Research on image classification based on Capsnet

Capsnet is a very innovative and innovative deep learning network. This paper uses a capsule network to classify and test the MNIST and CIFAR10 datasets, respectively. And in order to compare with the traditional convolutional neural network, two architectures are designed. Then the experiment is carried out and the results are compared. By comparing and analyzing the experimental results, Capsnet has obvious advantages over the CNN method regardless of the MNIST dataset or the CIFAR10 dataset. Also, Capsnet will greatly reduce the number of parameters, which has the certain guiding significance in the classification of the image.