IoT-Integrated Multi-Modal Deep Learning for Accurate Plant Leaf Disease Detection

Abstract

Plant leaf diseases significantly affect crop yield and quality, posing a major challenge to global agriculture. Conventional detection methods based on manual inspection or image-only deep learning often overlook critical environmental factors influencing disease development. This paper proposes an Edge-Aware Multi-Modal Attention Network (EMAN) that integrates IoT sensor data with high-resolution leaf images to detect and assess disease severity accurately in real time. EMAN fuses visual leaf features with environmental measurements, including temperature, humidity, soil moisture, and light intensity. Dynamic Dilated Residual Blocks (DDRB) extract multi-scale image features to identify small lesions, discolorations, and subtle disease indicators. A dual attention mechanism emphasizes disease-prone regions (spatial attention) and relevant environmental factors (sensor attention), enhancing predictive performance. The architecture supports edge-cloud hybrid inference, enabling lightweight edge models to provide instant alerts to farmers, while cloud analytics refine predictions and monitor temporal disease trends. Experiments on combined Plant Village and on-field datasets demonstrate that EMAN outperforms conventional CNN-based methods in both disease classification and severity estimation, achieving robust multi-modal fusion and real-time edge inference. The proposed system provides a scalable solution for precision agriculture, promoting proactive crop management, minimizing economic losses, and supporting sustainable farming practices.

Plant phenotyping relevance

葉画像とIoTセンサーを統合し、植物病害の検出・重症度推定手法を開発・評価しており、植物状態の取得が研究の中心である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.