SMART TECHNOLOGIES FOR AGRI-FOOD AUTOMATION: DEVELOPMENT OF A SMART CART FOR PRECISION AGROCHEMICAL APPLICATION AND HARVEST ASSISTANCE

The rapid advancement of precision agriculture has fostered the adoption of optical sensing, data-driven modelling, and robotic platforms to improve crop management, quality assessment, and resource efficiency. However, the practical deployment of these technologies in real agricultural environments remains constrained by sensor reliability, limited robustness of predictive models, computational restrictions of embedded systems, and insufficient integration between perception, decision-making, and actuation. This doctoral thesis addresses these challenges through the development of an integrated framework combining multispectral and Vis–NIR sensing, advanced statistical and machine learning models, and a modular robotic platform designed to support agrochemical distribution and harvest-related operations. The research focuses on short-range crop monitoring, early stress and disease detection, non-destructive fruit quality assessment, and system-level robotic integration within a Robot Operating System 2 (ROS 2) architecture. A sensor-less approach for pixel-level alignment of multispectral images is proposed and validated, enabling reliable vegetation index computation under proximal sensing conditions. Physiological monitoring and disease detection are investigated through the integration of biochemical measurements, spectral indices, and chemometric modelling, highlighting the strengths and limitations of Vis–NIR spectroscopy for quantitative and classification tasks. Harvest assistance is addressed via portable spectroscopy, demonstrating accurate non-destructive prediction of fruit quality attributes and effective classification of ripening stages and postharvest defects using reduced spectral information. These sensing and modelling strategies are embedded within a distributed robotic architecture, resulting in the development of a modular smart agricultural cart. The platform integrates heterogeneous sensors, embedded computing, and navigation capabilities, and is validated through proof-of-concept autonomous navigation experiments. The work further identifies hardware limitations of low-power computing platforms and outlines a scalable pathway toward next-generation embedded solutions for real-time, closed-loop agricultural applications. The proposed approach combines sensing, modelling, and robotics to support more efficient, sustainable, and technology-enabled agricultural practices within precision farming contexts.