Armando Egas Jose, Hanyurwimfura Damien, Gatera Omar, Kim Kwang Soo, Nduwumuremyi Athanase
African Center of Excellence in Internet of Things, University of Rwanda, Kigali, Rwanda.
Superior School of Rural Development, Eduardo Mondlane University, Vilankulo, Mozambique.
Sci Rep. 2025 Jul 1;15(1):20829. doi: 10.1038/s41598-025-05427-w.
Effective sensor denoising is crucial for accurate, real-time agricultural decision-support systems. This study explores the application of Unscented Kalman Filter (UKF) extensions on resource-constrained devices to improve sensor denoising and enhance the reliability of Internet of Things (IoT) based agricultural soil monitoring. The study was conducted in Ruhango district, Rwanda, utilizing a wireless sensor node equipped with a Raspberry Pi 5 (ARM v8) and an integrated seven-in-one soil sensor measuring temperature, humidity, electrical conductivity, pH, nitrogen, phosphorus, and potassium. The sensor was placed at a depth of 20 cm in ten cassava farms, collecting data every 30 min for eight months. Four real-time sensor denoising models were implemented: UKF, Cubature Kalman Filter (CKF), UKF with Artificial Neural Network (UKF_ANN), and UKF with Fuzzy Logic (UKF_FL). Models' performance was evaluated using boxplot, square root(R), mean absolute error (MAE), root mean square error (RMSE), computation memory (CM), and computation time (CT). Data analysis was performed using Python 3.12 on ARM v8. Results demonstrated that CKF outperformed the other models, reducing RMSE by up to 32% and lowering CM and CT by 75%. CKF and UKF_ANN maintained the integrity of the censored data while effectively removing Gaussian, uniform, and salt-and-pepper noise, making them suitable for IoT-based soil monitoring systems.
有效的传感器去噪对于准确的实时农业决策支持系统至关重要。本研究探讨了无迹卡尔曼滤波器（UKF）扩展在资源受限设备上的应用，以改善传感器去噪并提高基于物联网（IoT）的农业土壤监测的可靠性。该研究在卢旺达的鲁汉戈区进行，使用了一个配备树莓派5（ARM v8）的无线传感器节点和一个集成的七合一土壤传感器，用于测量温度、湿度、电导率、pH值、氮、磷和钾。该传感器放置在十个木薯农场20厘米深处，每30分钟收集一次数据，持续八个月。实施了四种实时传感器去噪模型：UKF、容积卡尔曼滤波器（CKF）、带人工神经网络的UKF（UKF_ANN）和带模糊逻辑的UKF（UKF_FL）。使用箱线图、平方根（R）、平均绝对误差（MAE）、均方根误差（RMSE）、计算内存（CM）和计算时间（CT）对模型性能进行评估。使用Python 3.12在ARM v8上进行数据分析。结果表明，CKF的性能优于其他模型，RMSE降低了32%，CM和CT降低了75%。CKF和UKF_ANN在有效去除高斯噪声、均匀噪声和椒盐噪声的同时，保持了审查数据的完整性，使其适用于基于物联网的土壤监测系统。
