[Paper Review] Using GAN to Enhance the Accuracy of Indoor Human Activity Recognition

Parisa Fard Moshiri, Hojjat Navidan+3 more|arXiv (Cornell University)|Apr 23, 2020

Context-Aware Activity Recognition Systems15 references21 citations

TL;DR

This paper proposes a semi-supervised deep learning framework that leverages a Generative Adversarial Network (GAN) to synthesize Channel State Information (CSI) data, enhancing the accuracy of indoor human activity recognition (HAR) using WiFi signals. By combining 50% real CSI data with GAN-generated data, and employing LSTM with PCA and STFT for feature extraction, the method improves classification accuracy by 3.4% and reduces log loss by 16% compared to using only real data.

ABSTRACT

Indoor human activity recognition (HAR) explores the correlation between human body movements and the reflected WiFi signals to classify different activities. By analyzing WiFi signal patterns, especially the dynamics of channel state information (CSI), different activities can be distinguished. Gathering CSI data is expensive both from the timing and equipment perspective. In this paper, we use synthetic data to reduce the need for real measured CSI. We present a semi-supervised learning method for CSI-based activity recognition systems in which long short-term memory (LSTM) is employed to learn features and recognize seven different actions. We apply principal component analysis (PCA) on CSI amplitude data, while short-time Fourier transform (STFT) extracts the features in the frequency domain. At first, we train the LSTM network with entirely raw CSI data, which takes much more processing time. To this end, we aim to generate data by using 50% of raw data in conjunction with a generative adversarial network (GAN). Our experimental results confirm that this model can increase classification accuracy by 3.4% and reduce the Log loss by almost 16% in the considered scenario.

Motivation & Objective

• To reduce the high cost and data collection burden of acquiring real CSI data for indoor human activity recognition.
• To improve classification accuracy in CSI-based HAR systems using limited real data.
• To explore the effectiveness of synthetic data via GANs in enhancing deep learning models for HAR.
• To develop a semi-supervised learning pipeline combining PCA, STFT, and LSTM for robust feature extraction and classification.

Proposed method

• A Generative Adversarial Network (GAN) is trained on 50% of real CSI amplitude data to generate synthetic CSI samples.
• Principal Component Analysis (PCA) is applied to reduce dimensionality of CSI amplitude data.
• Short-Time Fourier Transform (STFT) is used to extract frequency-domain features from CSI signals.
• An LSTM network is trained on a combination of real and GAN-generated CSI data for sequence modeling and activity classification.
• The model uses a semi-supervised learning strategy, where the GAN enhances data diversity while minimizing reliance on fully labeled real data.
• Classification performance is evaluated using accuracy and log loss metrics on a dataset of seven indoor human activities.

Experimental results

Research questions

• RQ1 Can GAN-generated synthetic CSI data effectively improve the performance of deep learning models in indoor human activity recognition?
• RQ2 To what extent does combining real and synthetic CSI data enhance classification accuracy compared to using only real data?
• RQ3 How does the integration of PCA and STFT with LSTM improve feature representation in CSI-based HAR?
• RQ4 Does the semi-supervised training strategy reduce the need for large-scale real CSI data collection?

Key findings

• The proposed method increases classification accuracy by 3.4% compared to training solely on real CSI data.
• The model reduces log loss by approximately 16% when using a combination of real and GAN-generated data.
• The use of GAN-generated data significantly improves model generalization, especially with limited real data.
• The integration of PCA and STFT enhances feature representation, supporting better LSTM performance.