A hybrid framework combining microwave sensing and deep learning for real-time soil volumetric water content estimation.

This study introduces an integrated, non-destructive method for real-time soil moisture measurement using a 2.4 GHz microwave/RF sensing system combined with machine learning. It overcomes the limitations of traditional gravimetric techniques, which are destructive, energy-intensive, and have a 24-hour delay, by providing quick, on-site estimates of volumetric water content (VWC) for urgent geotechnical decisions. The system pairs a 2.4 GHz radar platform with edge computing via a Raspberry Pi 5 to gather and analyze reflected signal power, which correlates strongly with soil dielectric properties. A convolutional neural network (CNN) classifies soil texture (sand, clay, or mixed), allowing for the selection of optimal polynomial regression models for precise prediction. Validation with target water contents of 0.05, 0.15, 0.25, 0.35 and 0.45 L showed high agreement between measured and predicted values, with average accuracies around 99.8%, 99.7%, and 99.4% for sand, clay, and mixed soil, respectively. The system dramatically reduces measurement time to real-time, maintaining reliability compared to the standard gravimetric method, which usually takes 24 h. This approach offers a cost-effective, scalable solution for continuous soil monitoring in civil engineering and agriculture, improving the assessment of foundational stability and enabling proactive risk management.