Optical sensors applied on vineyard monitoring

This doctoral thesis investigates the use of optical imaging systems (RGB, multispectral, and hyperspectral) for non-destructive vineyard monitoring, with particular emphasis on practical and cost-effective field applicability. While most studies in the literature rely on laboratory-based hyperspectral imaging (HSI) systems that are expensive and difficult to deploy under real outdoor conditions, this work focuses on the development and validation of more accessible multispectral solutions capable of operating directly on agricultural machines in the field. A custom-built multispectral imaging system, mounted on a tractor, was designed and integrated with a dedicated Matlab-based acquisition software. Laboratory experiments established a robust radiometric calibration procedure based on a fixed white reference, ensuring signal reliability under variable illumination. Field trials conducted in the Cadriano vineyard demonstrated that multispectral data, when properly calibrated and processed, can predict key grape quality indicators, such as soluble solids content, pH, and titratable acidity, with accuracy comparable to hyperspectral systems, yet at a fraction of their cost and complexity. The final part of the research extended this concept toward the next frontier: spectral super-resolution, or the reconstruction of hyperspectral information from RGB images. A deep residual neural network (HSCNN-R) was trained on vineyard datasets to explore how common RGB cameras could emulate the spectral richness of hyperspectral imaging. This approach points toward a new generation of lightweight, low-cost, and easily deployable sensing systems for precision viticulture. Overall, the thesis advances both the methodological understanding and the technological feasibility of optical sensing in the field, demonstrating that practical, economically sustainable systems can achieve performance levels once attainable only with laboratory-grade hyperspectral instruments.