, ResNet models were compared on cifar10, cifar100, mnist datasets. experimental results show that proposed 3.07% data set 2.08% 17.69% 28.47% cifar100.