Fusing Perceptual Vision Experts with Multimodal Large Language Models for Explainable Plant Disease Diagnosis: From Benchmark Imagery to Real-World Robotic Field Validation

2026-08-23

AI Summary

"This study addresses the challenge of enhancing the accuracy of field plant disease diagnosis by proposing a Hybrid Hierarchical Multi-Agent Framework (H²MAF). This framework integrates the decision outputs from EfficientNet-B3 and ConvNeXt-Tiny visual models, leveraging multimodal large language models such as Gemma 4 E4B and Qwen3.5 4B for semantic arbitration. The method generates explainable diagnoses, risk levels, and treatment urgency assessments based on JSON-formatted evidence. Experimental results demonstrate that H²MAF significantly improves diagnostic accuracy, particularly in conflict data subsets, with an increase from 63.9% to 68.5% on the PlantDoc dataset, and a 7.6% improvement in CNN prediction conflicts. Furthermore, the framework achieves high accuracy rates of 99.3% and 98.9% on real-world field datasets from Cornell University."

Abstract

Accurate field plant disease diagnosis requires reliable fusion of uncertain and conflicting perceptual evidence. We present the Hybrid Hierarchical Multi-Agent Framework (H$^{2}$MAF), combining decision-level fusion of EfficientNet-B3 and ConvNeXt-Tiny with semantic arbitration by open-weight multimodal large language models (MLLMs), Gemma 4 E4B and Qwen3.5 4B, using structured JSON evidence to generate explainable diagnoses, risk levels, treatment urgency, and financial exposure. (H$^{2}$MAF) is evaluated on 14,364 images (1,370 test images) across PlantDoc (2,922 images, 27 classes) and two non-public, continuously captured Cornell robot-acquired field datasets: Stage 2 (20 GB; 4,215 images) and Stage 4 (40 GB; 7,227 images), covering Early Blight, Late Blight, and Septoria Leaf Spot under uncontrolled field conditions. On PlantDoc, Gemma improves accuracy from 63.9% to 68.5%, achieving +7.6 points on the 41.7% CNN-conflict subset. Cornell accuracies reach 99.3% and 98.9%, with only 1.7-4.1% disagreement, demonstrating conflict-dependent MLLM utility. The critical-risk error of gemma is 0.14-0.5 points, whereas Qwen overflags by 3.5-14.4 points. These results establish MLLM arbitration as a promising, yet calibration-dependent, approach for explainable agricultural AI and robotic field decision support. Github Link: https://github.com/Applied-AI-Research-Lab/Explainable-AI-Plant-Disease-Detection

Problem

• plant disease diagnosis
• perceptual evidence
• multimodal large language models
• explainable diagnoses
• field conditions

Innovation

• Hybrid Hierarchical Multi-Agent Framework (H²MAF)
• multimodal large language models
• explainable diagnoses
• structured JSON evidence
• robotic field validation