Remote monitoring system based on cross-hole GPR and deep learning

This paper presents a high-voltage pulse-based radar design and a deep-learning method for soil moisture estimation. This study aims to develop a pulse-based radar system that can detect changes in soil moisture content using a cross-hole approach. The system consists of a pulse generator based on a Marx generator with an LC filter, three transmitting antennas placed in a 12 m deep borehole, and three receiving antennas located in a separate borehole 100 m away from the transmitter. The receiver used a high-frequency data acquisition card to acquire signals at 3 Giga Bytes per second. At the same time, the borehole antennas were designed to operate in a wide frequency band to ensure signal propagation throughout the soil. For volumetric soil moisture estimation using time-sampled signals, this paper proposes a deep regression convolutional network that models changes in wave propagation between the transmitted and received signals. The training dataset comprises soil moisture measurements taken at three points between the transmitter and receiver and 25 meters apart to provide ground truth data. Radar data and soil moisture measurements were collected for 73 days between the two boreholes. In an additional experiment, water was poured into several specially prepared boreholes between transmitter and receiver antennas to acquire additional data for training, validation, and testing of convolutional neural networks. Experimental results showed that the proposed system could detect changes in volumetric soil moisture using Tx and Rx antennas.