A Decentralized, Edge-AI Hydroponic System for Sustainable Urban Agriculture

Eldin MS and Shafie AA

DOI: 10.5281/zenodo.22816462

Global pressures from urbanization, shrinking arable land, and climate volatility necessitate efficient food production methods. This paper presents an Intelligent Hydroponic System (IHS) designed for autonomy and resource efficiency, integrating edge-based TinyML for closed-loop environmental control with a computervision layer for periodic plant-health diagnostics. The system employs an ESP32 microcontroller to host a lightweight Multi-Layer Perceptron (MLP) model that processes real-time sensor data—pH, electrical conductivity/total dissolved solids (EC/TDS), temperature, and humidity—to automate nutrient dosing and LED lighting. A secondary diagnostic module uses an NVIDIA Jetson Nano running a YOLO-based object detection model for twice-daily visual health assessments. In an eight-week Nutrient Film Technique (NFT) trial with lettuce (Lactuca sativa) and mustard greens (Brassica juncea), the IHS maintained target nutrient parameters with 96.3% accuracy and achieved 18–21% higher fresh biomass compared to a conventional, manually managed hydroponic system. The system also demonstrated a 21% reduction in water consumption and improved nutrient-use efficiency from 72% to 87%. Automated LED and pump scheduling further reduced total energy consumption by 8%. With a core smart component cost of approximately RM 244.25, the IHS offers a scalable, low-cost, and connectivity-resilient solution for precision agriculture in urban and resource-limited settings. These results establish a compelling case for decentralized, edge-native intelligence as a practical alternative to cloud-dependent smart farming platforms.