Akademik Veri Yönetim Sistemi
Aerospace Systems, 2025 (Scopus)
Air traffic control (ATC) systems traditionally rely on radio-based identification methods such as ADS-B, SSR, and IFF. Although effective under normal conditions, these systems are increasingly vulnerable to misidentification errors in non-cooperative scenarios, such as when aircraft deliberately suppress or spoof signals, or when technical and environmental factors degrade coverage. Such limitations are especially critical when distinguishing between commercial (friendly) and military (potentially hostile) aircraft, where classification errors directly compromise airspace security. To address this challenge, we propose a complementary identification layer based on satellite imagery and deep learning, explicitly designed as an operational backup for ATC systems. Unlike prior works that benchmark detection and classification algorithms in generic remote-sensing contexts, our study directly targets the safety-critical problem of friend-or-foe aircraft classification under degraded communication conditions. We implement convolutional neural networks (CNNs), residual networks (ResNets), and capsule networks (CapNets) to classify commercial and military aircrafts. Our curated dataset of more than 15,000 images is augmented with noise, illumination shifts, and viewpoint variations to simulate realistic ATC surveillance scenarios. The comparative results show that CapNet consistently outperforms CNN and ResNet, achieving 97% classification accuracy and the highest AUC score of 0.991. CapNet’s ability to preserve spatial hierarchies and pose information enhances robustness against noise, lighting changes, and perspective distortions, making it particularly suitable for non-cooperative identification tasks. While CapNet introduces higher computational demands, its resilience and accuracy demonstrate its potential as a safety-critical auxiliary tool for ATC, significantly reducing the risk of misidentification and unauthorized airspace penetration in scenarios where radar and radio-based systems may fail.