AI-Based Plant Disease Recognition System: A CNN Approach with Dual Deployment via Web and Telegram

Abstract

Early detection of plant diseases is crucial for improving crop yield and reducing economic losses. This paper presents a CNN-based plant disease recognition system using a hybrid dataset of over 87,000 images across 38 classes. An EfficientNet-based transfer learning model is employed to achieve high accuracy while maintaining computational efficiency. The system incorporates advanced features such as context-aware analysis using environmental data, economic loss estimation, and explainable AI through Grad-CAM. To ensure accessibility, it is deployed via a Streamlit web application and a Telegram chatbot, along with a multilingual voice interface. Experimental results show improved performance, achieving 96–97% accuracy, making the system suitable for real-world agricultural applications.

Plant phenotyping relevance

植物画像から病害状態を認識するCNN手法の開発・評価が中心であり、植物の病徴・病害状態を直接推定するため、植物フェノタイピング手法として適格です。

Code and data availability

The paper's only public asset is the PlantVillage dataset, which is a cited prior-work resource (Mohanty et al.), not a paper-specific deposit. The authors' in-field images, trained models, and code are not publicly deposited; in-field images are available only on request.

No evidence-backed public reproduction asset is currently recorded.