Optimal Intelligent Embedded Edge Computing for AIoT in Smart Agriculture

This paper presents an IoT-based smart agricultural monitoring system that integrates machine learning-driven edge computing to optimize resource efficiency and support sustainable agriculture. Utilizing a heterogeneous LoRa and NB-IoT network, the system enables real-time analysis of environmental, meteorological, and soil data. To enhance intelligent edge computing, multiple machine learning models were evaluated, with Random Forest selected for its balance of accuracy and computational efficiency. The feasibility of embedding this model on various microcontroller platforms was assessed based on memory usage, computational speed, and classification accuracy. A case study optimized model parameters to balance hardware constraints and predictive performance. Experimental results demonstrate that while increased computational resources improve performance, the enhancements are nonlinear. Selecting an embedded platform requires a trade-off between accuracy, power consumption, cost, and resource availability. This study provides a structured approach for deploying machine learning on edge devices, offering insights into model optimization, resource constraints, and future intelligent edge computing applications.