Jingzhen Zhu

Speckle removal is a crucial preliminary step for synthetic aperture radar (SAR) image processing. In recent years, the application of deep neural networks towards solving SAR image despeckling problems has yielded commendable outcomes. However, prevailing deep learning methods for SAR image despeckling rely on convolutional neural network (CNN) architectures, which inherently capture only local information within their receptive fields. Consequently, the despeckling performance can be further improved through advanced network structure design. More recently, the Transformers based methods show impressive performance in natural image denoising relying on the long-range dependency modeling capability of the self-attention mechanism. In this paper, we introduce a practical despeckling network that incorporates the global modeling capability of the Swin Transformers and the local modeling ability of the residual CNNs. Specifically, the proposed despeckling network is based on the widely-used U-Net architecture, wherein a Swin-Conv (SC) block is adopted to replace the convolutional layer in the baseline UNet. The SC block mainly comprises a Residual Convolutional (RConv) block and a Swin Transformer (SwinT) block, which are used to extract local features and long-range dependencies from images respectively. Furthermore, to deal with the spatially correlated real SAR speckle, a Pixel-shuffle Down-sampling (PD) post-processing strategy is adopted, which can significantly improve the practicality of the proposed method without additional real dataset for fine-tuning. Experimental results demonstrate that the proposed method achieves state-of-the-art performance on both synthetic and real SAR images, and outperforms CNN-based and Transformer-based methods (L=1) by an average PSNR of 0.94dB and 1.58dB, respectively.

Speckle suppression is a crucial preliminary step for synthetic aperture radar (SAR) image processing. Supervised despeckling approaches trained on synthetic datasets usually perform poorly in practice due to the unavailability of clean SAR images. Besides, the spatial correlation of speckle is rarely considered in many methods based on the fully developed speckle assumption. In this paper, we propose an unsupervised despeckling method to address these issues by combining online speckle generation and unpaired training. The method consists of two branches: the stop-gradient branch and the unpaired branch. First, the stop-gradient branch learns to generate the spatially correlated speckle. Then, the unpaired branch combines the generated speckle with the unpaired optical image to form pairs of training data for network parameter updates. More specifically, in order to generate the more realistic speckle in the stop-gradient branch, we design a speckle correction module with three SAR speckle priors: the threshold prior, the unit mean prior, and the correlation prior coupled with the weighted patch-shuffle. In the unpaired training, a hybrid loss function is employed, which takes spatial smoothness and detail protection into consideration. Afterwards, we combine the stop-gradient branch with the unpaired branch by the Siamese network to achieve alternate optimization of speckle generation and speckle removal. Finally, the optimization process in our method is analyzed theoretically. Qualitative and quantitative experiments demonstrate that the proposed method is comparable to the supervised despeckling approaches on synthetic datasets and outperforms several state-of-the-art unsupervised methods on real SAR datasets.

The frequency modulation continuous-wave (FMCW) radar has been widely concerned for its compact system structure, low hardware cost, and excellent imaging resolution. However, an FMCW radar suffer from the nonlinearity commonly, which leads to the deterioration of the imaging quality. A challenging problem is that the nonlinear reference beat signal contains multiple components with closely spaced and approximately equal amplitude, which has not been researched in the existing literature to the best of our knowledge. To deal with this trouble, a software-based correction method is proposed by utilizing the complementary ensemble empirical mode decomposition (CEEMD). First, several intrinsic model functions (IMFs) are generated by decomposing the beat signal obtained from a reference target by the CEEMD method. Then, an identification method is proposed to find the reference IMF from the IMFs by average period method, and a reference delay estimation method based on windowed fast Fourier transform is proposed. Finally, the higher-order ambiguity function (HAF) is employed to estimated the polynomial coefficients of the reference IMF, which are utilized to compensate the nonlinear beat signal by time resampling method. The effectiveness and superiority are verified by numerical simulation and real data. Compared with the state-of-the-art nonlinear correction algorithms, we obtain more outstanding radar imaging quality by adopting the proposed method.