ALDMAgents: An Integrated Multi-Agent Collaborative Framework for Apple Leaf Disease Detection and Treatment Plan Generation Based on LLMs

Apple leaf diseases pose a significant threat to apple yield and quality. While existing computer vision algorithms can accurately recognize disease types, subsequent treatment relies heavily on the subjective diagnosis and experience of human experts, resulting in an end-to-end disconnection. To address this gap, this paper proposes ALDMAgents, an integrated multi-agent collaborative framework based on Large Language Models (LLMs). Guided by Standard Operating Procedures (SOP), ALDMAgents structures a pipeline from image detection to treatment generation using four distinct expert roles. Furthermore, we introduce the Tree of Thoughts (ToT) reasoning method combined with a novel multi-expert LLM collaboration mechanism to overcome the single-path and single-perspective limitations of traditional agents. Experimental results demonstrate that ALDMAgents significantly outperforms mainstream Input-Output (IO), Chain-of-Thought (CoT) and Tree-of-Thoughts (ToT) methods, achieving an average treatment effectiveness of up to 89.43% and superior BLEU scores for complex diseases like Black rot, Scab, and Rust, providing an intelligent and comprehensive solution for modern orchard management.