Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AstroSpy: On detecting Fake Images in Astronomy via Joint Image-Spectral Representations

Mohammed Talha Alam, Raza Imam, Mohsen Guizani, Fakhri Karray

TL;DR

AstroSpy tackles the rising challenge of AI-generated astronomical images by jointly leveraging spatial and spectral cues to distinguish real NASA imagery from fakes. The method uses two ResNet50 pathways to extract image features and Fourier-magnitude spectral features, fused into a joint embedding for binary classification. A dataset of approximately 18k images (9k real, 9k synthetic) from NASA and FLARE underpins extensive in-domain and cross-domain evaluations, including comparisons with natural, medical, and face images generated by diverse tools. Results show state-of-the-art in-domain accuracy around 98.5% and strong cross-domain generalization, highlighting AstroSpy’s potential to safeguard scientific integrity and public trust in astronomy.

Abstract

The prevalence of AI-generated imagery has raised concerns about the authenticity of astronomical images, especially with advanced text-to-image models like Stable Diffusion producing highly realistic synthetic samples. Existing detection methods, primarily based on convolutional neural networks (CNNs) or spectral analysis, have limitations when used independently. We present AstroSpy, a hybrid model that integrates both spectral and image features to distinguish real from synthetic astronomical images. Trained on a unique dataset of real NASA images and AI-generated fakes (approximately 18k samples), AstroSpy utilizes a dual-pathway architecture to fuse spatial and spectral information. This approach enables AstroSpy to achieve superior performance in identifying authentic astronomical images. Extensive evaluations demonstrate AstroSpy's effectiveness and robustness, significantly outperforming baseline models in both in-domain and cross-domain tasks, highlighting its potential to combat misinformation in astronomy.

AstroSpy: On detecting Fake Images in Astronomy via Joint Image-Spectral Representations

TL;DR

AstroSpy tackles the rising challenge of AI-generated astronomical images by jointly leveraging spatial and spectral cues to distinguish real NASA imagery from fakes. The method uses two ResNet50 pathways to extract image features and Fourier-magnitude spectral features, fused into a joint embedding for binary classification. A dataset of approximately 18k images (9k real, 9k synthetic) from NASA and FLARE underpins extensive in-domain and cross-domain evaluations, including comparisons with natural, medical, and face images generated by diverse tools. Results show state-of-the-art in-domain accuracy around 98.5% and strong cross-domain generalization, highlighting AstroSpy’s potential to safeguard scientific integrity and public trust in astronomy.

Abstract

The prevalence of AI-generated imagery has raised concerns about the authenticity of astronomical images, especially with advanced text-to-image models like Stable Diffusion producing highly realistic synthetic samples. Existing detection methods, primarily based on convolutional neural networks (CNNs) or spectral analysis, have limitations when used independently. We present AstroSpy, a hybrid model that integrates both spectral and image features to distinguish real from synthetic astronomical images. Trained on a unique dataset of real NASA images and AI-generated fakes (approximately 18k samples), AstroSpy utilizes a dual-pathway architecture to fuse spatial and spectral information. This approach enables AstroSpy to achieve superior performance in identifying authentic astronomical images. Extensive evaluations demonstrate AstroSpy's effectiveness and robustness, significantly outperforming baseline models in both in-domain and cross-domain tasks, highlighting its potential to combat misinformation in astronomy.
Paper Structure (18 sections, 6 equations, 6 figures, 3 tables)

This paper contains 18 sections, 6 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Motivation
  3. Related Works
  4. Contribution
  5. Method: AstroSpy
  6. Dataset Preparation
  7. Joint Embeddings
  8. Training Procedure
  9. Results and Discussion
  10. Experiments
  11. Datasets
  12. Implementation Details
  13. In-Domain Generalization
  14. Generalization to Out-of-Domain Distributions
  15. Feature Shifts via AstroSpy
  16. ...and 3 more sections

Figures (6)

  • Figure 1: AstroSpy architecture: Real images from database archive, and fake images generated by Stable Diffusion, are transformed into spectral embeddings using Fourier Transform. Both image and spectral features are extracted using ResNet50, concatenated into joint embeddings, and used to train a binary classifier for detecting real and fake images.
  • Figure 3: Samples of real (left column) and synthetic (right column) astronomical images with their corresponding spectra.The spectra reveal distinct patterns that help in differentiating real images from fakes.
  • Figure 4: Samples of out-of-domain datasets used for testing AstroSpy's generalization capabilities. The top row shows real images from a specific class, while the bottom row shows corresponding generated images.
  • Figure : (a) Baseline (bs.)
  • Figure : (a) Baseline (bs.)
  • ...and 1 more figures