Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PECL: A Heterogeneous Parallel Multi-Domain Network for Radar-Based Human Activity Recognition

Jiuqi Yan, Chendong Xu, Dongyu Liu

TL;DR

PECL addresses the challenge of accurately recognizing human activities from radar signals by leveraging three complementary domain representations (Range-Time, Doppler-Time, Range-Doppler) processed in parallel through EfficientNet-B0 backbones augmented with CBAM attention. Temporal dynamics are captured via LSTM in the RT and DT branches, while the RD branch preserves spatial cues with a linear projection and max pooling, and the fused 384-dimensional features yield state-of-the-art accuracy with low computational cost. Comprehensive experiments on the Radar2 dataset show 96.16% accuracy and strong per-class performance, with ablations confirming the value of multi-domain fusion, CBAM attention, and selective temporal modeling. The work demonstrates significant potential for privacy-preserving, lighting-insensitive HAR in real-world settings and outlines clear paths toward robustness, generalization, and edge deployment.

Abstract

Radar systems are increasingly favored for medical applications because they provide non-intrusive monitoring with high privacy and robustness to lighting conditions. However, existing research typically relies on single-domain radar signals and overlooks the temporal dependencies inherent in human activity, which complicates the classification of similar actions. To address this issue, we designed the Parallel-EfficientNet-CBAM-LSTM (PECL) network to process data in three complementary domains: Range-Time, Doppler-Time, and Range-Doppler. PECL combines a channel-spatial attention module and temporal units to capture more features and dynamic dependencies during action sequences, improving both accuracy and robustness. The experimental results show that PECL achieves an accuracy of 96.16% on the same dataset, outperforming existing methods by at least 4.78%. PECL also performs best in distinguishing between easily confused actions. Despite its strong performance, PECL maintains moderate model complexity, with 23.42M parameters and 1324.82M FLOPs. Its parameter-efficient design further reduces computational cost.

PECL: A Heterogeneous Parallel Multi-Domain Network for Radar-Based Human Activity Recognition

TL;DR

PECL addresses the challenge of accurately recognizing human activities from radar signals by leveraging three complementary domain representations (Range-Time, Doppler-Time, Range-Doppler) processed in parallel through EfficientNet-B0 backbones augmented with CBAM attention. Temporal dynamics are captured via LSTM in the RT and DT branches, while the RD branch preserves spatial cues with a linear projection and max pooling, and the fused 384-dimensional features yield state-of-the-art accuracy with low computational cost. Comprehensive experiments on the Radar2 dataset show 96.16% accuracy and strong per-class performance, with ablations confirming the value of multi-domain fusion, CBAM attention, and selective temporal modeling. The work demonstrates significant potential for privacy-preserving, lighting-insensitive HAR in real-world settings and outlines clear paths toward robustness, generalization, and edge deployment.

Abstract

Radar systems are increasingly favored for medical applications because they provide non-intrusive monitoring with high privacy and robustness to lighting conditions. However, existing research typically relies on single-domain radar signals and overlooks the temporal dependencies inherent in human activity, which complicates the classification of similar actions. To address this issue, we designed the Parallel-EfficientNet-CBAM-LSTM (PECL) network to process data in three complementary domains: Range-Time, Doppler-Time, and Range-Doppler. PECL combines a channel-spatial attention module and temporal units to capture more features and dynamic dependencies during action sequences, improving both accuracy and robustness. The experimental results show that PECL achieves an accuracy of 96.16% on the same dataset, outperforming existing methods by at least 4.78%. PECL also performs best in distinguishing between easily confused actions. Despite its strong performance, PECL maintains moderate model complexity, with 23.42M parameters and 1324.82M FLOPs. Its parameter-efficient design further reduces computational cost.

Paper Structure

This paper contains 26 sections, 18 equations, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Ralated Works
  3. Methodology
  4. Echo Signal Model of Radar
  5. Multi-Domain Representation Learning
  6. Range-Time Domain
  7. Doppler-Time Domain
  8. Range-Doppler Domain
  9. Parallel-EfficientNet-CBAM-LSTM (PECL)
  10. Structure of PECL
  11. Attention Mechanism in MBConv Module
  12. Continuous Temporal Dynamics Extraction
  13. Implementation and Experimental details
  14. Data Source
  15. Data Augmentation
  16. ...and 11 more sections

Figures (6)

  • Figure 1: System flowchart of the radar HAR pipeline
  • Figure 2: Architecture of the proposed PECL framework with multi-domain processing branches, spatial attention (CBAM) modules, and temporal sequence (LSTM) module.
  • Figure 3: Enhanced MBConv module integrated with CBAM.
  • Figure 4: Heterogeneous design for multi-domain feature extraction: Featuring LSTM branches for Range-Time and Doppler-Time maps, and a linear layer for the Range-Doppler maps.
  • Figure 5: t-SNE visualization of feature embeddings across model variants: (a) raw data; (b–d) single-branch outputs (Range-Time, Doppler-Time, Range-Doppler); (e) Parallel-EfficientNet (PE); (f) PE with CBAM (PEC); (g) PEC with LSTM on all branches; (h) our final PECL model.
  • ...and 1 more figures