PDD-AGENT: Multimodal Large Language Model-Driven AI Agent for Enhanced Plant Disease Diagnosis

Multimodal large language models (MLLMs) have made remarkable progress across various domains, excelling in tasks such as question answering, segmentation, and detection. However, their performance in multitask operations—particularly in specialized applications like plant disease diagnosis—remains limited. Existing large-scale plant disease models are often confined to narrow task scopes and lack expert-level diagnostic capabilities. To address these challenges, we propose a novel MLLM-driven Plant Disease AI Agent System designed to deliver accurate, expert-grade diagnostic services. Our system integrates four key modules: a data preprocessing module, a decision-making module, a multifunctional action module, and a result aggregation module. By fine-tuning the MLLM with large-scale plant disease datasets, the system acquires extensive prior knowledge to support precise decision-making. It intelligently selects and orchestrates specialized diagnostic tools, enabling multidimensional analysis and comprehensive result synthesis. Experimental results demonstrate the system’s effectiveness in overcoming current diagnostic limitations, offering a robust solution for plant pathology tasks with enhanced accuracy and adaptability, supporting the advancement of smart agriculture.