A IoT-Based Smart Agriculture Disease Detection system

Abstract

K Rice cultivation faces significant challenges from foliar diseases, which can reduce yields by up to 40% if not detected early. Traditional disease monitoring relies heavily on manual field scouting, which is time-consuming, subjective, and often results in delayed intervention. This research presents an innovative IoT-enabled multimodal deep learning framework that integrates visual imagery with real-time environmental data for early-stage rice leaf disease detection and severity assessment. The proposed system employs a three-tier architecture comprising sensing, edge processing, and cloud analytics layers. Our multimodal fusion approach combines image features extracted through EfficientNet-B4 with contextual environmental parameters including temperature, humidity, and soil moisture levels. The fusion is mathematically formulated as F_total = αF_image + βF_env, where optimal weighting coefficients are learned through end-to-end training. We introduce a novel Disease Severity Index (DSI) that quantifies infection progression from 0-100%, enabling precision intervention strategies. Explainable AI techniques including Grad-CAM++ and LIME provide interpretable visual explanations, ensuring farmer trust and adoption. Experimental validation on a dataset of 12,847 images across eight disease classes demonstrates superior performance, achieving 97.3% accuracy, 96.8% precision, 97.1% recall, and an F1-score of 96.9%. The multimodal approach outperforms unimodal image-only baselines by 4.7% in accuracy and shows robust generalization across different geographic regions. Ablation studies confirm the significant contribution of environmental context, improving early-stage detection accuracy by 6.2%. The system demonstrates practical deployment feasibility with edge inference latency of 143ms per sample. This work advances the field by bridging the gap between laboratory research and field-deployable precision agriculture solutions, offering smallholder farmers an accessible, interpretable, and effective disease management tool.