Enhancing Agricultural Internet of Things Data Accuracy: Evaluating Kalman Filter-Based Sensor Denoising Techniques

Abstract

Purpose With the rapid adoption of the Internet of Things (IoT) in agriculture, monitoring ecological parameters has become increasingly important for farmers aiming to optimize crop yields. However, sensor noise often hampers the collection of accurate data. This pioneering paper addresses the gap by evaluating real-time sensor denoising techniques using Kalman filter models on a Raspberry Pi 4. Methods A sensor node was deployed in the Ruhango District, Rwanda, for 6 months to address this. We implemented three models: the Unscented Kalman Filter (UKF), the Unscented Kalman Filter combined with Fuzzy Logic (UKF_FL), and the Cubature Kalman Filter (CKF), to process real-time data. The environmental parameters monitored included non-methanic hydrocarbon concentration (NMHC), nitrogen oxide concentration (NOx), nitrogen dioxide concentration (NO2), and air temperature (T). The performance of these models was assessed using the coefficient of determination, root mean square error, mean absolute error, and computation time and memory usage.

Results Our findings indicate that the Unscented Kalman Filter and Fuzzy Logic (UKF_FL) outperformed the other models, achieving approximately 99% accuracy and a 25% reduction in computational memory usage compared to UKF.