The modern agricultural sector faces growing challenges related to climate variability, the timely management of plant diseases, and complex compliance with territorial regulations. The objective of this thesis is to design, develop, and validate a functional prototype of a Decision Support System (DSS) aimed at optimizing the operational and decision-making processes of agricultural operators, with a particular emphasis on usability (user-centric design). The architecture of the proposed system integrates various Artificial Intelligence technologies to lower technical barriers to entry. The core of the user interface is constituted by a Large Language Model (LLM), which enables fluid interaction in natural language. The LLM acts as a central orchestrator, dynamically querying external APIs for meteorological data and analyzing document databases to extract updated agronomic directives. Concurrently, the diagnostic module utilizes Machine Learning (Computer Vision) algorithms to analyze user-provided images, identify phytopathological diseases, and suggest targeted intervention procedures. To ensure the technical and economic scalability of the system, the project adopts a hybrid architectural approach, leveraging pre-existing frameworks and pre-trained models. The expected results aim to demonstrate how information centralization and the use of conversational interfaces can streamline crop management, transforming heterogeneous data into immediately actionable insights for the end user.
Il settore agricolo moderno affronta sfide crescenti legate alla variabilità climatica, alla gestione tempestiva delle patologie vegetali e all'articolata conformità alle normative territoriali. L'obiettivo di questo lavoro di tesi è progettare, sviluppare e validare un prototipo funzionale di Decision Support System (DSS) pensato per ottimizzare i processi operativi e decisionali degli operatori agricoli, ponendo particolare enfasi sull'usabilità (user-centric design). L'architettura del sistema proposto integra diverse tecnologie di Intelligenza Artificiale per abbattere le barriere tecniche all'ingresso. Il fulcro dell'interfaccia utente è costituito da un Large Language Model (LLM), che consente un'interazione fluida in linguaggio naturale. L'LLM funge da orchestratore centrale, interrogando dinamicamente API esterne per i dati meteorologici e analizzando banche dati documentali per estrarre direttive agronomiche aggiornate. Parallelamente, il modulo diagnostico sfrutta algoritmi di Machine Learning (Computer Vision) per analizzare immagini fornite dall'utente, identificare patologie fitosanitarie e suggerire procedure di intervento mirate. Per garantire la scalabilità tecnica ed economica del sistema, il progetto adotta un approccio architetturale ibrido, sfruttando framework pre-esistenti e modelli pre-addestrati. I risultati attesi mirano a dimostrare come la centralizzazione delle informazioni e l'utilizzo di interfacce conversazionali possano efficientare la gestione delle colture, trasformando dati eterogenei in informazioni immediatamente azionabili per l'utente finale.
PROGETTAZIONE USER-CENTRIC DI UN DECISION SUPPORT SYSTEM PER L'AGRICOLTURA 4.0: INTEGRAZIONE DI LARGE LANGUAGE MODELS, MACHINE LEARNING E COMPUTER VISION.
SINISI, SERAFINO
2024/2025
Abstract
The modern agricultural sector faces growing challenges related to climate variability, the timely management of plant diseases, and complex compliance with territorial regulations. The objective of this thesis is to design, develop, and validate a functional prototype of a Decision Support System (DSS) aimed at optimizing the operational and decision-making processes of agricultural operators, with a particular emphasis on usability (user-centric design). The architecture of the proposed system integrates various Artificial Intelligence technologies to lower technical barriers to entry. The core of the user interface is constituted by a Large Language Model (LLM), which enables fluid interaction in natural language. The LLM acts as a central orchestrator, dynamically querying external APIs for meteorological data and analyzing document databases to extract updated agronomic directives. Concurrently, the diagnostic module utilizes Machine Learning (Computer Vision) algorithms to analyze user-provided images, identify phytopathological diseases, and suggest targeted intervention procedures. To ensure the technical and economic scalability of the system, the project adopts a hybrid architectural approach, leveraging pre-existing frameworks and pre-trained models. The expected results aim to demonstrate how information centralization and the use of conversational interfaces can streamline crop management, transforming heterogeneous data into immediately actionable insights for the end user.I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14253/82186