Explainable AI for Agricultural Pest Diagnosis
Research Paper#Agricultural AI, Vision-Language Models, LLMs, Explainable AI🔬 Research|Analyzed: Jan 3, 2026 06:19•
Published: Dec 31, 2025 16:21
•1 min read
•ArXivAnalysis
This paper introduces a novel, training-free framework (CPJ) for agricultural pest diagnosis using large vision-language models and LLMs. The key innovation is the use of structured, interpretable image captions refined by an LLM-as-Judge module to improve VQA performance. The approach addresses the limitations of existing methods that rely on costly fine-tuning and struggle with domain shifts. The results demonstrate significant performance improvements on the CDDMBench dataset, highlighting the potential of CPJ for robust and explainable agricultural diagnosis.
Key Takeaways
- •Proposes a training-free framework (CPJ) for agricultural pest diagnosis.
- •Utilizes large vision-language models and LLMs for image captioning and refinement.
- •Achieves significant performance improvements on the CDDMBench dataset.
- •Provides transparent, evidence-based reasoning for diagnosis.
- •Offers a solution that avoids costly fine-tuning and addresses domain shift issues.
Reference / Citation
View Original"CPJ significantly improves performance: using GPT-5-mini captions, GPT-5-Nano achieves +22.7 pp in disease classification and +19.5 points in QA score over no-caption baselines."