Soil moisture classification using hyperspectral imaging and deep learning models on ground robot vehicles

Soil moisture monitoring is increasingly critical in precision agriculture due to growing concerns over food security and sustainable water use. While technologies like UAVs, satellite remote sensing, IoT, and wireless sensor networks have advanced single-depth soil moisture assessment, a major gap remains in leveraging on-the-go sensing systems, particularly unmanned ground vehicles (UGVs) with hyperspectral cameras. UGV-based hyperspectral systems offer real-time, high-resolution, and cost-effective monitoring potential across multiple soil depths. Although machine learning and deep learning (DL) models have shown strong performance in predicting soil properties like soil organic matter using hyperspectral imagery, their application to soil moisture classification remains limited. This study addresses that gap by developing one-dimensional convolutional neural network (1D-CNN) to classify real-time soil moisture at 6-24 inches (15.2-61.0 cm) depth using UGV-acquired hyperspectral data. Ground truth was determined using gravimetric water content, and hyperspectral data were captured at 6, 12, 18, and 24 inches (15.2, 30.5, 45.7, and 61.0 cm). The best results were achieved using full spectral datasets enhanced with spectral augmentation methods, standard normal variation, and first/second derivative Savitzky-Golay filtering, along with balanced class weighting. The 1D-CNN models achieved testing accuracies up to 0.67 and AUC values of 0.85, outperforming models using only the top spectral bands selected by random forest and partial least squares regression. Green visible (500-565 nm) and near-infrared (750-1000 nm) regions were consistently critical for classification. This research demonstrates the potential of integrating hyperspectral imaging and DL for timely and accurate soil moisture classification, supporting precision irrigation and offering valuable tools for researchers and policymakers in agricultural water management.