An optimized CNN-based intrusion detection system for reducing risks in smart farming

Abstract

Smart farming is a well-known and superior method of managing a farm, becoming more prevalent in today's contemporary agricultural practices. Crops are monitored for their overall health and productivity using various information and agricultural technology. This involves monitoring the state of the field crops and looking at various other indicators. The ultimate objective of smart farming is to decrease the amount of money spent on agricultural inputs while maintaining the same level of quality in the final result. The Internet of Things has enabled smart farming via different data collecting and storage methods. For example, intelligent sensor networks gather field data for optimum plant irrigation in contemporary irrigation systems. As the dependency of smart farming on the IoT environment increases, they become more vulnerable to cyber-attacks. The reason is that IoT networks include so many nodes with few resources available, making securing them a challenging problem. Therefore, the need for an intrusion detection system (IDS) that can adapt to such difficulties is of the utmost importance. This paper proposes a new system that detects intrusions in IoT networks used in agriculture. The NSL KDD data set is used to evaluate the proposed method, which starts by performing several pre-processing steps on the original feature set. Important features are selected using recursive feature elimination, then converted into square color images. Now input images are suitable to be learned by different CNN architectures such as VGG16, Inception, and Xception models. Performance comparisons of CNN models with classical machine learning algorithms are evaluated using accuracy, F1 Score, Recall, and precision metrics.