ASTRIANet Data for: Python Computational Inference from Structure (PyCIS)

摘要

<h1> Usage and Overview </h1> <p> This data set contains FITS-formatted files of annotated image data collected by the ASTRIANet telescope network, along with a technical specification document on the telescope sensor and an Excel-formatted table of example image annotations. Two full passes of data are provided as packaged tar.gz files containing contiguous observations of one LEO (Starlink-1422) and one GEO (Navstar-48) satellite over a 5-minute and 11-minute period respectively, and GIF-format video visualizations of each observation are provided. Individual zipped FITS files of single-frame data are also provided for the Starlink satellite. This data is provided as an example of typical ASTRIANet sensor capabilities, and may be used to run demonstrations of the related PyCIS object detection software. For more details on this software, the ASTRIANet telescope network, and the provided data and documents, please see the descriptions provided below. </p> <h1> ASTRIANet </h1> <p> ASTRIANet is a robotic telescope network operated by ASTRIA, designed to record time-series of optical observations of Anthropogenic Space Objects (ASOs) from all orbital regimes LEO-to-GEO. The network supports the Space Domain Awareness (SDA) mission of ASTRIAGraph by closing the information lifecycle from raw observation to trajectory data fusion. Per its operational strategy, the network collects electro-optical observations with significant imaging and tracking noise, and does not collect traditional calibration frames. Accurate ASO detection and tracking algorithms are required to enable Orbit Determination (OD) and bridge the ASTRIANet telescope network with the ASTRIAGraph database by delivering actionable trajectory information. </p> <h1> PyCIS Algorithm Details: A-Contrario Detection and Tracking </h1> <p> Under the so-called "a-contrario" paradigm, structures in data are considered "meaningful" if they are unlikely to occur by chance according to a background noise model. This "Helmholtz Principle" is formalized by a "Number of False Alarms" (or NFA) function which probabilistically defines a structure's “meaningfulness” given both a noise model assumption and some measurement function [2]. We use this dataset to demonstrate the feasibility of detecting ASOs from 16-Megapixel visual imagery observations under the “a-contrario” paradigm. </p> <p> Our motivation in implementing an “a-contrario” detection scheme is two-fold. Firstly, we seek to avoid traditional CCD sensor reductions (e.g. background subtraction, flat-fielding) which can reduce precision as well as operational observation time. Secondly, we seek to avoid making discrete point detections on each image fame, in contrast to many implementations of so-called “track-before-detect” schemes and other three-dimensional “a-contrario” algorithms. </p> <p> Our "Python Computation Inference from Structure" (PyCIS) algorithm [1] builds upon existing low-level implementations of the “a-contrario” principle, which detect edge features in 2D image data by probabilistically detecting meaningful regions of gradient vector field alignments [3, 4]. We have extended these implementations to extract centerline trajectories of of stars and other space objects moving over time. Maintaining the “a-contrario” framework, we then classify meaningful trajectory populations such as stars and sensor noise and identify potential anthropogenic space objects by rejection. For further details, please reference the related software. </p> <h1> Data </h1> <p> The curated set of data included in this submission is as follows: </p> <ul> <li> 20201220_45696_starlink-1422/*.zip: The raw optical image data from the ASTRIANet02 telescope (New Mexico Skies Observatory), stored as separate Flexible Image Transport System (FITS)-format files containing single Header/Data Units (HDUs) of singl