科技创新与应用2025，Vol.15Issue(18)：9-16,8.DOI:10.19981/j.CN23-1581/G3.2025.18.003
基于数据增强的实时人体动作识别
摘要
Abstract
Human action recognition(HAR)based on Channel State Information(CSI)has significant application prospects in fields such as human-computer interaction,healthcare,and intrusion detection.Although current research has made substantial progress in recognizing various types of activities and improving recognition accuracy,challenges remain in the need for a large number of activity samples to train models,and in improving the real-time performance of the recognition process.To address these issues,a real-time human action recognition(HAR)system,CSI-FHAR,is designed based on data augmentation.By augmenting a small number of real samples to generate synthetic samples,the system reduces the demand for real samples during model training.Additionally,by segmenting complete activity samples,the recognition speed is increased,enhancing real-time performance.To increase the inter-class feature differences,CSI-FHAR adds temporal encoding to the activity samples,thereby improving the model's recognition accuracy.The convolutional block attention module(CBAM)is embedded in the convolutional neural network(CNN)to further enhance the network's ability to extract effective features from activity samples.Experimental results demonstrate the effectiveness of CSI-FHAR:with only five samples per activity class for 10 types of activities,the proposed model achieved recognition accuracies of 95.1%for gestures and 92.5%for full-body activities.关键词
人体动作识别/信道状态信息/数据增强/时间编码/注意力机制Key words
human action recognition(HAR)/channel state information/data enhancement/temporal coding/attention mechanism分类
计算机与自动化引用本文复制引用
俞秀文,张文哲,刘钝,何飞,王昱洁..基于数据增强的实时人体动作识别[J].科技创新与应用,2025,15(18):9-16,8.基金项目
国家自然科学基金资助(62271188) （62271188）
