传感技术学报2025，Vol.38Issue(4)：695-704,10.DOI:10.3969/j.issn.1004-1699.2025.04.017
基于图卷积的注意力聚焦时空融合人体活动识别研究
Research on Human Activity Recognition Based on Graph Convolution Attention Focus Temporal-Spatial Feature Fusion
摘要
Abstract
Accurately identifying human activity data can offer significant assistance in various domains such as motion analysis and medi-cal rehabilitation training.Given the limitations of existing human activity recognition models in achieving high accuracy for human activity data with non-Euclidean spatial features,a novel approach to human activity recognition feature extraction,named GCN-AL is introduced,which combines graph convolution,graph attention mechanism(GAT),and long short-term memory network(LSTM).Based on the GCN-AL framework,the human activity recognition model is established as GCT-net.Comparative simulation experiments conducted on the pub-licly available DaLiAc dataset,the results demonstrate significant improvements in the overall accuracy,average precision,and average re-call of the GCT-net model.Specifically,these improvements are 2.0％,2.4％,and 2.4％respectively when compared with GAN model based on graph convolution and graph attention mechanism.Furthermore,enhancements of 2.3％,2.5％,and 3.1％are observed respectively when compared with graph convolution-based GCN model.Moreover,when compared to classification models in recent refer-ences,the GCT-net model also demonstrates advancements in terms of overall accuracy.关键词
可穿戴惯性传感器/人体活动识别/GCT-net模型/图卷积/图注意力机制/长短时记忆网络Key words
wearable inertial sensors/human activity recognition/GCT-net model/graph convolution/graph attention mechanism/long short-term memory network分类
计算机与自动化引用本文复制引用
刘艳,赵明,马萌,曹清清,刘芳,聂凯..基于图卷积的注意力聚焦时空融合人体活动识别研究[J].传感技术学报,2025,38(4):695-704,10.基金项目
湖南省自然科学基金青年人才联合基金项目(13JJB001) （13JJB001）
