CBCTL-IDS: A Transfer Learning-Based Intrusion Detection System Optimized With the Black Kite Algorithm for IoT-Enabled Smart Agriculture

Smart Agriculture(SA) stands as a critical application frontier for the evolution of the IoT, catalyzing the transformation of traditional agricultural practices into modern, data-driven, intelligent, and automated systems. This transition not only elevates the levels of automation and intelligence in agricultural production but also introduces a spectrum of cybersecurity challenges. In particular, the widespread deployment of interconnected devices within these systems poses significant risks to the security and integrity of data. As a pivotal defense mechanism against malicious activities and anomalous traffic, Network Intrusion Detection Systems (NIDS) play an indispensable role in ensuring the security of cyberspace. Nevertheless, existing methods for detecting anomalous traffic in IoT networks still exhibit limitations in accuracy, thereby hindering the efficiency of identifying potential threats. To enhance the capability of threat detection, this paper proposes a novel network intrusion detection method, CBCTL-IDS, based on transfer learning. This method integrates three core components: Convolutional Neural Networks (CNN), the Black Kite Algorithm (BKA), and a Confidence Averaging mechanism. By incorporating pre-trained models such as MobileNet, EfficientNet, Xception, VGG19, and Inception into the domain of IoT intrusion detection, CBCTL-IDS leverages transfer learning to achieve efficient feature extraction and classification, addressing the limitations of traditional methods in feature representation. Furthermore, the BKA is utilized to adaptively optimize the hyperparameters of multiple DL models, significantly enhancing their classification performance. Through ensemble learning, the top three performing models are selected, and the Confidence Averaging mechanism is applied to further improve the stability and reliability of detection results. Experimental results demonstrate that the CBCTL-IDS method achieves detection accuracy rates exceeding 99% on three IoT intrusion detection datasets: ToN-IoT, Edge-IIoTset, and WSN-DS significantly outperforming existing mainstream methods. This approach provides robust technical support for the security protection of IoT systems.