Multimodal Large Language Model for Intelligent Diagnosis and Management of Crop Nutrient Deficiencies and Environmental Stresses

Crop nutrient deficiencies and environmental stresses pose major challenges to modern agriculture, often leading to reduced yields and inefficient management. This study presents AgriHealth-LLM, a multimodal intelligent system designed for crop health diagnosis and management. The model combines a vision encoder, a modality alignment module, and a language model to analyze crop images and farmer queries, enabling precise identification of issues and generation of personalized recommendations. A domain-specific dataset, AgriDiagnose-MVD, is constructed to support model training, featuring diverse crops, annotated symptoms, and expert-curated question–answer pairs. Experimental evaluation shows that AgriHealth-LLM surpasses existing approaches in both diagnostic accuracy and quality of management suggestions, demonstrating its potential to support sustainable and data-driven agricultural practices.