Scholarlink

<p>This paper aims to propose a WiFi-based human activity recognition method addressing privacy protection, recognition accuracy, and cost. The technique achieves non-contact recognition by extracting human activity features from Channel State Information (CSI). Adaptive Kalman Filtering combined with the Savitzky–Golay filter enhances the CSI data quality and suppresses environmental interference. A variance-based sliding window threshold method is used to segment activity periods accurately. Third-order cumulant features are extracted via high-order spectral analysis, and the Mutual Information Feature Selection (MIFS) algorithm selects robust features to improve environmental adaptability and behaviour discrimination. Finally, a behaviour recognition model is constructed by integrating a Gaussian Mixture Model–Hidden Markov Model (GMM–HMM) with a Deep Q-Network (DQN) to classify segmented activities. Experimental results demonstrate that the proposed method achieves superior accuracy and robustness compared to baseline approaches.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Privacy-preserving activity recognition method based on WiFi signals: adaptive CSI enhancement and deep reinforcement learning

  • 期刊论文

  • 发表日期：

    2025年11月26日

查看全文
  • Jiai He,
  • Xue Zhao,
  • Xianqi Li

摘要

This paper aims to propose a WiFi-based human activity recognition method addressing privacy protection, recognition accuracy, and cost. The technique achieves non-contact recognition by extracting human activity features from Channel State Information (CSI). Adaptive Kalman Filtering combined with the Savitzky–Golay filter enhances the CSI data quality and suppresses environmental interference. A variance-based sliding window threshold method is used to segment activity periods accurately. Third-order cumulant features are extracted via high-order spectral analysis, and the Mutual Information Feature Selection (MIFS) algorithm selects robust features to improve environmental adaptability and behaviour discrimination. Finally, a behaviour recognition model is constructed by integrating a Gaussian Mixture Model–Hidden Markov Model (GMM–HMM) with a Deep Q-Network (DQN) to classify segmented activities. Experimental results demonstrate that the proposed method achieves superior accuracy and robustness compared to baseline approaches.