Research On Image Classification Based On Capsnet

The huge potential of capsule network in the field of computer vision has attracted widespread attention.The capsule network encodes the possibility of entity existence through the length of the vector,and maps the instantiation parameters to the vector direction.It can not only represent the image with the strength of the feature response,but also the information such as the direction and location of the image feature.The update between the two layers of the capsule network uses a dynamic routing algorithm to avoid losing the exact location information of the image.The specific research contents of this article are as follows:(1)The image classification problem based on the capsule network is studied.The capsule network is used to perform image classification experiments on the MNIST dataset and the CIFAR10 dataset,respectively,and a convolutional neural network of the same size is designed as a baseline for comparison.By comparing and analyzing the number of parameters,running speed,and classification accuracy,it is concluded that the capsule network has more advantages in image classification than the convolutional neural network.However,the classification effect of capsule network on complex images needs to be improved,which has certain guiding significance in image classification.(2)Aiming at the classification of capsule networks on complex images,an improved capsule network model is proposed.The network consists of a feature extraction network,a dynamic routing network and a reconstruction network.The feature extraction network uses a variety of convolution kernels of different sizes for convolution,which can make the obtained combined feature map have different receptive fields,increase diversity,capture more image features,and enhance the network’s feature expression ability.Finally,experiments on the CIFAR10 dataset prove that the improved model has a recognition accuracy rate that is 7.5% higher than the original model,and the number of parameters is lower than the original model,which provides a reference direction for the research and development of the capsule network.