Using Wi-Fi signals to sense target activity is a promising study field, accounting for convenience concerns. However, it remains challenging to recognize target activity as a way of high-precision and stability due to the multi-path effect in Wi-Fi signals. In this paper, we propose a robust framework named WiPD, for accurate activity recognition based on Wi-Fi phase difference data. Firstly, a novel feature representation mechanism named visualized spectrum matrix (VSM) for Wi-Fi activity recognition is proposed. VSM is generated by performing a Short Time Fourier Transform operation on Wi-Fi phase difference data. Then, we design a neural network with the input type of VSM, namely, WiPD-Net, in which the activity features are extracted by both four convolutional neural network submodules and two WiPD-Block submodules. Experiment results show that our proposed WiPD-Net outperforms the existing baselines on our dataset and one public dataset. In particular, WiPD-Net can reach up to an accuracy of 99.80%, and achieve a good migration performance among five Wi-Fi environments.

考虑到便利性问题，使用 Wi-Fi 信号来感知目标活动是一个很有前途的研究领域。然而，由于 Wi-Fi 信号中的多路径效应，将目标活动识别为高精度和稳定性的方式仍然具有挑战性。在本文中，我们提出了一个名为 WiPD 的强大框架，用于基于 Wi-Fi 相位差数据进行准确的活动识别。首先，提出了一种新的特征表示机制，称为可视化频谱矩阵（VSM），用于 Wi-Fi 活动识别。VSM 是通过对 Wi-Fi 相位差数据执行短时傅里叶变换操作生成的。然后，我们设计了一个输入类型为 VSM 的神经网络，即 WiPD-Net，其中活动特征由四个卷积神经网络子模块和两个 WiPD-Block 子模块提取。实验结果表明，我们提出的 WiPD-Net 在我们的数据集和一个公共数据集上优于现有基线。特别是WiPD-Net可以达到99.80%的准确率，在五种Wi-Fi环境之间实现了良好的迁移性能。