A low-cost AI-IoT integrated smart farming system with IoT sensor network simulation for Smallholder farms
Abstract
This study shows how to build, test, and put into use a low-cost, modular AI-IoT smart farming system that is perfect for smallholder farmers. The system uses Python-based sensor network simulation to imitate getting real-time data arising from sensors that measure soil moisture, temperature, humidity, light, pH, and the concentrations of nitrogen, phosphorus, and potassium in the soil. This cuts down on the requirement for costly hardware during development and training. Using edge computing on a Raspberry Pi with SQLite stored locally ensures that analytics occur in real-time, data is kept safe, and everything works well even when the internet is down. We created AI models, such as Random Forest classifiers and regressors, to check the health of plants, forecast how much they would produce, and find the best way to water them. We used both synthetic and real-world data for this. The system had 99.2% uptime, 98.6% data success rate, almost flawless AI accuracy, and low-latency inference (<200 ms) atop a 60-day field experiment. Economic analysis of scalable designs showed a quick return on investment (ROI of 2,079%–4,777%), and yearly benefits per hectare that are much higher than the original investment. This used to be due to yield increases of 15–25%, water savings of 20–30%, and labor cuts of around 30%. The system's modular, offline-capable design immediately solves problems with cost, technological complexity, and intermittent connection, which are common within rural areas. The results show that it is possible to make modern digital agricultural technologies available to everyone, giving smallholder farmers useful, data-driven information that will help them be more productive in the long term.