Review of SAR Image Generation Methods for Maritime Ships Under Complex Backgrounds

Synthetic Aperture Radar (SAR) ship images play a significant role in the maritime remote sensing field and are of great value for data-driven tasks such as ocean monitoring and ship detection. However, high-resolution SAR imagery of ships with comprehensive annotations is difficult and costly to acquire, and data scarcity has become a major constraint on the development of ship detection and recognition. To address these challenges, this paper reviews SAR ship image generation in complex backgrounds, covering conventional electromagnetic simulation and deep generative models, including variational autoencoders (VAEs), generative adversarial networks (GANs), and denoising diffusion probabilistic models (DDPMs). Their generation mechanisms, conditional control, computational cost, and physical consistency are compared, together with public datasets, multidimensional evaluation methods, and downstream detection applications. VAEs offer stable training but tend to lose high-frequency scattering details. GANs generate sharp images efficiently, but still face limitations in training stability, sample diversity, and physical consistency. Diffusion models provide better diversity and conditional control but require substantial computation and stronger constraints on scattering statistics and imaging geometry. Integrating electromagnetic priors, multi-condition control, and lightweight architectures is therefore important for improving the fidelity and practical utility of SAR ship image generation.