Distributed Intelligent Framework for Remote Area Observation on Multilayer Non-Terrestrial Networks

Satellite-based Remote Area Observation systems are becoming increasingly popular in the upcoming 6 G world. However, traditional Earth Observation (EO) systems suffer from communication requirements, reliability, and data privacy issues. To address these issues, we propose a multilayered Non-Terrestrial Network (NTN) based EO framework for remote area observation purposes. The proposed framework includes the air network along with traditional satellite networks for reliable and low-cost EO services. Additionally, with onboard edge computing facilities, the proposed EO framework can process data in space. Next, given the importance of intelligent services in the 6 G world, we extend the multi-layered EO framework and propose a novel Distributed Learning (DL) solution for federated training. The proposed framework is defined as Generalized Federated Split Transfer Learning (GFSTL), which can induce split and transfer learning tools into a federated learning framework for improving overall training performance and accuracy. Moreover, GFSTL uses Unmanned Aerial Vehicles (UAVs) for improved data accuracy and image quality in challenging terrains, ensuring increased accuracy in EO applications, and establishes a resilient model for efficient and secure training across distributed platforms, making it both efficient and accurate. In addition, SL helps resource-constrained UAVs perform the task efficiently, enhancing scalability and extensibility. Finally, we conduct experiments to provide theoretical and numerical insight into the performance of the proposed method.