Designing Multi-Task Convolutional Variational Autoencoder for Radio Tomographic Imaging

Radio tomographic imaging (RTI) emerges to model the environment and detect the passive targets by a wireless network. In this work, the received signal strength (RSS) measurements are collected from an uncalibrated network, and a multi-task convolutional variational autoencoder model is proposed to realize RTI. The presented model is trained end-to-end to denoise the RSS measurements, reconstruct the static tomographic images, estimate the parameters of the wireless network, and classify the measurement noise level, simultaneously. The multi-task variational learning strategy is able to improve the generalization of the model. Numerical experiments demonstrate the efficacy of our RTI method.