An LSTM–AE–Bayes embedded gateway for real-time anomaly detection in agricultural wireless sensor networks

Agricultural wireless sensor networks have become a critical data source for agricultural big data analytics. However, due to factors such as network transmission errors and human interference, multiple sensor nodes in these networks may generate anomalous data. Such anomalies can lead to inaccurate analyses, adversely affecting crop growth, and result in unnecessary energy consumption during data transmission. To address these challenges, this paper proposes a greenhouse gateway incorporating an anomaly detection algorithm. By integrating the detection algorithm directly into the gateway, only data deemed normal are forwarded. The gateway, built around the STM32F407ZGT6 microcontroller, utilizes LoRa and 4 G modules for wireless data transmission, enabling real-time data visualization on a cloud server. Experimental results demonstrate that the proposed anomaly detection algorithm not only outperforms conventional methods but also remains practical and effective when embedded within the gateway. We further observed that our approach reduced unnecessary data traffic by over 9 % and improved F1-score by about 5 % compared to state-of-the-art methods, confirming its efficiency in both performance and resource utilization.