DACDM-CR: Discriminative attention and cloud-aware dynamic mamba for SAR-assisted optical data cloud removal

Abstract

Cloud contamination significantly diminishes the potential applications of optical remote sensing images in geosciences, whereas Synthetic Aperture Radar (SAR) images remain unaffected by such interference. Numerous approaches have sought to leverage information from SAR images to restore affected areas in optical images. However, these methods still have room for improvement in fully leveraging the synergistic potential of SAR and optical images while preserving the global consistency of the reconstructed images. This paper proposes a novel SAR-assisted cloud removal network for optical remote sensing images, which comprises two key stages: feature extraction and image reconstruction. The feature extraction stage involves extracting deep features from optical and SAR images, which are then integrated into a Discriminative Attention Feature Interaction (DAFI) module. This enables multimodal feature collaboration, effectively recovering missing textural information in cloud-contaminated regions. In the image reconstruction stage, a Dynamic Cloud-Adaptive MAMBA Gated Spatial-Channel Attention (DMA) module is employed, efficiently reconstructing global contextual information with linear computational complexity while restoring spatial and channel details in cloud-affected areas. To further improve visual quality, this study introduces a multi-scale cloud-adaptive perceptual loss function based on VGG19, specifically targeting cloud-contaminated regions across different scales. The proposed method is validated on the SEN12MSCR dataset and M3M-CR dataset, with experimental results demonstrating superior performance over existing algorithms in terms of peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), spectral angle mapper (SAM), and mean absolute error (MAE).