Classification Of Synthetic Aperture Radar Data Research Articles

SAR Image Fusion Classification Based on Improved D-S Evidence Theory

Applications for classifying Synthetic Aperture Radar (SAR) images are critical to environmental monitoring, urban planning, and land resource surveying. Fusion approaches work well for increasing SAR image categorization accuracy. However, effective processing of uncertainty information is frequently overlooked by conventional fusion classification systems. This paper presented a fusion classification approach for SAR images based on improved Dempster-Shafer (D-S) evidence theory, taking into account the uncertainty and possible conflicts in the classifiers’ output. First, after adaptively identifying the optimal hyperparameters of models, the SAR data is identified using Gradient Boosting Machine (GBM), Multilayer Perceptron (MLP), and Random Forest (RF). Different weights are then allocated in accordance with the variations in classifier performance. Ultimately, enhanced D-S evidence theory is used to integrate the output of these classifiers. In addition, uncertainty in classification results is also modeled and visualized. Experiments with Flevoland and Oberpfaffenhofen datasets show that the accuracy of this method is 85.21% and 91.88% respectively, outperforming both the soft voting ensemble classifier (SVE) and D-S evidence theory fusion method. This approach enhances the comprehension and capacity to manage uncertainty in the model output in addition to increasing the accuracy and reliability of the classification of SAR data.

Fusion and Classification of SAR and Optical Data Using Multi-Image Color Components with Differential Gradients

This paper proposes a gradient-based data fusion and classification approach for Synthetic Aperture Radar (SAR) and optical image. This method is used to intuitively reflect the boundaries and edges of land cover classes present in the dataset. For the fusion of SAR and optical images, Sentinel 1A and Sentinel 2B data covering Central State Farm in Hissar (India) was used. The major agricultural crops grown in this area include paddy, maize, cotton, and pulses during kharif (summer) and wheat, sugarcane, mustard, gram, and peas during rabi (winter) seasons. The gradient method using a Sobel operator and color components for three directions (i.e., x, y, and z) are used for image fusion. To judge the quality of fused image, several fusion metrics are calculated. After obtaining the resultant fused image, gradient based classification methods, including Stochastic Gradient Descent Classifier, Stochastic Gradient Boosting Classifier, and Extreme Gradient Boosting Classifier, are used for the final classification. The classification accuracy is represented using overall classification accuracy and kappa value. A comparison of classification results indicates a better performance by the Extreme Gradient Boosting Classifier.

Complex-Valued End-to-End Deep Network With Coherency Preservation for Complex-Valued SAR Data Reconstruction and Classification

Deep learning models have achieved remarkable success in many different fields and attracted many interests. Several researchers attempted to apply deep learning models to Synthetic Aperture Radar (SAR) data processing, but it did not have the same breakthrough as the other fields, including optical remote sensing. SAR data are in complex domain by nature and processing them with Real-Valued (RV) networks neglects the phase component which conveys important and distinctive information. A Complex-Valued (CV) end-to-end deep network is developed in this study for the reconstruction and classification of CV-SAR data. Azimuth subaperture decomposition is utilized to incorporate physics-aware attributes of the CV-SAR into the deep model. Moreover, the correlation coefficient amplitude (Coherence) of the CV-SAR images depends on the SAR system characteristics and physical properties of the target. This coherency should be considered and preserved in the processing chain of the CV-SAR data. The coherency preservation of the CV deep networks for CV-SAR images, which is mostly neglected in the literature, is evaluated in this study. Furthermore, a large-scale CV-SAR annotated dataset for the evaluation of the CV deep networks is lacking. A semantically annotated CV-SAR dataset from Sentinel-1 Single Look Complex StripMap mode data (S1SLC_CVDL dataset) is developed and introduced in this study. The experimental analysis demonstrated the better performance of the developed CV deep network for CV-SAR data classification and reconstruction in comparison to the equivalent RV model and more complicated RV architectures, as well as its coherency preservation and physics-aware capability.

On the Role of Polarimetric Decomposition and Speckle Filtering Methods for C-Band SAR Wetland Classification Purposes

Previous wetlands studies have thoroughly verified the usefulness of data from synthetic aperture radar (SAR) sensors in various acquisition modes. However, the effect of the processing parameters in wetland classification remains poorly explored. In this study, we investigated the influence of speckle filters and decomposition methods with different combinations of filter and decomposition windows sizes on classification accuracy. We used a C-band Radarsat 2 image acquired over a wetland located in northeast Poland. We processed the SAR data using various speckle filters: boxcar, intensity-driven adaptive-neighborhood (IDAN), improved Lee sigma, refined Lee (in 5×5 to 11×11 pixel window sizes), and a nonlocal NL-SAR. Next, we processed the nonfiltered and filtered data using nine polarimetric decompositions, also in 5×5 to 11×11 pixel window sizes. The extracted polarimetric features were applied as an input dataset in the random forest classification model in single- and multidecomposition scenarios. In the single-decomposition scenario, the Cloude–Pottier decomposition produced the highest (72%) and the Touzi decomposition achieved the lowest (38%) accuracy. The IDAN filter with an 11×11 filter window and a 9×9 decomposition window had the highest, and the nonfiltered data with a 5×5 decomposition window had the lowest accuracy in the multidecomposition scenario. The most important features were the alpha parameter from the Cloude–Pottier decomposition, the polarimetric contribution of the Shannon entropy, and the volume backscattering components. The results stress the importance of appropriate processing parameters in the SAR data classification workflow, and the study guides in selecting the most suitable combination of radar image processing parameters for wetland classification.

SAR image classification using adaptive neighborhood-based convolutional neural network

ABSTRACT The convolutional neural network (CNN)-based pixel-wise synthetic aperture radar (SAR) data classification does not take fully use the spatial neighborhood information due to the fact that the impact of neighborhood pixels is not taken into consideration. The flaw of CNN-based classification method may lead to misclassification under some conditions. In this paper, we propose a novel adaptive neighborhood-based convolutional neural network (AN-CNN) for the single polarimetric synthetic aperture radar data classification. In the convolution layer, the neighborhood pixels are adaptively weighted based on their bilateral distance (spatial and feature distance) to the central pixel. In this way, different pixels have different impact on the classification result of the central pixel. The spatial distance-based weighting can reduce the misclassifications in the homogenous regions which are caused by speckle noise and the feature distance-based weighting is beneficial for the classification in the boundary regions. As a result, the misclassification is obviously reduced by the proposed AN-CNN which has a new cost function. Experimental results on simulated and real SAR data show that our proposed AN-CNN can notably improve the classification accuracy in both boundary regions and homogeneous regions compared with conventional CNN in different scenes especially when limited training samples are explored.

River Levels Derived with CryoSat-2 SAR Data Classification—A Case Study in the Mekong River Basin

In this study we use CryoSat-2 SAR (delay-Doppler synthetic-aperture radar) data in the Mekong River Basin to estimate water levels. Compared to classical pulse limited radar altimetry, medium- and small-sized inland waters can be observed with CryoSat-2 SAR data with a higher accuracy due to the smaller along track footprint. However, even with this SAR data the estimation of water levels over a medium-sized river (width less than 500 m) is still challenging with only very few consecutive observations over the water. The target identification with land–water masks tends to fail as the river becomes smaller. Therefore, we developed a classification approach to divide the observations into water and land returns based solely on the data. The classification is done with an unsupervised classification algorithm, and it is based on features derived from the SAR and range-integrated power (RIP) waveforms. After the classification, classes representing water and land are identified. Better results are obtained when the Mekong River Basin is divided into different geographical regions: upstream, middle stream, and downstream. The measurements classified as water are used in a next step to estimate water levels for each crossing over a river in the Mekong River network. The resulting water levels are validated and compared to gauge data, Envisat data, and CryoSat-2 water levels derived with a land–water mask. The CryoSat-2 water levels derived with the classification lead to more valid observations with fewer outliers in the upstream region than with a land–water mask (1700 with 2% outliers vs. 1500 with 7% outliers). The median of the annual differences that is used in the validation is in all test regions smaller for the CryoSat-2 classification results than for Envisat or CryoSat-2 land–water mask results (for the entire study area: 0.76 m vs. 0.96 m vs. 0.83 m, respectively). Overall, in the upstream region with small- and medium-sized rivers the classification approach is more effective for deriving reliable water level observations than in the middle stream region with wider rivers.

Land cover classification of RADARSAT-2 SAR data using convolutional neural network

In this paper, we propose a convolutional neural network (CNN) based on deep learning method for land cover classification of synthetic aperture radar (SAR) images. The proposed method consists of convolutional layers, pooling layers, a full connection layer and an output layer. The method acquires high-level abstractions for SAR data by using a hierarchical architecture composed of multiple non-linear transformations such as convolutions and poolings. The feature maps produced by convolutional layers are subsampled by pooling layers and then are converted into a feature vector by the full connection layer. The feature vector is then used by the output layer with softmax regression to perform land cover classification. The multi-layer method replaces hand-engineered features with backpropagation (BP) neural network algorithm for supervised feature learning, hierarchical feature extraction and land cover classification of SAR images. RADARSAT-2 ultra-fine beam high resolution HH-SAR images acquired in the rural urban fringe of the Greater Toronto Area (GTA) are selected for this study. The experiment results show that the accuracy of our classification method is about 90% which is higher than that of nearest neighbor (NN).

Polarimetric Classification of C-Band SAR Data for forest Density Characterization

Polarimetric classification is one of the most significant applications of synthetic aperture radar (SAR) remote sensing. Sensitivity of C-band SAR in discerning the variation in canopy roughness and limited penetration capability through forest canopy have been well studied at a given frequency, polarization and incidence angle. However, the scope of C-band SAR in characterizing and monitoring forest density has not been adequately understood with polarimetric techniques. The objectives of the present study were to understand the scattering behaviour of different landcover classes and evaluate the feasibility of polarimetric SAR data classification methods in forest canopy density slicing using C-band SAR data. The RADARSAT- 2 image with fine quad-pol obtained on 27 October 2011 over Madhav National Park, Madhya Pradesh, India and its surroundings was used for the analysis. Forest patches exhibit α-angle around 45°, which means the dominant scattering mechanism is volume; entropy of one or a value close to it denotes distributed targets and low anisotropy values than all other land units, which shows a dominant first scattering mechanism. This study comparatively analysed Wishart supervized classifier and Support Vector Machine (SVM) classifier for classification of the forest canopy density along with other associated land-cover classes for a better understanding of the class separability. All forest density classes showed comparatively good separability in Wishart supervized classification (73.8-84.7%) and in SVM classifier (82.3-84.8%). The results demonstrate the effectiveness of SVM classifier (88.7%) over Wishart supervized classifier (87.8%) with kappa coefficient of 0.86 and 0.85 respectively. The experimental results obtained with polarimetric C-band SAR data over dry deciduous forest area imply that SAR data have a significant potential for estimating stand density in operational forestry.

Fusion of Quickbird MS and RADARSAT SAR data for urban land-cover mapping: object-based and knowledge-based approach

The objective of this research is to evaluate Quickbird multi-spectral (MS) data, multi-temporal RADARSAT Fine-Beam C-HH synthetic aperture radar (SAR) data and fusion of Quickbird MS and RADARSAT SAR for urban land-use/land-cover mapping. One scene of Quickbird multi-spectral imagery was acquired on 18 July 2002 and five-date RADARSAT fine-beam SAR images were acquired during May to August 2002. Quickbird MS images and RADARSAT SAR data were classified using an object-based and rule-based approach. The results demonstrated that the object-based and knowledge-based approach was effective in extracting urban land-cover classes. For identifying 16 land-cover classes, object-based and rule-based classification of Quickbird MS data yielded an overall classification accuracy of 87.9% (kappa: 0.868). For identifying 11 land-cover classes, object-based and rule-based classification of RADARSAT SAR data yielded an overall accuracy: 86.6% (kappa: 0.852 ). Decision level fusion of Quickbird classification and RADARSAT SAR classification was able to take advantage of the best classifications of both optical and SAR data, thus significantly improving the classification accuracies of several land-cover classes (25% for pasture, 19% for soybeans, 17% for rapeseeds) even though the overall classification accuracy of 16 land-cover classes increased only slightly to 89.5% (kappa: 0.885).

Object-based Classification of High Resolution SAR Images for Within Field Homogeneous Zone Delineation

Delineating management zones is important in agriculture for implementing site-specific practices. We delineated within-field homogeneous zones over a corn and a wheat field using high spatial resolution multi-temporal airborne C-band synthetic aperture radar (SAR) imagery with an object-based fuzzy k-means classification approach. Image objects were generated by a segmentation procedure implemented in eCognition® software, and were classified as basic processing units using SAR data. Results were evaluated using analysis of variance and variance reduction of soil electrical conductivity (EC), leaf area index (LAI), and crop yield. The object-based approach provided better results than a pixel-based approach. The variance reduction in LAI, and soil EC varied with SAR acquisition time and incidence angle. Although the variance reduction of yield was not as significant as that of LAI and EC, average yield among the delineated zones were different in most cases. The SAR data classification produced interpretable patterns of soil and crop spatial variability, which can be used to infer within-field management zones.

Polarimetric classification of land cover for Glen Affric radar project

In recent years two polarimetric decompositions methods, the A/E method (Cloude and Pottier, 1997) and the three-component decomposition method (Freeman and Durden, 1998) have become the main methods of the polarimetric synthetic aperture radar data classification in radar imagery applications. In both methods the structural composition of the land cover is modelled using the main backscatter responses, as a function of the polarised channels, the level of energy received and the associated backscatter angles. To enhance the coherence in the polarimetric decompositions, data are generally multi-looked and de-speckled. The authors have applied and compared the 'Freeman' three-component decomposition followed by the Wishart complex classifier with the A/E decomposition algorithms to the de-speckled and edge-enhanced Glen Affric, fully polarimetric L-band radar data. The Glen Affric radar site was chosen for its extensive and diverse semi-natural woodlands of native Scots Pine, and its varied topography. The application results demonstrate the response of the radar classification from a semi-natural land cover on a rapid topography.

Comparison of JERS-1 SAR and Landsat TM Data for Estimating Water Areas in the Amazon.

This paper analyzes landcover in the Amazon using local fractal dimensions in Synthetic Aperture Radar (SAR) data compared to SAR and Thematic Mapper (TM) brightness information. To improve SAR data classification accuracy, we apply local fractal dimensions to classifying water areas in the Amazon.SAR, unlike TM, cannot distinguish water types. Local fractal dimensions in SAR data classify two landcover classes-varzea and others-that SAR brightness information cannot classify. Local fractal dimensions also extract detailed landcover at data acquisition more accurately than TM brightness information. The estimation ratio of the water-1 & water-2 in two study sites is about 34.2% using local fractal dimensions in SAR data.