SAS Vision
Description
SAS VISION is a convolutional neural network (CNN)-based computer vision model designed to identify previously undocumented astronomical objects such as AGNs, NGCs, and stellar sources using large volumes of XMM-Newton observational data. This system fine-tunes open source CNN through transfer learning using XMM-Newton raw FITS files and their corresponding SAS pipeline-processed outputs. It performs a two-stage process: Denoising & Enhancement (using DnCNN, AutoEncoders, U-Net/W-Net, or DDPMs to preprocess and enhance noisy X-ray data), and Object Identification (through CNNs with instance segmentation (Mask R-CNN), enabling detection and classification of celestial objects. In addition to this, cross-observations will enable spatial matching of object features to confirm repeat detections. This model could enable large-scale automated cataloging of undocumented objects not present in SIMBAD, Vizier, or the databases queried by Aladin.
Detailed example
Denoised Images, Segmentation Masks, Object Classifications, Multi-Observation Object Tracker
AI / analytics pattern
Computer Vision: AI that processes and interprets visual data (e.g., images and videos).
Automation level / stage
a) Pre-deployment – The use case is in a development or acquisition status.
Expected benefit
Discovery of New Objects, Accelerates Identification tasks, Improved Catalogs
Controls / human review
ATO: Not reported; PIA: Not published