IoT-Based Rice Leaf Disease Detection and Automated Treatment System Using Edge AI

Tóm tắt

This paper presents an IoT-based system for automated rice leaf disease detection and treatment using edge AI. The system employs a three-tier architecture: (1) distributed sensors and actuators on Arduino and ESP32 microcontrollers, (2) NVIDIA Jetson Nano for real-time AI inference and control logic, and (3) Flask web interface with Firebase cloud synchronization. The TensorRT-optimized deep learning model achieves 35.76 ms inference latency (27.96 FPS) enabling real-time video processing. Six environmental sensors (temperature, humidity, soil moisture, light, wind, pH) provide context for intelligent treatment decisions. Disease-specific pesticide delivery operates through dual pumps controlled via ESP32 with HTTP communication. The system demonstrates 96% correct pump selection and maintains operational autonomy during network outages with automatic cloud synchronization upon  restoration. Laboratory validation over 72 hours confirmed reliable environmental monitoring, sub-second treatment response, and stable inference performance. The edge-first architecture eliminates cloud dependencies while enabling remote monitoring, addressing critical gaps in precision agriculture for resource-constrained rural environments. Future work requires field trials, mobile platform integration, and economic viability studies for smallholder farmer adoption.