In this master’s thesis, we developed an autonomous, cost-effective, and energy-efficient system for non-contact soil moisture measurement on small‑scale agricultural fields. The measuring system combines a camera, illuminance and reference moisture sensors, and a multimodal multi‑branch neural network, which predicts soil moisture from digital images, illuminance data and categorical labels of location and weather conditions. We iteratively optimized and tested the system across multiple locations. The results demonstrate high accuracy, with the model’s robustness found to be highly dependent on the diversity of the training dataset.
|
