Differential SAR Interferometric Phase Map Denoising Using Hybrid Convolutional-Attention Network

Abstract

This paper presents a hybrid deep learning approach for speckle noise reduction from differential synthetic aperture radar (SAR) interferometric phase maps. Differential interferometric SAR (DInSAR) is a powerful technique for detecting and quantifying surface deformations; however, the obtained phase maps are corrupted by speckle noise, topographic contributions, and atmospheric artifacts. Effective speckle denoising is crucial for the accurate extraction of the desired deformation information. The proposed method, called Hybrid Convolutional-Attention Network (HCA-Net), combines convolutional neural networks (CNN) and vision transformers (ViT) in a two-stage architecture. First, a modified U-Net model performs initial despeckling of the input DInSAR phase map. Then, the denoised phase map is fed into a Swin Transformer adapted with a masked self-attention mechanism, which further refines the denoising while preserving fine details and discontinuities related to surface deformations. Experimental results on simulated and real DInSAR data demonstrate the effectiveness of our approach, outperforming traditional techniques in terms of quantitative metrics such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and edge preservation index (EPI). Our HCA-Net achieves improvements of 4-6 dB in PSNR, and significantly higher SSIM (0.96) and EPI (0.94) compared to classical methods (0.85 and 0.84 respectively), enabling more reliable deformation measurements for geological and environmental monitoring applications.