The existing deep residual network, as a variant of a convolutional neural network, is used in various fields due to its sound performance. Although the depth residual network obtains higher accuracy by increasing the depth of the neural network, there are still other ways to improve the accuracy at the same depth. In this study, three optimization methods are used to optimize the depth residual network. (1) Dimension filling by mapping through a convolutional network. (2) Building a residual module based on the SELU activation function. (3) Learning rate decays with the number of iterations. Testing the improved network on the dataset Fashion-MNIST, the experimental results show that the proposed network model is superior to the traditional deep residual network in accuracy.