Hossein Aghababaei

Translating Synthetic aperture radar (SAR) images into optical images is intrinsically ill-posed because microwave backscatter and optical reflectance describe different physical properties of the observed scene. Although frequency-domain modeling has been introduced into diffusion-based translation, existing methods mainly rely on independent weighting of individual Fourier coefficients and provide limited modeling of interactions among neighboring frequencies and feature channels. To address this limitation, we propose dualCDM, a conditional diffusion model that jointly exploits spatial- and frequency-domain representations. In the diffusion backbone, a spatial-frequency hybrid residual block (SFHRB) combines a spatial convolution branch with complex-valued convolution in the Fourier domain. The complex convolution aggregates neighboring Fourier coefficients across all input feature channels, enabling local cross-frequency and cross-channel modeling, while its response is modulated by the diffusion timestep. In the SAR conditional encoder, an adaptive frequency-domain feature recalibration block (AFFRB) predicts input-dependent real-valued gains from magnitude and trigonometric phase representations of intermediate GRD features. These gains adaptively recalibrate the complex frequency responses without introducing an additional phase shift, while the residual connection preserves the original conditional information. A dual-domain objective further constrains both the predicted diffusion noise and the one-step optical reconstruction in the spatial and frequency domains. We also construct the S1S2 dataset using 16-bit Sentinel-2 reflectance data, retaining the original 0–10,000 value range and including the near-infrared band. Experiments on SEN1-2 and S1S2 show that dualCDM improves radiometric accuracy, spectral consistency, and structural preservation over six representative methods. Paired statistical tests further confirm significant improvements over the strongest competing method across all six evaluation metrics on both datasets.

Tropical forests are a key component of the global carbon cycle and crucial in understanding the amounts of biomass. With plans for upcoming space-borne missions like BIOMASS to monitor forestry, several airborne missions, including TropiSAR and AfriSAR campaigns, have been successfully launched and experimented. Thanks to these campaigns and the availability of SAR and LiDAR date, recently deep learning solution for the forest height estimation have been proposed. Indeed, interesting results have been achieved and have proven the effectiveness of deep learning in accurately estimating forest and ground heights by utilizing the elements of the covariance matrix derived from multichannel SAR data on a pixel-by-pixel basis. The aim of this work is to go beyond the pixel-wise approach and introducing a patch-wise deep learning solution. The idea is to utilize a three-dimensional array composed of elements from the covariance matrix of a patch of multichannel SAR data as its input data. The solution is able to leverage patch-based information and extract spatial features from neighboring pixels’ heights to improve estimation accuracy. In experiments, the training is conducted by considering SAR data as the input and Light Detection and Ranging (LiDAR) values as the ground truth, and results show striking advantages in both performance and generalization ability by leveraging on spatial information.

Synthetic Aperture Radar (SAR) systems capture both amplitude and phase information, producing complex-valued images widely used in Earth observation applications. Among SAR modalities, Polarimetric SAR (PolSAR) and Polarimetric Interferometric SAR (PolInSAR) systems leverage polarimetric and interferometric information, typically represented by a data coherency matrix comprising real-valued diagonal elements and complex-valued off-diagonal elements. Existing deep learning approaches process the entire coherency matrix either through purely real-valued or purely complex-valued networks, which fails to fully exploit its heterogeneous structure. In this paper, we propose a structurally decoupled modeling strategy for coherency matrices, which explicitly separates and processes diagonal and off-diagonal components based on their distinct structural roles. A real-valued network is employed to model the diagonal scattering power components, while a complex-valued network is used to capture the cross-channel correlation, orientation, and coherence structures from the off-diagonal components. This design aligns well with the inherent organization of the SAR coherency matrix, enabling more targeted and effective feature learning. The extracted real and complex features are subsequently fused via a cross-domain enhancement fusion block to achieve robust representation learning. Experiments on DLR's FSAR dataset demonstrate that the proposed method consistently outperforms six state-of-the-art techniques across three different SAR modalities, achieving superior performance in both quantitative accuracy and qualitative robustness.

In the above article [1], the correct reference associated with equation (2) on page 3, left column, is the below paper: 1)P. Imperatore and G. Fornaro, “Joint Phase-Screen Estimation in Airborne Multibaseline SAR Tomography Data Processing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–14, 2024, Art. no. 4412614, doi: 10.1109/TGRS.2024.3446186.

Multi-baseline and tomographic Synthetic Aperture Radar (SAR) data are often affected by phase distortions known as phase screens. These distortions stem either from atmospheric effects or residual errors in platform motion. Calibrating and compensating for the phase screen is crucial to prevent spreading and defocusing in multi-dimensional tomographic imaging. Given the growing interest in artificial intelligence and deep learning, we aim to utilize their potential to develop a phase calibration process for SAR tomographic data. Our proposed framework is based upon a convolutional neural network and generates training patches directly from the tomographic images under consideration, without relying on external references or resources. Once trained, the network effectively estimates phase distortions across the entire image; these are then used to calibrate the tomographic data. Experimental results from AfriSAR and UAVSAR tomographic datasets are included to showcase the effectiveness of the proposed solution.

Satellite images of synthetic aperture radar (SAR) sensors are contaminated by speckles from the coherent imaging mechanism. Although removing or mitigating speckle has been a critical issue for SAR applications, effective reduction continues to be a significant challenge for existing methods when preserving the intricate structures within SAR images. To address this issue, this work proposes a novel conditional diffusion model for SAR despeckling (DiffusionSAR). The new method explicitly learns data distributions by forward diffusion towards multiplicative gamma noise. The logarithmic and Yeo-Johnson (log-Yeo-Johnson) transformation are harnessed in preprocessing for fine-tuning or hybrid training. A prolonging steps technique is suggested in fine-tuning to match the preprocessing. A new synthetic dataset is designed for satellite SAR despeckling. The proposed method is compared with eight state-of-the-art methods using both synthetic and real-world SAR satellite images. The qualitative and quantitative evaluations confirm the effectiveness of the proposed method in structural preservation as well as noise reduction. A fine-tuning experiment using stacked multitemporal data shows the necessity of tine-tuning training in bridging the domain gap when trained with synthetic data and tested with real-world SAR data.

Tropical forests are a key component of the global carbon cycle. With plans for upcoming space-borne missions like BIOMASS to monitor forestry, several airborne missions, including TropiSAR and AfriSAR campaigns, have been successfully launched and experimented. Typical Synthetic Aperture Radar Tomography (TomoSAR) methods involve complex models with low accuracy and high computation costs. In recent years, deep learning methods have also gained attention in the TomoSAR framework, showing interesting performance. Recently, a solution based on a fully connected Tomographic Neural Network (TSNN) has demonstrated its effectiveness in accurately estimating forest and ground heights by exploiting the pixel-wise elements of the covariance matrix derived from TomoSAR data. This work instead goes beyond the pixel-wise approach to define a context-aware deep learning-based solution named CATSNet. A convolutional neural network is considered to leverage patch-based information and extract features from a neighborhood rather than focus on a single pixel. The training is conducted by considering TomoSAR data as the input and Light Detection and Ranging (LiDAR) values as the ground truth. The experimental results show striking advantages in both performance and generalization ability by leveraging context information within Multiple Baselines (MB) TomoSAR data across different polarimetric modalities, surpassing existing techniques.