Interpretable Multi-modal Plant Disease Detection Using CNN Fusion and LLM-Based Explanations with Meteorological Insights

Document Type

Article

Source of Publication

Lecture Notes on Data Engineering and Communications Technologies

Publication Date

5-1-2026

Abstract

Early detection of crop diseases can minimize losses for farmers and enhance production. Traditional approaches are mainly based on CNN methods that learn visual symptoms from plant leaf images. However, visual symptoms lack contextualized prediction and explanation power. To address these limitations, we propose a multimodal disease classification and explanation framework. To enhance classification accuracy, we combine visual embeddings extracted from DenseNet121 with meteorological features. To ensure farmer-friendly interpretability, a multi-modal disease explanation method is designed based on decision tree to extract meteorological causal rules, grad-cam to extract visual cues and LLM to generate textual explanations. The experiments demonstrated that the multi-modal predictions outperform single-modal baselines by achieving high accuracy (i.e., 0.987) and provide, furthermore, contextualized farmer-friendly explanations.

ISSN

2367-4512

Publisher

Springer Nature Switzerland

Volume

298

First Page

292

Last Page

303

Disciplines

Computer Sciences

Keywords

classification, multi-modal fusion, Plant disease, XAI

Scopus ID

105046722341

Indexed in Scopus

yes

Open Access

no

Share

COinS