International Conference on Computer and Knowledge Engineering

Home / 13th International Conference on Computer and Knowledge Engineering

Efficient Sub-Carrier Relationship Extraction for Human Activity Recognition via EEGNet in Wireless Sensing

Authors :

Siavash Zaravashan¹ Sadegh ArefiZadeh² Sajjad Torabi³

1- Part AI Research Center 2- Part AI Research Center 3- Part AI Research Center

Keywords :

WiFi-sensing،Channel State Information (CSI)،Signal Processing،Signal Processing،Human Activity Recognition

Abstract :

Human activity Recognition (HAR) through wireless sensing is a complex task due to the dynamic nature of human movements. Deep learning models have proven to be highly effective in this area. However, DL methods come with a high computational cost for training and deployment. This has led to the development of more efficient models that can deliver real-time performance while maintaining high accuracy. In Channel State Information (CSI) data, a noteworthy correlation exists among the subcarriers of an antenna system. This makes it necessary to capture and analyze these relationships among subcarriers. To address these challenges, we utilize EEGNet, a Convolutional Neural Network (CNN) architecture typically used in Brain-Computer Interface (BCI) tasks. This method is capable of extracting both temporal and spatial features from EEG data. Given the structural similarity between CSI and EEG data, we apply EEGNet to CSI data. The advantage of EEGNet is its low parameter count, which results in quicker inference times and improved transfer learning and fine-tuning procedures, especially when the amount of training data is limited. In our study, we showcase EEGNet's superior performance compared to Recurrent Neural Network (RNN) approaches. Additionally, despite having notably fewer parameters, EEGNet can hold its own when pitted against renowned CNN techniques such as ResNet.