Underground root tuber sensing (RTS) is important for monitoring crop phenotypic traits in crop breeding and other smart agriculture applications. This paper proposes a novel RTS framework with a radio frequency (RF) sensor network and deep learning models, demonstrating the “see-through soil” capability of RF sensor networks in underground RTS. We build upon an RF tomography network system and propose a novel data-driven RTS model, TD-RTS, that uses transformer and diffusion neural networks for imaging cross-sections of potato root tubers. Furthermore, we propose a biomass sensing model by combining the transformer network in TD-RTS with a multilayer perceptron (MLP) to estimate the biomass of underground tubers. To achieve accurate sensing, we use both the frequency and spatial diversities of the networked sensing system in RTS, and use fade-level to facilitate the selection of RF channels in order to reduce the data processing overhead. We perform extensive experiments, demonstrating the efficacy of the RTS framework.
Wang et al. (Mon,) studied this question.