MAE-GAN: A Multifeature Adaptive Enhancement Approach for Multiangle Spaceborne InSAR Complex-Valued Images

Multiangle spaceborne interferometric synthetic aperture radar (InSAR) offers comprehensive observational capabilities by capturing additional angle-dependent information from multiple observation angles. It enables the retrieval of parameters that cannot be obtained from a single-angle observation, such as 3-D deformation. However, variations in observation geometries and platform configurations often lead to noticeable differences in image features, including spatial resolution, texture, and noise level, which present significant challenges for joint processing and consequently degrade the accuracy of parameter retrieval. To address these challenges, a multifeature adaptive enhancement method based on a generative adversarial network (MAE-GAN) is proposed. The network adaptively enhances the InSAR complex-valued image while simultaneously preserving additional angle-dependent information such as phase and scattering characteristics. Specifically, a cross-domain cycle (CDC) adversarial learning is employed to align images from low-quality angles with those from the highest-quality angle in the deep feature space, thereby enhancing the quality of low-quality observations. Furthermore, a fractal loss is introduced to protect additional angle-dependent information by maintaining self-similarity across scales. The experimental results on both simulated multiangle datasets and real multiangle space-surface bistatic synthetic aperture radar (SS-BSAR) measurements demonstrate that MAE-GAN achieves balanced improvements in texture consistency, signal-to-noise ratio (SNR), spatial resolution, and phase accuracy. These results confirm the effectiveness of the proposed method in enhancing the quality and reliability of multiangle InSAR complex-valued images.