Indoor CSI Identity Recognition Based on Residual Networks

Identity recognition has been extensively studied over the years, with traditional methods like camera and iris recognition becoming less favored due to their vulnerability to ambient light and privacy concerns. Recently, there has been a shift towards using CSI (channel state information) from wireless devices, such as commercial wireless cards, for identity recognition. This approach has applications in various fields, including smart homes and virtual reality. However, to achieve high identification accuracy, it is crucial to mitigate random noise from indoor environments. This paper proposes using dwt (discrete wavelet transform) to reduce noise in CSI data. Additionally, to extract robust high-level features from CSI data and ensure precise identity recognition, it is essential to deepen the training network. This often leads to challenges like gradient vanishing and explosion. To address these issues, this paper employs the ResNet18 (Residual Network) for feature extraction and identification of pre-processed CSI data, effectively preventing non-convergence and enhancing identification accuracy.