Skip to main content
SearchLoginLogin or Signup

Trans-Neptunian Objects Detection from Ground-Based Telescopes: A Fast and Efficient Approach using Deep Learning CNN with the Deep Ecliptic Exploration Project (DEEP)

Presentation #219.01 in the session TNO (Poster + Lightning Talk)

Published onOct 23, 2023
Trans-Neptunian Objects Detection from Ground-Based Telescopes: A Fast and Efficient Approach using Deep Learning CNN with the Deep Ecliptic Exploration Project (DEEP)

The trans-Neptunian region offers a unique opportunity to study the same planetesimals from which the planets in our solar system formed. The Deep Ecliptic Exploration Project (DEEP), a multiyear survey program investigating the trans-Neptunian region, characterizing the size and shape distribution of faint TNOs and their physical properties in relation to dynamical class and size, while also following the candidate TNOs over two or more years. The project employs the Dark Energy Camera (DECam) on the 4-meter Cerro Tololo Inter-American Observatory (CTIO) Blanco telescope, identifying and characterizing the orbits of thousands of Trans-Neptunian Objects (TNOs).

The DEEP search for TNOs was performed using a shift-and-stack moving object detection algorithm and recovered 110 new objects. We chose to pursue a different method, aiming to provide a fresh perspective on the detection and characterization of TNOs.

Here we present an AI-based moving object detection technique. Instead of following the traditional shift and stack method, we perform simple stacking on single field night images using different statistics, which later are seached for TNOs trails with the help of AI. The primary objective of the project is to develop an optimized algorithm for expedited analysis of large datasets, enabling the detection of faint TNOs using the DEEP results as benchmark magnitude limit of R ̴ 26.2.

The algorithm comprises two key components. The first involves pre-processing, which encompasses six major steps ranging from background subtraction to the combination of images using different statistics. The processes effectively generates discernible trails in the final combined image. Subsequently, these images are fed into an AI framework where the YOLOv8 deep learning convolutional neural network is trained to identify moving object trails out of noise. Notably, the algorithm’s strength lies in its ability to quickly process large volumes of data while minimizing false positives. The ultimate aim is to enable real-time detection of moving objects as soon as the data becomes available. In this presentation, I will delve into the key methods used in this innovative technique and discuss some of the challenges faced during its development, with a focus on the role of AI in enabling the fast and efficient detection of moving objects in the trans-Neptunian region. Also, I will discuss how this can be implemented in deep drilling fields from LSST, and other wide field surveys, possibly enhancing the yield of new Solar System Objects, particularly for those that exhibit a sky motion unsuitable for deep drilling.

Comments
0
comment
No comments here