Tangut characters have the characteristics of complex structure,numerous strokes,high similarity between categories,and uneven distribution of samples in each category.In view of these characteristics,we demonstrated the feasibility and superiority of applying the CapsNet network architecture to Tangut text recognition,and the A-CapsNet network was proposed.The superior performance of the AlexNet network to extract image information at a deep level was used to make up for the missing feature information received by the CapsNet high-level capsule.From the experimental analysis of the AlexNet module and the Capsule module,the experimental results show that the proposed A-CapsNet network can achieve 94％recognition rate of Tangut characters,which is 3 percentage points higher than the original capsule network.And A-CapsNet network is better than deep learning convolutional neural network,and it has good applicability.This research has made a certain contribution to the study of Tangut characters.
国家科技期刊平台
NSTL
万方数据
