Attribute-Controlled SAR Target Image Generation Driven by Diffusion Models

Sufficient Synthetic Aperture Radar (SAR) target images are of great importance for the advancement of SAR systems. However, the necessity for professional annotation of SAR target images results in a high acquisition cost, which in turn restricts the availability of SAR images in practical applicat...

Full description

Saved in:
Bibliographic Details
Published in2024 5th International Conference on Electronic Communication and Artificial Intelligence (ICECAI) pp. 462 - 468
Main Authors Chen, Ziyi, Dai, Mingcheng, Cong, Yulai, Lu, Rongwei
Format Conference Proceeding
LanguageEnglish
Published IEEE 31.05.2024
Subjects
attribute-controlled generation
Azimuth
Costs
diffusion model
Diffusion models
image generation
Image synthesis
Statistical analysis
Training
Transformer
Visualization
Online AccessGet full text

Cover

More Information
Summary:Sufficient Synthetic Aperture Radar (SAR) target images are of great importance for the advancement of SAR systems. However, the necessity for professional annotation of SAR target images results in a high acquisition cost, which in turn restricts the availability of SAR images in practical applications and further impedes the development of SAR technology applications. In order to effectively expand the labeled SAR target image dataset, many scholars have devoted themselves to the research of GAN-based SAR image generation technology. However, few GAN-based SAR image generation methods can achieve controllable generation of multi-attribute combinations of SAR images, and the inherent instability during GAN training also limits the practicality and popularity of the methods. To overcome these limitations, we propose an innovative model named Attribute-controlled Diffusion Model (AcDM), which leverages the superior image generation capabilities of diffusion models to enhance both the quality and controllability of SAR image synthesis. Specifically, to achieve precise control over the azimuth and category attributes in SAR target image generation, we design a module called the Attribute Information Fusion Module (AIFM). This module facilitates an in-depth integration of SAR target images with their respective attribute information, enabling the generation of SAR target images with highly controllable attributes. The AcDM has proven to be highly effective in accomplishing SAR image generation tasks.
DOI:10.1109/ICECAI62591.2024.10675205