Related Experiment Video
Updated: Aug 15, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.0K
STC-NLSTMNet: An Improved Human Activity Recognition Method Using Convolutional Neural Network with NLSTM from WiFi
Md Shafiqul Islam1, Mir Kanon Ara Jannat1, Mohammad Nahid Hossain1
1Department of Electronics Engineering, Kwangwoon University, Seoul 01897, Republic of Korea.
Sensors (Basel, Switzerland)
|January 8, 2023
Summary
This study introduces a new deep learning model for human activity recognition (HAR) using WiFi signals. The novel STC-NLSTMNet model achieves high accuracy by integrating spatial and temporal features from Channel State Information (CSI).
Area of Science:
- Computer Science
- Artificial Intelligence
- Signal Processing
Background:
- Human Activity Recognition (HAR) is crucial for applications like healthcare and assisted living.
- WiFi Channel State Information (CSI) offers a privacy-preserving method for indoor HAR, eliminating the need for extra devices.
- Existing deep learning models struggle to simultaneously extract and integrate spatial and temporal features from CSI data, limiting accuracy.
Purpose of the Study:
- To propose a novel deep learning model, STC-NLSTMNet, for accurate human activity recognition using WiFi CSI.
- To effectively extract and integrate both spatial and temporal features from CSI signals simultaneously.
- To improve the accuracy of human activity recognition in indoor environments.
Main Methods:
- Developed a novel deep learning model named spatio-temporal convolution with nested long short-term memory (STC-NLSTMNet).
- Utilized depthwise separable convolution (DS-Conv) blocks for spatial feature extraction from CSI signals.
- Incorporated a feature attention module (FAM) to highlight essential features and nested long short-term memory (NLSTM) to capture temporal dependencies.
Main Results:
- The STC-NLSTMNet model achieved high accuracies of 98.20% on the Multi-environment dataset and 99.88% on the StanWiFi dataset.
- Demonstrated significant improvements in activity recognition accuracy compared to existing methods, with gains of 4% and 1.88% respectively.
- The model effectively extracts and integrates spatial and temporal features for robust HAR.
Conclusions:
- The proposed STC-NLSTMNet model offers a superior approach for human activity recognition using WiFi CSI.
- The integration of spatial and temporal feature extraction significantly enhances recognition accuracy.
- This model holds promise for advancing privacy-preserving indoor sensing applications.
