Devanna, Rosa Pia (2025) Robotics and Artificial Intelligence for Precision Agriculture: Innovative vision-based solutions for in-field crop monitoring and assessment. [Tesi di dottorato]

Documento PDF Rosa_Pia_Devanna_37CICLO.pdf Visibile a [TBR] Amministratori dell'archivio Download (91MB) | Richiedi una copia

Abstract

Precision agriculture appears to be one of the most interesting strategies to address the issues posed by population growth and the need for ethical resource utilization today. Indeed, in this scenario, the efficiency and timing of agronomic decisions can make a significant difference, and yet they can also be enhanced by the benefits of both robotic equipment and artificial intelligence (AI) techniques. Nonetheless, despite the fact that such topics are appealing and represent the latest frontiers of human growth, there remain open questions regarding the accuracy and generalizability of sophisticated sensing technologies in real-world applications. The crop monitoring and evaluation methods described in this thesis aim to extend the literature knowledge about current AI- ased technologies by leveraging: autonomous robotic vehicles equipped with multi-sensor systems; 3D spatial mapping and monitoring, including infrared data and point cloud reconstruction; agronomic image analysis and classification employing AI algorithms developed with the latest machine and deep learning models as well as traditional computer vision approaches, depending on the context and goal; techniques for data integration that merge data from various sensors to earn precise estimations of vegetative indexes. The proposed solutions are developed and validated through seven case studies conducted in different agronomic contexts, highlighting the technical feasibility and practical impact in supporting farmers in field management. The findings indicate that robotics with embedded AI algorithms plays a critical role in achieving more effective and sustainable decisions in agriculture by enabling continuous and non-invasive crop monitoring. At the same time, the experimental activities reveal some severe challenges—like the need for solid data infrastructures and accommodation of the environmental variability to sufficiently scale the models—and ultimately provide important guidance for further research and development efforts.

Downloads

Actions (login required)

Modifica documento