Securing smart agriculture networks using bio-inspired feature selection and transfer learning for effective image-based intrusion detection

The smart agricultural system integrates advanced technologies to optimize farming practices and increase productivity. It gathers images from sensors, drones, and other sources to create detailed maps of crop yield variability, soil composition, and vegetation indices. In the realm of network security, the rise of smart agricultural systems presents both challenges and opportunities. This study addresses the crucial need for a tailored Intrusion Detection System (IDS) for agricultural networks, focusing on image-based traffic. Employing advanced techniques such as transfer learning and bio-inspired algorithms, we propose a novel IDS architecture adept at handling the unique characteristics of image traffic. Our methodology includes imbalanced data handling, transformation, feature extraction utilizing the pre-trained VGG16 model, feature selection via the bio-inspired Binary Greylag Goose (BGGO) algorithm, and classification employing a Random Forest classifier. Evaluation on the new CICIoT2023 dataset showcases high accuracy, with 99.41% for multiclass and 99.83% for binary classifications with a reduced number of features from 25088 to 6327. These results underscore the significance of efficient IDS solutions for the evolving landscape of agricultural technologies, promising enhanced network security in smart agriculture environments.