Machine learning-based classification for Single Photon Space Debris Light Curves

TL;DR

This work aims to classify Single Photon Space Debris using the ML framework, and successfully classified Space Debris LCs captured on Single Photon basis, obtaining accuracies as high as 90.7%.

Abstract

The growing number of man-made debris in Earth's orbit poses a threat to active satellite missions due to the risk of collision. Characterizing unknown debris is, therefore, of high interest. Light Curves (LCs) are temporal variations of object brightness and have been shown to contain information such as shape, attitude, and rotational state. Since 2015, the Satellite Laser Ranging (SLR) group of Space Research Institute (IWF) Graz has been building a space debris LC catalogue. The LCs are captured on a Single Photon basis, which sets them apart from CCD-based measurements. In recent years, Machine Learning (ML) models have emerged as a viable technique for analyzing LCs. This work aims to classify Single Photon Space Debris using the ML framework. We have explored LC classification using k-Nearest Neighbour (k-NN), Random Forest (RDF), XGBoost (XGB), and Convolutional Neural Network (CNN) classifiers in order to assess the difference in performance between traditional and deep models. Instead of performing classification on the direct LCs data, we extracted features from the data first using an automated pipeline. We apply our models on three tasks, which are classifying individual objects, objects grouped into families according to origin (e.g., GLONASS satellites), and grouping into general types (e.g., rocket bodies). We successfully classified Space Debris LCs captured on Single Photon basis, obtaining accuracies as high as 90.7%. Further, our experiments show that the classifiers provide better classification accuracy with automated extracted features than other methods.

Figures (16)

• Figure 1: Raw LC measurements of TOPEX/Poseidon (NORAD 22076), a decommissioned satellite in LEO with a spin period of 10.73 sec as per 2016 kucharski2017.
• Figure 2: A LC of satellite Sentinel-3b (NORAD 43437), captured on 2022/4/30. The first half of the measurement experiences interruptions in the time domain. Though the measurement file can be considered faulty, the latter half of the measurement can still potentially be used for analysis.
• Figure 3: The schematic diagram of the one-dimensional CNN architecture used for the LCs classification problem. Three convolution and max pooling layers are used between the input and output layers. All the layers used a fixed kernel and filter size of $64$ and $3$, respectively. The tanh activation function is used for convolution layers, whereas the softmax activation function has been chosen for output layers. Batch normalization is used between the convolution and max pooling layers.
• Figure 4: The schematic representation of methods used in this work at example of the Types subset.
• Figure 5: Mapping examples of a lock-step measure (e.g., Euclidean Distance) versus an elastic measure (e.g., Dynamic Time Warping) Wang2013, graphic inspired by Abanda2019.