Abstract:In recent years, diffusion models (DMs) have become a popular method for generating synthetic data. By achieving samples of higher quality, they quickly became superior to generative adversarial networks (GANs) and the current state-of-the-art method in generative modeling. However, their potential has not yet been exploited in radar, where the lack of available training data is a long-standing problem. In this work, a specific type of DMs, namely denoising diffusion probabilistic model (DDPM) is adapted to the SAR domain. We investigate the network choice and specific diffusion parameters for conditional and unconditional SAR image generation. In our experiments, we show that DDPM qualitatively and quantitatively outperforms state-of-the-art GAN-based methods for SAR image generation. Finally, we show that DDPM profits from pretraining on largescale clutter data, generating SAR images of even higher quality.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the lack of a large amount of training data in Synthetic Aperture Radar (SAR) image generation. Specifically, the paper explores how to use the Denoising Diffusion Probability Model (DDPM) to generate high - quality SAR images, in order to overcome the problems such as instability, low diversity, mode collapse and difficulty in hyper - parameter optimization in the training process of traditional Generative Adversarial Networks (GANs). By adapting DDPM to the SAR field, researchers aim to improve the quality of generated SAR images and show that DDPM pre - trained on large - scale clutter datasets can generate higher - quality SAR images. The key points of the paper include: - **Problem Background**: Although SAR technology provides high - resolution imaging data, its high acquisition cost leads to scarce available data, which limits the development speed of applications such as object recognition based on deep learning. - **Solution**: Introduce DDPM as a new generation model for generating high - quality SAR images. Compared with traditional GANs, DDPM performs better in terms of the quality and diversity of generated images. - **Methodology**: The paper describes in detail the working principle of DDPM, including the forward diffusion process and the reverse denoising process, and discusses the choices of network design, such as the number of diffusion steps, noise scheduling schemes, etc. - **Experimental Verification**: Experiments are carried out using the MSTAR dataset to evaluate the performance of different models (including DDPM and several GAN variants). The results show that, whether in unconditional or conditional generation, DDPM is significantly superior to GANs in terms of the quality and diversity of generated SAR images. - **Innovation**: Propose a method of pre - training the model using large - scale unlabeled clutter data, which further improves the quality of generated SAR images. In conclusion, this paper effectively solves the problem of insufficient data in SAR image generation by introducing and optimizing the DDPM model, providing new ideas and technical support for subsequent research and applications.

SAR Image Synthesis with Diffusion Models

Synthetic Aperture Radar Deep Statistical Imaging Through Diffusion Generative Model Conditional Inference

DDM-CGAN: a Modified Conditional Generative Adversarial Network for SAR Target Image Generation

Diffusion Models for Interferometric Satellite Aperture Radar

Conditional Diffusion for SAR to Optical Image Translation

SatDM: Synthesizing Realistic Satellite Image with Semantic Layout Conditioning using Diffusion Models

SAR Image Generation Method Using DH-GAN for Automatic Target Recognition

SAR to Optical Image Translation with Color Supervised Diffusion Model

SAR Despeckling via Regional Denoising Diffusion Probabilistic Model

Conditional Brownian Bridge Diffusion Model for VHR SAR to Optical Image Translation

Learning SAR-to-Optical Image Translation Via Diffusion Models with Color Memory

Semantic Image Synthesis Via Diffusion Models

Synthetic Aperture Radar Image Generation With Deep Generative Models

Attribute-Controlled SAR Target Image Generation Driven by Diffusion Models

Generating Simulated SAR Images Using Generative Adversarial Network

Accelerating Diffusion for SAR-to-Optical Image Translation via Adversarial Consistency Distillation

Efficient End-to-End Diffusion Model for One-step SAR-to-Optical Translation

Single image super-resolution with denoising diffusion GANS

SAR Image Despeckling Based on Denoising Diffusion Probabilistic Model and Swin Transformer

SAR Image Target Recognition Using Diffusion Model and Scattering Information

Progressive Image Synthesis from Semantics to Details with Denoising Diffusion GAN