Hallucination Behavior in Multimodal LLMs Across Agricultural Image Interpretation and Generation Tasks

2026-05-26

AI Summary

This study addresses the prevalence of hallucinations in multimodal large language models (MLLMs) when applied to agricultural image understanding and generation—errors that often violate biological or agronomic principles and risk misleading decision-making. The work presents the first systematic evaluation of such hallucinatory behavior across both image-to-text interpretation and text-to-image generation tasks, introducing a fine-grained assessment framework tailored to agriculture that encompasses biological consistency, contextual accuracy, and agronomic plausibility. Leveraging mainstream models including Gemini, GPT-5, and LLaVA, the authors conduct multimodal evaluations under zero-shot and few-shot settings, complemented by human validation. Results reveal that zero-shot image interpretation achieves only 63%–75% accuracy, improving to 86.8% with few-shot learning, while 91% of generated images exhibit biological inconsistencies, underscoring significant reliability limitations and high hallucination risks in current MLLMs for agricultural applications.

Abstract

Large Language Models (LLMs) are being rapidly adopted in agricultural imaging applications, ranging from crop interpretation to synthetic field image generation. However, these models frequently exhibit hallucinations outputs that appear confident yet deviate from biological or environmental reality potentially leading to misinformed agronomic insights. This study investigates such hallucinations in two complementary directions: image-to-text, where LLMs interpret crop or field imagery to describe conditions such as biotic and abiotic stresses, and text-to-image, where models generate synthetic agricultural scenes based on descriptive prompts. We examine errors involving biological inconsistency, contextual inaccuracy, and agronomic implausibility, evaluating the outputs under domain-informed criteria across multiple imaging modalities. Our analysis identifies recurring hallucination patterns within both interpretive and generative tasks. In image interpretation, LLMs (e.g., Gemma, LLAVA, Qwen, and MiniCPM) achieved modest zero-shot accuracy (63 to 75 percent), whereas few-shot prompting improved performance up to 86.8 percent, exhibiting false detections and missed infections, indicating residual hallucination effects. In text-to-image tasks, advanced models such as GPT-5 and Gemini 2.5 Flash generate up to 91 percent biologically inconsistent scenes under relaxed prompt constraints, revealing fundamental weaknesses in current LLMs. This systematic assessment of visual reasoning and generation offers critical insights toward enhancing the reliability and trustworthiness of LLM-based agricultural imaging platforms.

Problem

• hallucination
• multimodal LLMs
• agricultural imaging
• image interpretation
• image generation

Innovation

• multimodal hallucination
• agricultural imaging
• biological consistency
• few-shot prompting
• synthetic image generation