[1]李志新,商樊淇,郇 战,等.基于混合特征图卷积神经网络的人体行为识别方法[J].郑州大学学报(工学版),2024,45(04):46-52.[doi:10.13705/ j.issn.1671-6833.2024.04.002]
　LI Zhixin,SHANG Fanqi,HUAN Zhan,et al.Human Activity Recognition Based on Hybrid Feature Graph Convolutional Neural Network[J].Journal of Zhengzhou University (Engineering Science),2024,45(04):46-52.[doi:10.13705/ j.issn.1671-6833.2024.04.002]
点击复制

基于混合特征图卷积神经网络的人体行为识别方法()
分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
45
期数:
2024年04期
页码:
46-52
栏目:
出版日期:
2024-06-16

文章信息/Info

Title:
Human Activity Recognition Based on Hybrid Feature Graph Convolutional Neural Network
文章编号:
1671-6833(2024)04-0046-07
作者:
李志新1;  商樊淇1;  郇 战1;  陈 瑛1;  梁久祯2
1.常州大学 微电子与控制工程学院,江苏 常州 213000;2.常州大学 计算机与人工智能学院,江苏 常州 213000
Author(s):
LI Zhixin1;  SHANG Fanqi1;  HUAN Zhan1;  CHEN Ying1;  LIANG Jiuzhen2
1.School of Microelectronics and Control Engineering, Changzhou University, Changzhou 213000, China; 2.School of Computer Sci ence and Artificial Intelligence, Changzhou University, Changzhou 213000, China
关键词:
图卷积神经网络;  可穿戴设备;  人体行为识别;  时空图;  特征提取
Keywords:
graph convolutional neural networks;  wearable device;  human activity recognition;  spatio-temporal graph;  feature extraction
分类号:
TP391
DOI:
10.13705/ j.issn.1671-6833.2024.04.002
文献标志码:
A
摘要:
基于可穿戴传感器的人体行为识别方法不能很好地处理时间序列数据采样点之间的结构信息,也忽略了 数据样本之间的潜在联系。针对这一问题,提出了混合时频特征和结构特征的图卷积神经网络模型进行人体动作 识别。首先,通过小波包变换获取原始信号的时频特征,进一步构建时空图提取信号的结构特征以挖掘采样点间 的动态特性,并在结构特征中加入距离约束,弱化时空图中远距离邻居对中心节点的影响。其次,考虑到结构特征 提取时受时空图拓扑关系影响较大,选择样本的时频特征构造图卷积神经网络的输入拓扑,混合时频特征和结构 特征作为网络输入特征。最后,输入特征沿着输入拓扑结构传播,得到最终分类结果。为了评估所提模型的性能, 在WHARF和DataEgo数据集上进行了实验验证。实验结果表明:所提模型的F1分数相比已有的基于卷积神经网 络模型在WHARF和DataEgo上均有提升,WHARF数据集上F1最高提升19.58百分点,DataEgo数据集上F1最高 提升26.44百分点,证明所提出模型通过挖掘动态特性能够有效提高动作识别能力。
Abstract:
The existing methods for human activity recognition using wearable sensors could not capture the struc tural information between the sampling points of time series effectively and might ignore the potential connections between samples. To address this issue, a graph convolutional neural network model with hybrid time-frequency and structural characteristics was proposed for human activity recognition. Firstly, the time-frequency characteris tics of the original signal were obtained by wavelet-packet transform, and the spatio-temporal graph was further con structed to extract the structural characteristics to identify the dynamic characteristics between the sampling points. The distance constraint was added to the structural characteristics to weaken the influence of long-distance neighbors on the central node on the spatio-temporal graph. Considering that the extraction of structural characteristics was greatly affected by the topological relationship of the spatio-temporal graph. The time-frequency characteristics of the samples to construct the input topology of the graph convolutional neural network were selected, and the time frequency and structural characteristics were combined as the input features of the network. Finally, the input fea tures propagated along the input topology to obtain the final classification result. To evaluate the performance of the proposed model, experiments were conducted on the WHARF and DataEgo datasets. Results in terms of F1 scores indicated that the proposed model outperformed existing convolutional neural network-based methods, achieving a maximum improvement of 19.58 percentage points on the WHARF dataset and 26.44 percentage points on the Da taEgo dataset. It demonstrated that the proposed model could effectively enhance the capability of activity recogni tion by exploiting dynamic characteristics.

参考文献/References:

[1] HUAN Z, LV S Y, HOU Z J, et al. An evaluation strat egy for the symmetry and consistency of lower limb seg ments during upper limb loading[J]. IEEE Sensors Jour nal, 2021, 21(5): 6440-6449. 

[2] PHAM C, NGUYEN-THAI S, TRAN-QUANG H, et al. SensCapsNet: deep neural network for non-obtrusive sens ing based human activity recognition[J]. IEEE Access, 2020, 8: 86934-86946. 
[3] ZHU R, XIAO Z L, LI Y, et al. Efficient human activity recognition solving the confusing activities via deep en semble learning[J]. IEEE Access, 2019, 7: 75490-75499. 
[4] CHEN Y Q, XUE Y. A deep learning approach to human activity recognition based on single accelerometer[C]∥ 2015 IEEE International Conference on Systems, Man, and Cybernetics. Piscataway: IEEE, 2015: 1488-1492.
[5] IGNATOV A. Real-time human activity recognition from accelerometer data using convolutional neural networks [J]. Applied Soft Computing, 2018, 62: 915-922. 
[6] WANG L K, LIU R Y. Human activity recognition based on wearable sensor using hierarchical deep LSTM net works[J]. Circuits, Systems, and Signal Processing, 2020, 39(2): 837-856. 
[7] SU T T, SUN H Z, MA C M, et al. HDL: hierarchical deep learning model based human activity recognition using smartphone sensors[C]∥2019 International Joint Conference on Neural Networks (IJCNN). Piscataway: IEEE, 2019: 1-8. 
[8] ZHOU J, CUI G Q, HU S D, et al. Graph neural net works: a review of methods and applications[J]. AI Open, 2020, 1: 57-81. 
[9] AHMAD N, CHOW S H C, LEUNG H F. Beyond the gates of Euclidean space: temporal-discrimination-fusions and attention-based graph neural network for human activ ity recognition[EB/OL]. (2022-06-10)[2023-11-02]. https:∥arxiv.org/abs/2206.04855. 
[10] DONNER R V, NORBERT M, YONG Z, et al. Nonlin ear time series analysis by means of complex networks [J]. Scientia Sinica Physica, Mechanica & Astronomi ca, 2020, 50(1): 010509. 
[11] ZHU S L, GAN L. Specific emitter identification based on horizontal visibility graph[C]∥2017 3rd IEEE Inter national Conference on Computer and Communications(ICCC). Piscataway: IEEE, 2017: 1328-1332. 
[12] LI C Y, MO L F, YAN R Q. Fault diagnosis of rolling bearing based on WHVG and GCN[J]. IEEE Transac tions on Instrumentation and Measurement, 2021, 70: 3519811. 
[13]高金峰, 秦瑜瑞, 殷红德. 基于小波包变换和支持向 量机的故障选线方法[J]. 郑州大学学报(工学版), 2020, 41(1): 63-69. 
GAO J F, QIN Y R, YIN H D. Fault line selection based on wavelet packet transform and support vector machine [J]. Journal of Zhengzhou University (Engineering Sci ence), 2020, 41(1): 63-69. 
[14] LOTFOLLAHI-YAGHIN M A, KOOHDARAGH M A. Examining the function of wavelet packet transform (WPT) and continues wavelet transform (CWT) in rec ognizing the crack specification[J]. KSCE Journal of Civil Engineering, 2011, 15(3): 497-506. 
[15] BRUNO B, MASTROGIOVANNI F, SGORBISSA A. Wearable inertial sensors: applications, challenges, and public test benches[J]. IEEE Robotics & Automation Magazine, 2015, 22(3): 116-124. 
[16] POSSAS R, CACERES S P, RAMOS F. Egocentric ac tivity recognition on a budget[C]∥2018 IEEE/CVF Con ference on Computer Vision and Pattern Recognition. Pis cataway: IEEE, 2018: 5967-5976. 
[17]郭毅博, 孟文化, 范一鸣, 等. 基于可穿戴传感器数 据的人体行为识别数据特征提取方法[J]. 计算机辅 助设计与图形学学报, 2021, 33(8): 1246-1253. 
GUO Y B, MENG W H, FAN Y M, et al. Wearable sen sor data based human behavior recognition: a method of data feature extraction[J]. Journal of Computer-Aided De sign & Computer Graphics, 2021, 33(8): 1246-1253. 
[18] JORDAO A, TORRES L A B, SCHWARTZ W R. Novel approaches to human activity recognition based on accel erometer data[J]. Signal, Image and Video Processing, 2018, 12(7): 1387-1394. 
[19] HUANG Y, YANG X S, GAO J Y, et al. Knowledge driven egocentric multimodal activity recognition[J]. ACM Transactions on Multimedia Computing, Communi cations, and Applications, 16(4): 133.

更新日期/Last Update: 2024-06-14