A Neural Network analysis of single-channel Light Curves for the characterization of Resident Space Objects

The increasing congestion of Earth’s orbital environment due to the rising number of satellite launches has heightened the risk of in-orbit collisions and the accumulation of space debris, posing significant challenges to space sustainability. Distinguishing between active and inactive space objects is a crucial activity in the Space Situational Awareness field to prevent fragmentation events and support orbital slot management. Recent advancements in data-driven techniques have enhanced the ability to analyze photometric data, allowing for automated and robust feature extraction. In this context, this paper presents a novel approach based on Deep Neural Networks to manage real light curves, provided by open-source databases, in order to classify resident space objects as active or inactive. A comparison between different Fully Convolutional Neural Networks and Hybrid Recurrent Convolutional Neural Networks is made under several tuning parameters and combined with different pre-processing techniques.