ABSTRACT Deep learning‐based synthetic aperture radar (SAR) ship image processing is valuable for civilian and military applications, yet training deep models requires abundant high‐resolution SAR ship images. Due to the high cost of SAR sensors and limited data availability, constructing large‐scale, high‐quality datasets remains challenging. SAR image generation offers a promising solution. However, most existing generation methods focus on natural images, with few targeting SAR ship targets. This paper proposes a high‐resolution SAR ship image generation method using an improved denoising diffusion generative adversarial network. To enhance detail generation, a context‐aware upsampling filter is designed to incorporate SAR‐specific information, enabling the backbone network to capture more effective features during upsampling. Experiments show that our method achieves a Fréchet inception distance of 55.21 and a structural similarity index of 0.7245, producing high‐quality SAR‐consistent images with rich details.
Ji et al. (Thu,) studied this question.