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.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is **the speckle noise suppression problem in polarimetric synthetic aperture radar (PolSAR) images**. Specifically, the author aims to develop a convolutional neural network (CNN) framework based on deep learning to effectively remove speckle noise in PolSAR images while maintaining the spatial resolution of the images and the integrity of polarization information. ### Problem Background 1. **Challenges of Speckle Noise**: - Speckle noise is a common phenomenon in SAR images, caused by the interference of multiple scatterers within the SAR resolution cell. - Although speckle noise is not noise in the traditional sense, it will reduce the quality and interpretability of the image, affecting subsequent polarization processing and applications. 2. **Limitations of Existing Methods**: - Traditional speckle suppression methods include multi - look processing, spatial filtering, and transform - domain filtering, etc., but these methods face challenges when applied to polarimetric SAR images. - Deep learning techniques have made significant progress in single - polarization SAR image despeckling, but their application in polarimetric SAR images is still limited, mainly because of the complexity of the polarimetric covariance matrix and the need to preserve its characteristics. ### Core Contributions of the Paper 1. **Proposing a Complete Framework**: - The author proposes a complete framework based on convolutional neural network (CNN) for removing speckle noise in polarimetric SAR images. - This framework includes converting the original complex covariance matrix into a set of real - value intensity bands for input into the neural network for processing. 2. **Introducing a Change Detection Strategy**: - In order to prevent the neural network from learning wrong features in regions with large temporal changes, the author introduces a change detection strategy to ensure that the network only learns the speckle component and is not affected by scene changes. 3. **Experimental Verification**: - This method was tested on Sentinel - 1 dual - polarization images. The experimental results show that this method performs well in speckle suppression and resolution preservation, and does not introduce artifacts or biases in the filtered image. ### Formula Summary 1. **Definition of Polarimetric Covariance Matrix**: \[ C_2=\begin{bmatrix} C_{11}&C_{12}\\ C_{21}&C_{22} \end{bmatrix}=\begin{bmatrix} E\{S_{VV}S_{VV}^*\}&E\{S_{VV}S_{VH}^*\}\\ E\{S_{VH}S_{VV}^*\}&E\{S_{VH}S_{VH}^*\} \end{bmatrix} \] where \( S_{VV} \) and \( S_{VH} \) represent the complex SAR images of the vertical - vertical and vertical - horizontal polarization channels respectively. 2. **Conversion of Intensity Bands**: \[ \begin{cases} \hat{c}_{vv}=E\{|S_{VV}|^2\}\\ \hat{c}_i = E\{|S_{VV}+S_{VH}|^2\}\\ \hat{c}_q = E\{|S_{VH}+jS_{VV}|^2\}\\ \hat{c}_{vh}=E\{|S_{VH}|^2\} \end{cases} \] 3. **Inverse Transformation**: \[ \begin{cases} C_{11}=\hat{c}_{vv}\\ C_{12}=0.5\cdot[(\hat{c}_i - SPAN)+j(\hat{c}_q - SPAN)]\\ C_{21}=0.5\cdot[(\hat{c}_i - SPAN)-j(\hat{c}_q - SPAN)]\\ C_{22}=\hat{c}_{vh} \end{cases} \]

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

Sentinel-1 Dual-Polarization SAR Images Despeckling Network Based on Unsupervised Learning

Adaptive Speckle Filter for Multi-Temporal PolSAR Image with Multi-Dimensional Information Fusion

Exploiting multi-temporal information for improved speckle reduction of Sentinel-1 SAR images by deep learning

An Advanced SAR Image Despeckling Method by Bernoulli-Sampling-Based Self-Supervised Deep Learning

PolMERLIN: Self-Supervised Polarimetric Complex SAR Image Despeckling with Masked Networks

A PolSAR Image Speckle Filtering Method Preserving Point Targets and Dominant Scattering Mechanisms

Speckle2Void: Deep Self-Supervised SAR Despeckling with Blind-Spot Convolutional Neural Networks

SAR2SAR: a semi-supervised despeckling algorithm for SAR images

Speckle Reduction in Polarimetric SAR Imagery with Stochastic Distances and Nonlocal Means

Fast strategies for multi-temporal speckle reduction of Sentinel-1 GRD images

A New Methodology for Assessing SAR Despeckling Filters

SAR Despeckling Using Multiobjective Neural Network Trained with Generic Statistical Samples.

Multi-Frequency Analysis of Fully Polarimetric Speckle Filter Based on Image Morphology for Land Use Classification

SAR Image Despeckling by Deep Neural Networks: from a pre-trained model to an end-to-end training strategy

Multi-temporal speckle reduction with self-supervised deep neural networks

Feature-Based Nonlocal Polarimetric Sar Filtering

SAR Image Colorization: Converting Single-Polarization to Fully Polarimetric Using Deep Neural Networks

SAR Image Despeckling Based on Denoising Diffusion Probabilistic Model and Swin Transformer

A Self-supervised SAR Image Despeckling Strategy Based on Parameter-sharing Convolutional Neural Networks

Convolutional Neural Network with a Learnable Spatial Activation Function for SAR Image Despeckling and Forest Image Analysis