Abstract:Synthetic aperture radar (SAR) and optical sensing are two important means of Earth observation. SAR can be used for all-day and all-weather Earth observation, but it has the disadvantages of speckle noise and geometric distortion, which are not conducive to human eye recognition. Optical image conforms to the characteristics of human visual observation, but it is easily affected by climate and time. Therefore, to integrate the advantages of the two, researchers have carried out extensive work on SAR-to-optical (S2O) image translation. Most of the existing methods for S2O image translation are supervised and need paired training samples, limiting its large-scale application in remote sensing field. Thus, we give priority to an unsupervised S2O image translation method. Meanwhile, we find that the images generated by unsupervised methods suffer from significant detail deficiencies. To solve this problem, we propose a fine-grained generative adversarial network (FG-GAN) introducing three strategies to enhance the detailed information in generated optical images. First, we design an unbalanced generator (UBG) with complex encoder networks and relatively simple decoder networks. The complex encoder extracts abundant feature information, while the decoder obtains key details by filtering these features. Second, to match the learning ability of the generator, we present a multiscale discriminator (MSD) to enhance the discriminant ability of the network. Third, we propose a comprehensive normalization group (CNG) to promote the physical representation consistency of SAR and optical images. Extensive experiments have been conducted, and the results show that our method is superior to the state-of-the-art (SOTA) methods on both subjective and objective evaluation indicators. Moreover, our FG-GAN has a significant improvement on classification accuracy, indicating its potential in facilitating the performance of practical remote sensing tasks.

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

A SAR-to-Optical Image Translation Method Based on Conditional Generation Adversarial Network (cGAN)

Improved SAR Image Generation with Double Top-K Training Method on Auxiliary Classifier GAN

A Target SAR Image Expansion Method Based on Conditional Wasserstein Deep Convolutional GAN for Automatic Target Recognition

Sar Image Synthesis Based On Conditional Generative Adversarial Networks

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

Improved Cgan for SAR-to-Optical Image Translation

Parallel Connected Generative Adversarial Network with Quadratic Operation for SAR Image Generation and Application for Classification

DMSC-GAN: A C-Gan-based Framework for Super-Resolution Reconstruction of SAR Images

Semi-Supervised SAR ATR Via Conditional Generative Adversarial Network with Multi-Discriminator

A Distribution and Structure Match Generative Adversarial Network for SAR Image Classification

A Deep Convolutional Generative Adversarial Networks (DCGANs)-Based Semi-Supervised Method for Object Recognition in Synthetic Aperture Radar (SAR) Images

A Two-Stage SAR Image Generation Algorithm Based on GAN with Reinforced Constraint Filtering and Compensation Techniques

Dialectical GAN for SAR Image Translation: From Sentinel-1 to TerraSAR-X

Generation of SAR Images with Features for Target Recognition

Hybrid Cgan: Coupling Global and Local Features for SAR-to-Optical Image Translation

FG-GAN: A Fine-Grained Generative Adversarial Network for Unsupervised SAR-to-Optical Image Translation.

A-CGAN based transformation from ISAR to optical image

IAM-ACGAN: A High-Accuracy Approach for SAR Image Augmentation

Synthetic aperture radar ground target image generation based on improved Wasserstein generative adversarial networks with gradient penalty

Integrated GANs: Semi-Supervised SAR Target Recognition