Human activity recognition based on modal decomposition and multiscale attention

WANG Hao; RONG Hailong; ZOU Ling

Chinese

您当前的位置：

首页 >

文章列表页 >

Human activity recognition based on modal decomposition and multiscale attention

Research & Approach | 更新时间：2026-06-24

- Human activity recognition based on modal decomposition and multiscale attention
- Vol. 45, Issue 6, Pages: 74-79(2026)
- 作者机构：
  
  1. 常州大学机械与轨道交通学院,江苏,常州,213000
  2. 常州大学微电子与控制工程学院,江苏,常州,213000
- 作者简介：
- 基金信息：
- DOI：
  CLC： TP183;TP212
- Published：2026
- 稿件说明：
移动端阅览
WANG Hao, RONG Hailong, ZOU Ling. Human activity recognition based on modal decomposition and multiscale attention[J]. 2026, 45(6): 74-79.
DOI：

WANG Hao, RONG Hailong, ZOU Ling. Human activity recognition based on modal decomposition and multiscale attention[J]. 2026, 45(6): 74-79. DOI：

摘要

为了解决惯性传感器原始信号在人体行为识别（HAR）任务中特征提取困难的问题，提出了一种基于变分模态分解（VMD）和多尺度通道注意力(MCA)机制优化的RepVGG人体行为识别网络模型。该模型首先使用VMD将惯性传感器原始信号依据中心频率的不同分解为若干本征模态函数（IMF）。接着，利用通道注意力机制，进一步优化这些IMF，突出有助于提高识别精度的特征。最后，使用具有结构重参数化的高效卷积神经网络RepVGG对优化后的IMF进行特征提取和活动分类。实验结果表明：该模型在UCI-HAR数据集中的准确率达到了97.42%。该模型将为不同场景下的活动识别提供可靠的技术支持。

Abstract

In order to solve the difficulty of feature extraction from raw signals of inertial sensors in human activity recognition(HAR) task

a RepVGG human activity recognition network model optimized by variational mode decomposition(VMD)and multiscale channel attention(MCA) mechanism is proposed.Firstly

the model uses VMD to decompose the raw signals of inertial sensors into several intrinsic mode functions(IMFs)according to different center frequencies.Next

these IMFs are further optimized using the channel attention mechanism to highlight features that contribute to enhancing the recognition precision.Finally

feature extraction and activity classification are performed on the optimized IMFs using RepVGG

an efficient convolutional neural network with structural re-parameterization.Experimental results show that the model achieves an accuracy of 97.42% on the UCI-HAR dataset.This model will provide reliable technical support for activity recognition in different scenarios.

关键词

Keywords

references

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Research on hemiplegic gait phase recognition based on inertial sensing

Signal processing method for wear particle sensor based on DA-VMD-SG

Gesture recognition method based on multi-stream feature fusion

Abnormal gait evaluation model based on inertial sensor data fusion

Arrester segmentation algorithm based on transfer learning and improved ResNet

Related Author

GUO Shijie

CAO Feng

DENG Huanyu

TIAN Haochen

WANG Jiaxin

WANG Zhenglin

LIU Yan

CHENG Chong

Related Institution

School of Mechanical Engineering,Hebei University of Technology

School of Artificial Intelligence and Data Science,Hebei University of Technology

College of Civil Aviation,Nanjing University of Aeronautics and Astronautics

College of Electronic Information and Automation,Civil Aviation University of China

Department of Orthopedics,Changzhou Second People’s Hospital,Nanjing Medical University

⁰