This paper introduces soil moisture estimation using novel convolutional neural networks with attention mechanism and Unmanned Aerial Vehicle (UAV)-based Synthetic Aperture Radar (SAR) system. The presented system consists of a compact light weight radar using carrier frequency of 5GHz and bandwidth of 600MHz attached to a UAV, capable of generating data in HH and VV polarization. The acquired data in Strip Map mode using a dual channel acquisition are used in experiment for soil moisture estimation using Dubois and Integral Equation Model (IEM) for soil moisture detection, and a custom-designed convolutional network with an attention mechanism for soil moisture estimation. The neural network consists of encoder layered convolutional network using stacked encoder architecture and dual branch attention mechanism. The data are classified into 20 different classes from the dry area with volumetric moisture of 10 % to wet class that represents 50 % of volumetric moisture. The results of neural network for soil moisture estimation are presented using Dubois and IEM model and ground truth measurements. Experimental results showed that the neural network provides results with smallest deviation from the ground truth values.

The despeckling of Synthetic Aperture Radar images using two different convolutional neural networks architectures is presented in this paper. The first method presents a dilated convolutional network inserted into Siamese convolutional neural network. Recently, attention mechanism have been introduced to convolutional networks to better model and recognize features. Therefore, we propose a design of convolutional neural network using attention mechanism for encoder-decoder type of network. The framework consists of multi scale spatial attention network to improve modeling of semantic information at different spatial level and additional attention mechanism to optimize feature propagation. Both proposed methods are different in design, but they provided very comparable despeckling results in subjective and objective measurements, when correlated speckle noise was concerned. The experimental results were evaluated on synthetically generated speckled images and real SAR images. Proposed methods in this paper are able to despeckle SAR images and preserve SAR features.

This paper presents an echo cancellation (EC) algorithm for ground penetrating radar (GPR) that effectively suppresses multiple static targets while maintaining visibility of dynamic targets. In continuous wave (CW) radars like the SFCW or FMCW radars, where transmitting time is much longer than the echo response time, all echoes are summed together and cannot be separated once received by the antenna. The proposed EC system generates an EC signal, matched in amplitude and with a phase difference of π radian, to physically cancel the undesired static echoes. The system is particularly useful for applications such as GPR and through-wall imaging, where static echoes are problematic. The paper discusses the simulation of the static target echo cancellation algorithm and presents implementation results. The EC system can be calibrated to adapt to changes in the static scene, ensuring optimal performance. Overall, the EC algorithm offers a practical solution for mitigating static echoes in GPR systems and similar applications.

In this paper we present 3D convolutional networks, which represent a computational formulation of spatial-temporal 3D CNNs for extracting both, spatial and temporal features at the same time. The convolutional neural networks are a type of deep models that can extract features directly from the SAR data. In processing of a single polarimetric SAR image, a regression framework for soil moisture estimation has been a proven concept for soil moisture estimation. In this paper we extracted a regression feature extraction to 3D deep regression network for soil moisture detection using single or a temporal series of SAR data. 4 acquisition of the same scene using L band SAR data were used to estimate the soil moisture. The model showed capabilities of soil moisture extraction with 40% accuracy compared to the ground measurements.

In this letter, a model-based approach to information extraction and despeckling is presented using a maximuma posteriori(MAP) estimate. An autobinomial model (ABM) is used as a prior and Nakagami model for a likelihood probability density function (pdf). ABM parameter estimation using the evidence maximization makes this method very computationally demanding and unuseful for practical applications. This letter proposes a deep convolutional neural network (CNN) for parameter estimation of ABM. A nine-layer CNN was used, consisting of convolutional, pooling, dropout, fully connected, and regression layers. The MAP estimation and CNN interactions were fused to obtain the best texture parameters with the highest evidence. Experimental results showed that the ABM using evidence maximization is 55 times slower than the ABM with CNN. Texture parameters are estimated better with the proposed technique because the estimated evidence obtained with the proposed method is much higher compared with the previous method. The improved parameter estimates subsequently used for MAP despeckling purposes provided favorable results.

This paper presents a visualization SAR patch categorization using complex valued deep learning approach. SAR patch categorization of a SAR product provides image categories with a sharp border between categories. This paper presents visualization of SAR categories using gradient weighted class activation mapping of final convolutional layer in order to produce a coarse localization map highlighting important region. The reconstructed categorization provides visually smoother categorization than the classical patch based deep learning categorization.

This paper presents despeckling of Synthetic Aperture Radar (SAR) detected data using deep convolutional networks. A discriminative model learning using a deep convolutional neural network (DCNN) was used. A DCNN was used to learn speckle statistical properties to process SAR data. The idea is to use two identical subnetworks, very similar to Siamese deep neural networks , which consisted of 15 layers. The network used a residual learning strategy using a large SAR database. SAR images were multilooked in order to model a noiseless SAR images. Experimental results demonstrated promising results using synthetic and real images.

This paper presents Synthetic Aperture Radar (SAR) patch categorization using unsupervised feature learning framework. It is based on layer based sparse coding, which extends a sparse coding to a multi layer architecture. A contribution of this paper is a framework which consists of 3 layers of sparse coding, local spatial pooling layer, normalization layer, map reduction layer and a classification layer. The new method is able to learn several levels of sparse representation of the image which capture features at a variety of abstraction levels and simultaneously preserve the spatial smoothness between the neighboring image patches. The proposed method achieved promising results in SAR patch categorization.

This letter proposes a change detection algorithm for damage assessment caused by fires in Ireland using Sentinel 1 data. The novelty, in this letter, is a feature extraction within tunable Q discrete wavelet transform (TQWT) using higher order log cumulants of fractional Fourier transform (FrFT), which were fed into a stacked autoencoder (SAE) to distinguish changed and unchanged areas. The extracted features were used to train the SAE layerwise using an unsupervised learning algorithm. After training the decoding layer was replaced by a logistic regression layer to perform supervised fine-tuning and classification. The proposed algorithm was compared with the algorithm that used log cumulants of FrFT within the oriented dual-tree wavelet transform using support vector machine (SVM) classifier. The experimental results showed that the proposed combination of algorithms decreased the overall error (OE) for real synthetic aperture radar images by 6%, when TQWT was used instead of oriented dual-tree wavelet transform and OE was decreased by another 5% when SAE was used instead of the SVM classifier.

Synthetic aperture radar (SAR) images are affected by a speckle noise, which is a consequence of random fluctuations in the return signal from an object that is no bigger than a single image processing element and it is caused by coherent processing of backscattered signals from multiple distributed targets. Speckle within SAR images can be reduced using filtering methods. To preserve features within the SAR images, this paper proposes a noise removal based on scene and SAR data modeling. The proposed method is a model-based total variational optimization with the minimization of a cost function. The cost function consisted of energy and data fidelity terms. The energy term was modeled using optimal-dual-based l1analysis. The data fidelity term modeled the amplitude of the SAR data, which was approximated using a Nakagami distribution. The minimization of the cost function was solved using a quasi-Newton approach. The experimental results showed good results in SAR feature preservation. The proposed method was evaluated numerically using quality metrics for synthetic generated data and real amplitude SAR data.

This letter presents sparse coding for interferometric synthetic aperture radar (InSAR) patch categorization. Motivated by the fact that an optimal dual based l1analysis can achieve better recognition rates, this letter proposes sparse coding with optimal dual-based l1analysis, which is applied to the amplitude and phase of the InSAR patches. The minimization of cost functions for amplitude and phase was designed and solved differently. The cost function for the amplitude part of InSAR data was modeled using the optimal dual-based l1analysis, and the minimization of cost function was solved using the forward-backward splitting algorithm. The phase was coded sparsely using the l1minimization approach and it was solved using the gradient descent algorithm. The experimental results showed that the proposed method outperforms the complex-valued methods for SAR patch categorization and outperforms the bag of visual words method as well.

This paper presents change detection using very high resolution SAR data. Small patches of SAR data were used for graphical Lasso based algorithm. The graphical Lasso for time series is defined as solution of an l1-regularized maximum likelihood problem. The optimization problem was solved using alternating direction method of multipliers (ADMM). The time series of patches was observed. The efficiency of change detection of the ADMM algorithm was observed using different window sizes. The time series of 10 interferometric SAR Very high resolution Spotlight data was used, which were acquired over period of 3 years. Experimental results showed that ADMM can be successfully applied to the time series in order to detect changes using complex valued SAR data.

This paper presents information extraction from very high resolution SAR data. Small patches of SAR data were used for graph based regularized sparse coding. The auto regressive model was used to extract sparse graphs from the data. The sparse graph features were used for dictionary learning using Graph-regularized Non-negative Matric Factorization. The algorithm was used categorization of database consisted of 20 classes and each class contained 100 patches. The algorithm was not superior to the supervised categorization using spectral features.

We propose a new compressed sensing MRI approach that uses the discrete nonseparable shearlet transform (DNST) as a sparsifying transform and the fast iterative soft thresholding algorithm (FISTA) for reconstruction. FISTA has a simple design and has shown good convergence behavior. The DNST transform has excellent localization properties within the space domain and excellent directional selectivity. We utilize the frequency representation of the DNST canonical dual filters to obtain a memory efficient modified FISTA based algorithm with a simple and efficient way of calculating the update, tuned to the non tight frame DNST transform. The proposed approach shows improved performance and similar execution time when compared with other state of the art reconstruction approaches.

This paper presents a categorization of SAR patches using supervised approach within a dictionary learning sparse representation framework. Dictionary learning algorithms represent matrix factorization of data matrix X as the product of Dictionary and sparse coefficients Z. The dictionary learning algorithm was implemented using well known K-SVD algorithm. The trained dictionaries were used for sparse representation and classification. Experimental results showed superior results for Dictionary-Learning Sparse Representation framework for categorization of SAR patches.

This letter presents synthetic aperture radar (SAR) image classification based on feature descriptors within the discrete wavelet transform (DWT) domain using parametric and nonparametric features. The DWT enables an efficient multiresolution description of SAR images due to its geometric and stochastic features. A 2-D DWT, a real 2-D oriented dual tree wavelet transform (2-D RODTWT) and an oriented dual tree complex wavelet transform (2-D ODTCWT) were used for the estimation of subband features. First and second moments, entropy, coding gain, and fractal dimension were used for the nonparametric approach. A parametric approach considers a Gauss Markov Random Field model for feature extraction. A database with 2000 images representing 20 different classes with 100 images per class was used for classification efficiency assessment. Several SAR scenes were divided into small patches with dimension of 200 × 200 pixels. 10% and 20% of the test images per class were used during the learning stage. Supervised learning using a support vector machine was used for all experiments. The experimental results showed that the proposed methods had superior performances compared with (GLCM) and log comulants of Fourier transform. Amongst the proposed methods, the nonparametric features within oriented dual tree complex wavelet transform gave the best results for classes when categorizing SAR images.

This paper presents an information extraction and image enhancement technique using single-look complex (SLC) synthetic aperture radar data. The novelty of this method is the proposed complex-domain despeckling stage. Tikhonov-like optimization is used for minimizing the cost function, which consists of a Gauss-Markov random field (GMRF) prior. The GMRF model is used for texture modeling. The texture parameters of the GMRF are estimated using the evidence maximization framework. The experimental results showed that despeckled SLC images have well-preserved textural features, structures, and point scatterers. The phase of the reconstructed image is well preserved and provides good-quality interferograms of high-resolution spotlight images.

This paper presents SAR image classification based on feature descriptors within the dual tree oriented discrete wavelet transform. The non-parametric approach to the feature extraction and supervised learning is presented in this paper. The spectral features, known from sound processing were used for subband features within the wavelet domain. Those features characterizing each subband of oriented wavelet transform were used for supervised classification using support vector machine. The database with 1300 images with 200 × 200 pixels was designed using 30 different high resolution TerraSAR-X spotlight images. 10 percent of all features for each class were used for training. The efficiency of presented method was compared with Gray Level Co-occurrence Matrix (GLCM) method and log commulants of Fourier transform. The experimental results showed improved classification results compared to the state-of-the-art methods used in this paper.

This paper introduces soil-moisture parameter retrieval using high-resolution vertically polarized (VV) Spotlight TerraSAR-X data. The soil-moisture estimation of bare and vegetated areas is considered by using volumetric scattering, which is modeled with a bare-soil component and a component reflecting vegetation. The unknown coefficients of the soil-moisture model are estimated using the Tikhonov regularization scheme. A neural network is used in order to distinguish volumetric scattering from all the other types of scattering. The estimated volumetric-soil-moisture parameters are further enhanced by using a supervised feedforward backpropagation neural network. The proposed algorithm based on the Tikhonov regularization scheme, in combination with neural networks, provides good results for estimating volumetric-soil-moisture in an area covered with a small vegetation canopy.

This paper presents a Bergman iteration of SLC Synthetic Aperture Radar image despeckling and information extraction. A split Bergman iteration for solving minimization of cost function is presented in this paper and the information extraction using Auto-binomial model is incorporated into the algorithm. Experimental results showed that parameters of ABM well characterize the spatial characteristics of SAR SLC data.

Speckle hinders information in synthetic aperture radar (SAR) images and makes automatic information extraction very difficult. The Bayesian approach allows us to perform the despeckling of an image while preserving its texture and structures. This model-based approach relies on a prior model of the scene. This paper presents an evaluation of two despeckling and texture extraction model-based methods using the two levels of Bayesian inference. The first method uses a Gauss–Markov random field as prior, and the second is based on an auto-binomial model (ABM). Both methods calculate a maximum a posteriori and determine the best model using an evidence maximization algorithm. Our evaluation approach assesses the quality of the image by means of the despeckling and texture extraction qualities. The proposed objective measures are used to quantify the despeckling performances of these methods. The accuracy of modeling and characterization of texture were determined using both supervised and unsupervised classifications, and confusion matrices. Real and simulated SAR data were used during the validation procedure. The results show that both methods enhance the image during the despeckling process. The ABM is superior regarding texture extraction and despeckling for real SAR images.

This paper proposes a method using non-quadratic regularization. The Auto-binomial model (ABM) is used for a prior. The texture parameters of ABM are estimated using the evidence maximization framework. The regularization parameters are kept constant during despeckling. The experimental results showed that the SLC images can be despeckled using the non-quadratic regularization, because it is an iterative procedure. The textural parameters of the ABM well describes the SAR scene and well separates the textural properties of the scene, when texture parameters are classified using an unsupervised classification.

High-resolution and dual polarized Spotlight TerraSAR-X images are assessed for soil moisture parameter retrieval. This letter presents bare soil moisture estimation and estimation of moisture of vegetated areas. The bare soil moisture estimation is based on the Shi model. The Minimum Mean Square Error approach is used to determine the unknown parameters of the Shi model using ground measurements of volumetric moisture and SAR data. The soil moisture of vegetated areas is estimated using the vegetation and soil backscattering coefficients. The unknown parameters of vegetation and soil backscattering models were estimated using Tikhonov optimization. The experimental results showed that the used models provide good results for estimating bare soil moisture and moisture of vegetated areas.

Synthetic aperture radar (SAR) images are affected by multiplicative noise called speckle. This noise makes automatic image classification and image interpretation difficult. Thus, many methods have been developed to remove speckle from SAR images while preserving the useful information of the scene such as texture and geometry. In this letter, a comparison between three different despeckling methods based on a Bayesian approach and Gibbs random fields is made. The used methods are Gauss–Markov random field (GMRF) and autobinomial modeling, which operate in the image domain, and the GMRF approach, which operates in the wavelet domain. Our methods are evaluated with synthetic and real SAR data (TerraSAR-X images). The experimental results show that, with these three methods, the speckle is well removed while structures are preserved; quantitative measures show that the autobinomial method provides the best smoothness and sharpness criteria in real SAR data, while the wavelet-based method generates the smallest bias.

This letter presents the despeckling of synthetic aperture radar (SAR) images within the bandelet and contourlet domains. A model-based approach is presented for the despeckling of SAR images. The speckle-reduced estimate is found using the first-order Bayesian inference, and the best model's parameters are estimated using the second-order Bayesian inference. Synthetic and real images are used for evaluating the qualities of the despeckling methods. The experimental results showed that the combination of Bayesian inference and bandelet transform outperforms the contourlet-based despeckling algorithm using synthetic data and objective measurements.

This letter presents the despeckling of single-look complex (SLC) synthetic aperture radar (SAR) images using nonquadratic regularization. The objective function consists of an image model, a gradient, and a prior model. The Huber–Markov random field (HMRF) models the prior. A numerical solution is achieved through extensions of half-quadratic regularization methods using complex-valued SAR data. The proposed method using the HMRF prior together with nonquadratic regularization shows the superior results on SLC synthetic and actual SAR images.

This paper presents a model-based despeckling (MBD) of synthetic aperture radar (SAR) images using Bayesian analysis. The SAR image is despeckled using first-order Bayesian inference. The novelty in this paper is an autobinomial model (ABM), which models apriorprobability density function (pdf); meanwhile, thelikelihoodpdf is modeled as a gamma distribution. Analytically, a solution for a maximuma posterioriestimate using an autobinomial prior cannot be computed; therefore, an approximation is introduced using differential. The best ABM for approximating the texture parameters in SAR images is found by using second-order Bayesian inference. The edges in the SAR images are detected using region borders, which have statistically different properties. Coefficient of variation is used to distinguish between homogeneous and heterogeneous areas. The experimental results show that the proposed method preserves the textural features and removes noise significantly in the homogeneous and heterogeneous regions. The proposed despeckling method is good regarding objective measures for synthetic images and better despeckles the real SAR images, when compared with the state-of-the-art MBD methods.

This paper proposes a new-wavelet-based synthetic aperture radar (SAR) image despeckling algorithm using the sequential Monte Carlo method. A model-based Bayesian approach is proposed. This paper presents two methods for SAR image despeckling. The first method, called WGGPF, models a prior with Generalized Gaussian (GG) probability density function (pdf) and the second method, called WGMPF, models prior with a Generalized Gaussian Markov random field (GGMRF). The likelihood pdf is modeled using a Gaussian pdf. The GGMRF model is used because it enables texture parameter estimation. The prior is modeled using GG pdf, when texture parameters are not needed. A particle filter is used for drawing particles from the prior for different shape parameters of GG pdf. When the GGMRF prior is used, the particles are drawn from prior in order to estimate noise-free wavelet coefficients and for those coefficients the texture parameter is changed in order to obtain the best textural parameters. The texture parameters are changed for a predefined set of shape parameters of GGMRF. The particles with the highest weights represents the final noise-free estimate with corresponding textural parameters. The despeckling algorithms are compared with the state-of-the-art methods using synthetic and real SAR data. The experimental results show that the proposed despeckling algorithms efficiently remove noise and proposed methods are comparable with the state-of-the-art methods regarding objective measurements. The proposed WGMPF preserves textures of the real, high-resolution SAR images well.

In this paper we present two algorithms for information extraction from Single Look Complex (SLC) Synthetic Aperture Radar (SAR) images. The first algorithm is based on Tikhonov regularization with Total Variation (TV) and a Point-Based Feature (PBF) term. Based on the equivalence of Tikhonov and the Bayesian estimate, the second algorithm is a Maximum A Posteriori (MAP) estimation with a complex-valued Gauss-Markov Random Field (GMRF) in addition to the TV prior. The first algorithm produces a despeckled image preserving fine details and texture. The second algorithm gives a denoised image and in addition the estimated feature parameter vector θ.

In this paper, a wavelet-based speckle-removing algorithm is represented and tested on synthetic aperture radar (SAR) images. The SAR image is first transformed using a dyadic wavelet transform. The noise in the wavelet-transformed image is modeled as an additive signal-dependent noise with Gaussian distribution. The distribution of a noise-free image in a wavelet domain is modeled as a generalized Gauss–Markov random field (GGMRF). An unsupervised stochastic model-based approach to image denoising is represented. If the observed area is homogeneous, the parameters of the Gaussian distribution and GGMRFs are estimated from incomplete data using mixtures of wavelet coefficients. An expectation–maximization algorithm is used to estimate the parameters of both noisy and noise-free images. The unknown parameters are estimated using image and noise models that are defined in the wavelet domain for heterogeneous areas. Different inter- and intrascale dependences of wavelet coefficients were used to estimate the unknown parameters. The represented wavelet-based method efficiently removes noise from SAR images.

This letter presents synthetic aperture radar (SAR) image despeckling using dyadic wavelet transform. Maximum a posteriori (MAP) estimation is used to despeckle a SAR image in the wavelet domain. A wavelet transformed speckle-free image is approximated with a Gauss–Markov random field, and a Gaussian model is chosen to approximate speckle in the wavelet domain. A speckle-free wavelet coefficient is estimated with Bayesian inference using image and noise model parameters, which produce the highest evidence. The experimental results showed that the despeckling algorithm removes speckle noise in the homogeneous areas better than the state-of-the-art methods, which operate in the wavelet and image domain. The proposed method is very simple and computationally not demanding.

In this paper a wavelet based method for SAR data denoising and compression is presented. An unsupervised stochastic model based approach to image denoising is presented. SAR image is modeled in wavelet domain Gauss Markov random field and noise is considered as Gaussian with unknown variance. The parameters are estimated from incomplete data using mixtures of wavelet coefficients, and expectation maximization algorithm. The expectation maximization algorithm is used to efficiently compute a maximum a posteriori estimate. Observed wavelet coefficient is estimated using inter and intra scale of wavelet coefficients to estimate image and noise model parameters. Presented wavelet based method efficiently removes noise from SAR images. The second step is to design an entropy coder that efficiently codes despeckled image. The texture parameters obtained at the despeckling stage are used in the compression process. The image coder is tested on X-SAR data with and achieves comparable compression results with the wavelet based state-of-the art coders for SAR data compression.

In this letter, we propose a new image coding technique, which is a combination of space frequency quantization and a context-based modeling using fuzzy logic. The compression results showed that the proposed coder outperforms the state-of-the-art coders in the rate-distortion sense for compression of processed synthetic aperture radar amplitude data.

In this paper we propose a new image coding technique, which consists of the space frequency quantization (SFQ), and a fuzzy context based modeling. The space frequency quantization (SFQ) is an effective coding technique, which uses a zerotree pruning of wavelet coefficients. After the space frequency quantization is applied to the wavelet coefficients the non-pruned coefficients are quantized using context modeling. The context based quantization technique uses the partitioning proposed by the trellis coded quantization and fuzzy logic to predict the next TCQ state from the context. Quantized indices are entropy coded using an arithmetic coder. The probability estimation is based on the observation of the past coded bits. The results of the proposed compression scheme showed that the proposed coder outperforms the state-of-the-art coders in the RD sense for the SAR data compression.

The authors propose a new wavelet image coding technique for synthetic aperture radar (SAR) data compression called a progressive space-frequency quantization (PSFQ). PSFQ performs spatial quantization via rate distortion-optimized zerotree pruning of wavelet coefficients that are coded using a progressive subband coding technique. They compared the performances of zerotree-based methods: EZW, SPIHT, SFQ, and PSFQ with the classical wavelet-based method (CWM), which uses uniform scalar quantization of subbands followed by recency rank coding. The performances of the methods based on zerotree quantization were better than the CWM in the rate distortion sense. The embedded coding techniques perform better SNR results than the methods using scalar quantization. However, the probability density function (PDF) of the reconstructed amplitude SAR data compressed using CWM, better corresponded to the PDF of the original data than the PDF of the reconstructed data compressed using the zerotree based methods. The amplitude PDF of the reconstructed data obtained using PSFQ compression algorithm better corresponded to the original PDF than the amplitude PDF of the data obtained using the multilook method.