A Modular IoT-TinyML Architecture for Early Detection of Citrus Diseases

Abstract

Citrus diseases such as Huanglongbing (HLB), citrus canker, and Citrus Tristeza Virus (CTV) cause substantial economic losses, yet conventional detection methods remain costly and inaccessible to smallholder farmers in developing regions. This paper presents a modular IoT-TinyML architecture for early disease detection that integrates low-cost sensors with edge-based machine learning. The three-layer architecture (Perception, Processing, Application) enables real-time environmental monitoring and on-device visual classification using dual TinyML models for leaf and fruit analysis. A confidence-based decision fusion strategy balances sensitivity and specificity. Solar power and lightweight protocols ensure energy autonomy and remote operability without internet dependency. This edge-computing approach reduces detection latency and costs compared to cloud-based solutions, offering a scalable and practical precision agriculture tool for resource-constrained environments.