Hybrid CNN and LLM for Image-Based Classification of Plant Leaf Diseases

Abstract

Plant diseases have continued to threaten agricultural productivity, while manual inspection methods have remained inefficient and prone to subjectivity. This study proposed and assessed a hybrid framework integrating a Convolutional Neural Network (CNN) with a Large Language Model (LLM) to perform image-based plant leaf disease classification accompanied by interpretable diagnostic explanations. EfficientNetV2-M was employed as the visual backbone and trained on 11 selected classes of apple, grape, and potato leaf images derived from the PlantVillage dataset. A structured data splitting strategy was applied to ensure reliable model validation and unbiased testing. The classification capability of the CNN component was examined through standard multi-class evaluation indicators, including class-wise predictive consistency and error distribution analysis. Experimental results indicated that the model delivered highly consistent predictions, reaching a peak test accuracy of 99.79%, reflecting its robustness in distinguishing visually similar disease patterns. To overcome the black-box limitation, prediction outputs were transformed into structured prompts and processed by GPT-4o to generate contextual explanations. The generated narratives systematically described observable symptoms, highlighted distinguishing characteristics, and suggested initial management actions. Overall, the proposed hybrid system demonstrated that combining high-performance visual recognition with language-based reasoning enhanced both diagnostic reliability and interpretability in digital agriculture applications.

Code and data availability

The paper uses the public PlantVillage color dataset (11 apple/grape/potato classes, 9,385 images) and explicitly points to it in the DATA AVAILABILITY statement as the replication package data. The authors also provide a public Streamlit demonstration of their hybrid CNN–LLM system. No author analysis code or trained-