Research on Image Recognition Based on Improved ResNet

With the increasingly easy access to data, the increasing amount of data and the improvement of computer performance have promoted the development of deep learning. To improve the performance, the neural network has been continuously deepened, the proposal of residual network makes great contribution to the breakthrough of bottleneck of traditional network. Subsequently, many improvements have been made to the internal structure of the residual network, such as wide residual network, pre-activated residual network, etc. However, there are some problems such as network, redundancy, poor performance on small data sets, and high requirement on computer performance. For these problems, this paper use to replace the convolution in the residual network, reduce the number of model parameters, increases the number of shortcut connection, and improves the flow of network information. Moreover, experiments are conducted on cifar-10 and cifar-100. The experiment shows that the error rate drops to 4.89% and 23.63%, which proves that our modification gets better results.