An edge AIoT framework for optimizing the placement of integrated soil sensors in agricultural wireless sensor networks

Abstract

Edge Artificial Intelligence of Things (AIoT) optimization of integrated soil sensor networks is essential for intelligent, data-driven precision agriculture. However, achieving adequate field coverage while minimizing deployment costs is challenging, especially with integrated soil sensors that measure multiple soil attributes simultaneously. Existing optimization techniques focus solely on single-sensor designs and require fragmented steps, including cleaning historical data, making them unsuitable for near-real-time edge AIoT systems and limiting the effectiveness of kriging-based geostatistical analysis. To address this gap, this study presents an edge AIoT framework for global optimization of integrated soil sensor networks within an agricultural wireless sensor network. The framework combines Cubature Kalman Filtering (CKF) for dynamic state estimation with Kriging (K), Artificial Neural Network (ANN), Genetic Algorithms (GA), and Particle Swarm Optimization (PSO) to enhance edge AIoT deployment in a Wireless Sensor Network (WSN). A WSN was deployed over an 8.5-hectare cassava field using a 25 × 25 m grid, resulting in 135 sampling points at a depth of 0.2 m. The CKF–Kriging framework integrated with Artificial Neural Networks achieved the highest spatial coverage (C ≥ 99%) while maintaining minimal sensor redundancy and low computational cost. The optimized deployment reduced the required number of sensors from five to three for temperature, moisture, and pH (40% reduction), and from five to four for phosphorus (20% reduction), outperforming the Genetic Algorithm and Particle Swarm Optimization approaches. CKF-K-ANN is a potential two-phase Deploy-Learn-Optimize global optimization approach that enables cost-effective, high-performance deployment of resource-constrained edge AIoT-integrated soil sensors for near-real-time agricultural systems.