Enhancing IoT communication security in smart agriculture using artificial intelligence
Abstract
The increased use of IoT systems in farming has led to more efficient farming practices that leverage data; however, it has also made communication security more vulnerable. Static security technologies, such as fixed encryption and intrusion detection, are ineffective in farms due to the rapid pace of technological advancements and limited resource availability. In this case, AI is applied in a novel way to secure IoT communication by identifying unusual events and selecting the most suitable type of encryption. To achieve this, an LSTM-based network utilizes attention mechanisms to detect abnormal traffic as it occurs, and a Deep Q-Learning algorithm matches encryption requirements based on the detected risk, the device's energy level, and the additional time required for the process. The system was designed and assessed using the Smart Agriculture Traffic Dataset, and its performance was corroborated using the NSL-KDD benchmark. As the results also demonstrate, the LSTM with attention modeling achieves an accuracy of 94.3% while reducing the likelihood of false positives. Additionally, the adjustable encryption module reduces energy and latency usage by approximately 18.7% and 26.0% compared to fixed AES-256 encryption. Therefore, applying interpretable anomaly detection in conjunction with context-aware crypto policies is effective and utilizes fewer resources in safeguarding smart agriculture. We can use the framework in real-time, and it has a high chance of benefiting many IoT applications.
Authors

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.