Skeleton-based Action Recognition Based on Feature Enhancement and Reparameterization
LI Doudou
LI Wanggen
XIA Yichun
GE Yinkui
WANG Zhige
Abstract：With the development of deep learning,the model of skeleton-based action recognition based on deep learning is becoming more and more complex,the network level is getting deeper and deeper,and the model weight is getting larger and larg-er,so there is a disadvantage of slow inference speed,so this paper proposes a skeleton-based action recognition(RepGCN)based on feature enhancement and reparameterization method.In order to improve the performance of the model,a feature enhancement method combining bone joint information,bone information,and bone angle information is proposed.Then,an adaptive map based on multi-scale is proposed for the input of graph convolution,which guides the graph convolution to extract deeper action features to improve the recognition performance of the model,and finally this paper proposes a way to train and test understanding coupling to simplify the complexity of the model and improve the inference speed of the model.It can reduce the amount of model parameters,facilitate the deployment of models and the application of mobile terminals.
Keywords：graph convolutional neural networkreparameterizationhuman skeletal action recognition
Publication Date：2025-04-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 1008-1014 )