StreakMind and StreakMind Workbench: A Complete Pipeline for Streak Detection and Analysis in Ground-Based Astronomical Surveys

The growing presence of artificial satellites and space debris in low Earth orbit poses significant and well-documented challenges for ground-based astronomical observations. Telescopes originally dedicated to asteroid detection and other celestial surveys routinely capture satellite tracks as unintended by-products. Rather than treating these features purely as noise, this work explores their potential as a source of residual scientific information. When properly identified and characterised, these tracks can contribute to orbit determination, population studies, and analyses of the temporal evolution of the low Earth orbit (LEO) environment.

The work presented at CPESS-8 (Málaga, May 2025) described the early stages of this effort, including a manual detection and cross-matching pipeline applied to archival images from the La Sagra observatory, the construction of a manually annotated dataset, and the initial development of a machine-learning-based detection system. This dataset has since become the foundation of a fully automated analysis pipeline.

The automated detection engine, now formally named StreakMind, is based on the YOLO11 Oriented Bounding Box (OBB) architecture implemented through the Ultralytics framework. The system, recently published in Astronomy & Astrophysics (April 2026), has been trained on the manually annotated dataset of satellite streak images from La Sagra and produces detections in the form of oriented bounding boxes that closely follow the geometry of each streak. From these detections, StreakMind derives astrometric parameters, including the streak centroid and endpoints via World Coordinate System (WCS) transformation, as well as the position angle. These measurements can be formatted into Minor Planet Center (MPC) 80-column observation records, enabling compatibility with standard astrometric reporting workflows.

Since CPESS-8, the pipeline has been extended to incorporate calibrated photometry. While in earlier stages photometry was used only as part of a post-processing procedure, without yielding calibrated magnitudes, the current approach aims to provide a calibrated magnitude for each detected satellite passage. This is achieved through aperture photometry performed along each detected streak, with calibration based on reference stars extracted from the Gaia DR3 catalogue. This addition enables not only geometric detection but also photometric characterisation of the observed objects.

In parallel, further developments are being explored to enhance the scientific scope of the system. In particular, cross-matching detected tracks against asteroid catalogues is currently under investigation. This would allow StreakMind to distinguish between artificial and natural objects and potentially contribute to minor planet detection and validation, thereby extending its applicability beyond satellite tracking.

A major milestone in the project has been the development of StreakMind Workbench, a desktop graphical interface that provides an interactive implementation of the full StreakMind pipeline. Developed using PyQt5, the Workbench is designed to offer a structured and reproducible workflow for observers and researchers without requiring direct interaction with Python code. The interface integrates image loading and visualisation (with native FITS support), model inference, calibrated photometry, database management, and training functionalities within a single integrated framework. The system includes configurable inference parameters, validation of MPC observatory codes, and an integrated SQLite database that stores detections, MPC-formatted observation records, camera metadata extracted from FITS headers, and image-level statistics. Additionally, a dedicated training module allows users to retrain or fine-tune the YOLO11 model on newly annotated datasets, ensuring adaptability to different observational setups.

StreakMind Workbench represents a transition from a research-oriented pipeline to an operational tool, facilitating large-scale data processing and improving accessibility for the astronomical community. A detailed description of the Workbench is currently being prepared as a companion paper.