融合注意力机制的通道拓扑细化改进的图卷积网络  

Improved Graph Convolutional Network with Channel Topology Refinement Based on Attention

作　　者：李昊璇[1] 李旭涛 LI Haoxuan;LI Xutao(College of Physics and Electronic Engineering,Shanxi University,Taiyuan 030006,China)

机构地区：[1]山西大学物理电子工程学院,山西太原030006

出　　处：《测试技术学报》2023年第2期120-126,共7页Journal of Test and Measurement Technology

摘　　要：针对目前骨架动作识别网络数据处理的处理方式、不同动作肢体的动态表述以及时间特征提取的不足,本文提出AA-SGN网络模型,将原始的关键点特征转换成骨骼特征,将动态的非拓扑网络,改进为动态拓扑细化的图卷积网络,在充分利用语义信息的基础上,在时间尺度上融合注意力机制,关联空间和时间信息,使网络充分利用骨骼位置信息和时间流的运动信息,在NTU60 RGB+D的CS和CV设定上提高了1%和0.4%,在NTU120 RGB+D的CSub和CSet分别提高了5.7%和4.6%,证明本文提出的AA-GCN能有效识别骨架动作。Aiming at the deficiencies of data processing methods,dynamic representation of different limbs and time feature extraction of skeleton action recognition network at present.AA-SGN network model is proposed in this paper,the dynamic non-topological network is improved to graph convolutional network with dynamic topology refinement.Transforming the original human key-point features into skeletal features.On the basis of making full use of semantic information,the attention mechanism is integrated to implicitly model the weight of bone position and motion features,so that the network makes full use of bone position information and motion information of time flow.Improving CS and CV Settings by 1%and 0.4%in NTU60 RGB+D,CSub and CSet by 5.7%and 4.6%in NTU120 RGB+D,respectively.It is proved that the AA-GCN proposed in this paper can recognize skeleton actions effectively.

关 键 词：图卷积网络 语义信息 拓扑结构 人体骨架 注意力机制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]