Deep Learning Based Speckle Filtering for Polarimetric SAR Images. Application to Sentinel-1

TL;DR

This work tackles speckle removal in polarimetric SAR (PolSAR) by introducing an end-to-end CNN framework that operates on four real-valued intensity bands derived from the complex polarimetric covariance via a bijective Gamma transformation. A change-detection strategy based on the Omnibus test is integrated during training to ignore temporally changing areas, enabling the network to learn the underlying speckle statistics from real data. The approach demonstrates exceptional speckle suppression and resolution preservation on Sentinel-1 dual-pol data, while preserving polarimetric content and enabling reliable downstream processing like model-based polarimetric decomposition. The method yields competitive performance against state-of-the-art despeckling methods, with practical computational costs and strong generalization to unseen scenes, highlighting its potential for operational PolSAR analysis.

Abstract

Speckle suppression in synthetic aperture radar (SAR) images is a key processing step which continues to be a research topic. A wide variety of methods, using either spatially-based approaches or transform-based strategies, have been developed and have shown to provide outstanding results. However, recent advances in deep learning techniques and their application to SAR image despeckling have been demonstrated to offer state-of-the-art results. Unfortunately, they have been mostly applied to single-polarimetric images. The extension of a deep learning-based approach for speckle removal to polarimetric SAR (PolSAR) images is complicated because of the complex nature of the measured covariance matrices for every image pixel, the properties of which must be preserved during filtering. In this work, we propose a complete framework to remove speckle in polarimetric SAR images using a convolutional neural network. The methodology includes a reversible transformation of the original complex covariance matrix to obtain a set of real-valued intensity bands which are fed to the neural network. In addition, the proposed method includes a change detection strategy to avoid the neural network to learn erroneous features in areas strongly affected by temporal changes, so that the network only learns the underlying speckle component present in the data. The method is implemented and tested with dual-polarimetric images acquired by Sentinel-1. Experiments show that the proposed approach offers exceptional results in both speckle reduction and resolution preservation. More importantly, it is also shown that the neural network is not generating artifacts or introducing bias in the filtered images, making them suitable for further polarimetric processing and exploitation.

Figures (9)

• Figure 1: General scheme of the proposed polarimetric speckle filter based on the DnCNN neural network architecture.
• Figure 2: Images of the dataset of Murcia (Spain): (a) single image at a randombly selected date, and (b) temporal average of the whole time series. Representation: RGB false colour composite formed with the backscattering coefficient at the linear channels with $R = VV$, $G = VH$, $B = VV/VH$.
• Figure 3: Omnibus change detection test applied to an excerpt of the datasets over Murcia (top row) and Toscana (orbit 117) (bottom row).
• Figure 4: Filtering results of a Sentinel-1 image acquired over the city of Munich (Germany). Image dimensions are 1500$\times$3000 pixels in azimuth and range, respectively. Representation: RGB false colour composite formed with the backscattering coefficient at the linear channels with $R = VV$, $G = VH$, $B = VV/VH$.
• Figure 5: Filtering results of a Sentinel-1 image acquired over the city of Bay of San Francisco (USA). Image dimensions are 1000$\times$2000 pixels in azimuth and range, respectively. Representation: RGB false colour composite formed with the backscattering coefficient at the linear channels with $R = VV$, $G = VH$, $B = VV/VH$.