Application of Artificial Neural Networks to Identify Orbital Resonances in the Motion of Artificial Earth Satellites

Traditional manual analysis of graphical data reflecting the behavior of critical arguments of orbital resonances is a very labor-intensive process. To address this problem, an attempt was made to apply convolutional neural networks (CNNs) to automatically classify orbital resonances based on graphical representations of critical arguments. It is shown that the model trained on the resonance argument plots obtained for objects from the 1 : 2 orbital resonance region demonstrated high accuracy on test data from this resonance region but showed a significant decrease in performance when applied to other resonance regions of 2 : 1 and 3 : 1 due to the presence of previously unseen graphical examples. After retraining the model on a more diverse dataset including all three resonance regions investigated, a more accurate model was developed, achieving an average classification accuracy of 98%. The study not only confirmed the high effectiveness of CNN for automating resonance classification but also highlighted the critical importance of using a diverse training dataset to build a highly accurate classification model.