Agri-CPJ: A Training-Free Explainable Framework for Agricultural Pest Diagnosis Using Caption-Prompt-Judge and LLM-as-a-Judge

arXiv:2604.23701v1 Announce Type: cross Abstract: Crop disease diagnosis from field photographs faces two recurring problems: models that score well on benchmarks frequently hallucinate species names, and when predictions are correct, the reasoning behind them is typically inaccessible to the practitioner. This paper describes Agri-CPJ (Caption-Prompt-Judge), a training-free few-shot framework in which a large vision-language model first generates a structured morphological caption, iteratively