Classification for geosynchronous satellites with deep learning and multiple kernel learning | JoVE Visualize

Area of Science:

Space Situational Awareness
Astrodynamics
Machine Learning for Space Applications

Background:

The geostationary orbit (GEO) belt is a critical space asset requiring robust monitoring.
Accurate identification of GEO satellites is essential for ensuring the safety of space objects.
Understanding the shape and attitude of GEO satellites is a key component of space situational awareness.

Purpose of the Study:

To develop and validate a deep learning algorithm for synchronous identification of GEO satellite shape and attitude using light curve data.
To evaluate the performance of a combined Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) architecture with Multiple Kernel Learning (MKL).
To compare the proposed method against conventional machine learning techniques for GEO object identification.

Main Methods:

Acquisition of light curve data for five GEO satellites over one year from laboratory and simulation sources.
Construction of a CNN-RNN network architecture for automatic extraction of multi-scale features from light curves.
Application of Multiple Kernel Learning (MKL) to fuse extracted features.
Classification and recognition of GEO satellite shape and attitude using a Support Vector Machine (SVM).

Main Results:

The proposed deep learning model successfully identified the shape and attitude of GEO satellites synchronously.
The CNN-RNN architecture effectively extracted relevant features from light curve data.
Multiple Kernel Learning demonstrated superior performance in feature fusion compared to single kernels.
The developed algorithm outperformed conventional methods like Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA).

Conclusions:

The integrated deep learning approach (CNN-RNN with MKL) provides an effective method for identifying GEO satellite shape and attitude from optical light curves.
This method significantly enhances the capabilities of space situational awareness by enabling accurate and synchronous classification of GEO objects.
The study highlights the advantage of using multiple kernels over single kernels for improved classification accuracy in this domain.