R. Hafid Hardyanto, Moh. Khairudin, Rustam Asnawi
Efficient and sustainable fertilizer management is critical for enhancing rice productivity while minimizing environmental degradation caused by excessive nitrogen, phosphorus, and potassium use. Conventional fertilization methods often rely on intuition and ignore sitespecific variability in soil, crops, and weather conditions. This paper presents an Edge Artificial Intelligence of Things (Edge AIoT) framework that employs a Temporal Convolutional Network (TCN) with attention for site-specific fertilizer recommendation in rice farming. The system integrates multimodal IoT sensor data-comprising soil moisture, temperature, humidity, rainfall, and light intensityprocessed in a simulated low-power edge environment (Raspberry Pi 4). The TCN, implemented with dilated causal convolutions and attention, captures both short-term fluctuations and long-term seasonal nutrient trends. The trained model was quantized into TensorFlow Lite (22.51 KB) for efficient on-device inference. Simulation experiments using field-acquired IoT sensor data demonstrate that the quantized TCN+Attention model achieves MAE =0.025, RMSE =0.021, and R2=0.906, confirming its suitability for real-time edge deployment. These findings highlight the feasibility of scalable, low-latency AIoT-based nutrient recommendation systems that can support sustainable fertilizer management in rural rice farming. © 2025 IEEE.
Engineering Science, Universitas Negeri Yogyakarta, Yogyakarta, Indonesia; Universitas Negeri Yogyakarta, Department of Electrical Engineering, Yogyakarta, Indonesia
