Capsule Network Based on Self-Attention Mechanism

Capsule network (CapsNet) is well-known as an evolution of classical convolution neural network (CNN), which is good at recognizing the postures, orientations, and textures, thus achieving promising results in areas such as image classification. However, the high computational complexity hinders them from building larger models. Fortunately, reducing the complexity of the routing of CapsNet through the channel attention mechanism has achieved significantly good results, whereas there is a possibility of further optimization of the storage and precision. Therefore, in this paper, we propose an improved version based on the self-attention mechanism on the basis of the traditional channel attention routing. Specifically, the global information of the capsule is captured by the self-attention routing, which leads to fewer misclassification errors. Experimental results on MNIST, affNIST, and Cifar10 show that the proposed methods can obtain better performances with fewer training parameters compared to traditional methods on average.