Application of Graph Pooling in Skeleton Based on Behavior Recognition
LI Zhuo
WU Chunlei
Abstract：Graph convolutional network(GCN)has been achieved excellent performance in skeleton-based action recognition tasks.However,not all the nodes are closely related to the action,and these irrelevant nodes must hinder the accuracy of recogni-tion.Therefore,graph pooling is applied to skeleton-based action recognition.Specifically,the feature is extracted by one graph convolutional layer,then the self-attention graph pooling is employed to remove irrelevant nodes,and finally the graph convolution-al layer is used for feature extraction and classification results are obtained.In this way,the network pays more attention to the nodes related to the action,while ignoring the influence of the irrelevant node information,and the recognition accuracy is corre-spondingly improved.The effectiveness of the method is verified on two large public datasets,NTU RGB-D and Kinetics skeleton.
Keywords：skeleton-based action recognitiongraph convolutional networkgraph poolingattention mechanism
Publication Date：2023-11-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：6( 2557-2562 )