RepDNet:A re-parameterization despeckling network for autonomous underwater side-scan sonar imaging with prior-knowledge customized convolution

Side-scan sonar(SSS)is now a prevalent instrument for large-scale seafloor topography measurements,deployable on an autonomous underwater vehicle(AUV)to execute fully automated underwater acoustic scanning imaging along a predetermined trajectory.However,SSS images often suffer from speckle noise caused by mutual interference between echoes,and limited AUV computational resources further hinder noise suppression.Existing approaches for SSS image processing and speckle noise reduction rely heavily on complex network structures and fail to combine the benefits of deep learning and domain knowledge.To address the problem,Rep DNet,a novel and effective despeckling convolutional neural network is proposed.Rep DNet introduces two re-parameterized blocks:the Pixel Smoothing Block(PSB)and Edge Enhancement Block(EEB),preserving edge information while attenuating speckle noise.During training,PSB and EEB manifest as double-layered multi-branch structures,integrating first-order and secondorder derivatives and smoothing functions.During inference,the branches are re-parameterized into a 3×3 convolution,enabling efficient inference without sacrificing accuracy.Rep DNet comprises three computational operations:3×3 convolution,element-wise summation and Rectified Linear Unit activation.Evaluations on benchmark datasets,a real SSS dataset and Data collected at Lake Mulan aestablish Rep DNet as a well-balanced network,meeting the AUV computational constraints in terms of performance and latency.

PAN-DeSpeck:A Lightweight Pyramid and Attention-Based Network for SAR Image Despeckling

SAR images commonly suffer fromspeckle noise,posing a significant challenge in their analysis and interpretation.Existing convolutional neural network(CNN)based despeckling methods have shown great performance in removing speckle noise.However,these CNN-basedmethods have a fewlimitations.They do not decouple complex background information in amulti-resolutionmanner.Moreover,they have deep network structures thatmay result in many parameters,limiting their applicability tomobile devices.Furthermore,extracting key speckle information in the presence of complex background is also a major problem with SAR.The proposed study addresses these limitations by introducing a lightweight pyramid and attention-based despeckling(PAN-Despeck)network.The primary objective is to enhance image quality and enable improved information interpretation,particularly on mobile devices and scenarios involving complex backgrounds.The PAN-Despeck network leverages domainspecific knowledge and integrates Gaussian Laplacian image pyramid decomposition for multi-resolution image analysis.By utilizing this approach,complex background information can be effectively decoupled,leading to enhanced despeckling performance.Furthermore,the attention mechanism selectively focuses on key speckle features and facilitates complex background removal.The network incorporates recursive and residual blocks to ensure computational efficiency and accelerate training speed,making it lightweight while maintaining high performance.Through comprehensive evaluations,it is demonstrated that PAN-Despeck outperforms existing image restoration methods.With an impressive average peak signal-to-noise ratio(PSNR)of 28.355114 and a remarkable structural similarity index(SSIM)of 0.905467,it demonstrates exceptional performance in effectively reducing speckle noise in SAR images.The source code for the PAN-DeSpeck network is available on GitHub.

A Downsampled SAR-BM3D Despeckling Approach for Single-Look SAR Images in High Resolution

SAR-BM3D is one of the state of the art despeckling algorithms for SAR images. However, when tackling with high resolution SAR images, it often has an unsatisfying despeckling performance in the homogeneous smooth regions, together with a high time complexity. In this paper, a novel downsampled SAR-BM3D despeckling approach combined with edge compensation is proposed. The proposed algorithm consists of two steps. First, despeckle the image which is a downsampled version of original image with SAR-BM3D. Then, compensate edges in each level when upsampling. This approach not only utilizes the good ability of feature preservation, but also improves performance of smoothing homogenous regions. When it comes to high resolution SAR images, the efficiency can be raised by six to seven times, compared to original SAR-BM3D. Experiments on simulated and real SAR images show that the proposed method reaches a high level in terms of visual quality and act more efficiently.

SAR image despeckling with a multilayer perceptron neural network

Speckle noise in synthetic-aperture radar (SAR) images severely hindersremote sensing applications;therefore, the appropriate removal ofspeckle noise is crucial. This paper elaborates on the multilayerperceptron (MLP) neural-network model for SAR image despeckling byusing a time series of SAR images. Unlike other filtering methods thatuse only a single radar intensity image to derive their parameters andfilter that single image, this method can be trained using archivedimages over an area of interest to self-learn the intensitycharacteristics of image patches and then adaptively determine theweights and thresholds by using a neural network for imagedespeckling. Several hidden layers are designed for feedforwardnetwork training, and back-propagation stochastic gradient descent isadopted to reduce the error between the target output and neuralnetwork output. The parameters in the network are automaticallyupdated in the training process. The greatest advantage of MLP is thatonce the despeckling parameters are determined, they can be used toprocess not only new images in the same area but also images incompletely different locations. Tests with images from TerraSAR-X inselected areas indicated that MLP shows satisfactory performance withrespect to noise reduction and edge preservation. The overall imagequality obtained using MLP was markedly higher than that obtainedusing numerous other filters. In comparison with other recentlydeveloped filters, this method yields a slightly higher image quality,and it demonstrates the powerful capabilities of computer learningusing SAR images, which indicate the promising prospect of applyingMLP to SAR image despeckling.

Information compression and speckle reduction for multifrequency polarimetric SAR images based on kernel PCA

Multifrequency polarimetric SAR imagery provides a very convenient approach for signal processing and acquisition of radar image. However, the amount of information is scattered in several images, and redundancies exist between different bands and polarizations. Similar to signal-polarimetric SAR image, multifrequency polarimetric SAR image is corrupted with speckle noise at the same time. A method of information compression and speckle reduction for multifrequency polarimetric SAR imagery is presented based on kernel principal component analysis （KPCA）. KPCA is a nonlinear generalization of the linear principal component analysis using the kernel trick. The NASA/JPL polarimetric SAR imagery of P, L, and C bands quadpolarizations is used for illustration. The experimental results show that KPCA has better capability in information compression and speckle reduction as compared with linear PCA.

Bayesian-Based Speckle Suppression for SAR Image Using Contourlet Transform

A novel and efficient speckle noise reduction algorithm based on Bayesian contourlet shrinkage using contourlet transform is proposed.First,we show the sub-band decompositions of SAR images using contourle transforms,which provides sparse representation at both spatial and directional resolutions.Then,a Bayesian contourlet shrinkage factor is applied to the decomposed data to estimate the best value for noise-free contourle coefficients.Experimental results show that compared with conventional wavelet despeckling algorithm,the proposed algorithm can achieve an excellent balance between suppresses speckle effectively and preserve image details,and the significant information of origina image like textures and contour details is well ma intained.

A non-local vectorial total variational model for multichannel SAR image speckle suppression

Abstract This paper aims at the multichannel synthetic aperture radar （SAR） image speckle reduc- tion. This paper proposes a novel energy minimized regularization model for multichannel image denoising, which is an extension of the non-local total variational model for gray-scale image. It contains two terms, namely the vectorial data fidelity term and the non-local vectorial total variation term. The latter is constructed by high-dimensional non-local gradient that contains the structure information of the multichannel image. The existence and the uniqueness of the solution of the model are proved. A fixed point iterative algorithm is designed to acquire the solution of this model. The convergence property of this algorithm is proved as well. This model is applied to the multipolarimetric and multi-temporal RAI）ARSAT-2 images despeckling. The result shows that this model performs better than the original vectorial total variational model on texture preserving.