Land Cover Semantic Annotation Derived from High-Resolution SAR Images

The classification of high-resolution (HR) synthetic aperture radar (SAR) images is of great importance for SAR scene interpretation and application. However, the presence of intricate spatial structural patterns and complex statistical nature makes SAR image classification a challenging task, especially in the case of limited labeled SAR data. This paper proposes a novel HR SAR image classification method, using a multi-scale deep feature fusion network and covariance pooling manifold network (MFFN-CPMN). MFFN-CPMN combines the advantages of local spatial features and global statistical properties and considers the multi-feature information fusion of SAR images in representation learning. First, we propose a Gabor-filtering-based multi-scale feature fusion network (MFFN) to capture the spatial pattern and get the discriminative features of SAR images. The MFFN belongs to a deep convolutional neural network (CNN). To make full use of a large amount of unlabeled data, the weights of each layer of MFFN are optimized by unsupervised denoising dual-sparse encoder. Moreover, the feature fusion strategy in MFFN can effectively exploit the complementary information between different levels and different scales. Second, we utilize a covariance pooling manifold network to extract further the global second-order statistics of SAR images over the fusional feature maps. Finally, the obtained covariance descriptor is more distinct for various land covers. Experimental results on four HR SAR images demonstrate the effectiveness of the proposed method and achieve promising results over other related algorithms.

The application of high-resolution synthetic aperture radar (SAR) images from aerial and satellite sensors challenges the researchers for new and effective interpretation tools. Although most of the main axes in the road network may be detected by a skilled human observer looking for dark or bright linear structures, automatic detection remains a difficult task. In this paper, we proposed a new simple approach to extract main road network in high-resolution SAR images automatically. The approach is based on three steps. The first step is pre-filtering the initial SAR images by using a two-way thresholding process in order to discard uninteresting parts of image. In the second step, the results of the first step are input and use the Hough transform to identify the roads respectively. The last step is based on a feature fusion technique. The road networks detected are combined with the fusion operators. We show the improved results from some examples. It is proved that this approach is effective for the straight highways in high-resolution images

Due to its great application value in the military and civilian fields, ship detection in synthetic aperture radar (SAR) images has always attracted much attention. However, ship targets in High-Resolution (HR) SAR images show the significant characteristics of multi-scale, arbitrary directions and dense arrangement, posing enormous challenges to detect ships quickly and accurately. To address these issues above, a novel YOLO-based arbitrary-oriented SAR ship detector using bi-directional feature fusion and angular classification (BiFA-YOLO) is proposed in this article. First of all, a novel bi-directional feature fusion module (Bi-DFFM) tailored to SAR ship detection is applied to the YOLO framework. This module can efficiently aggregate multi-scale features through bi-directional (top-down and bottom-up) information interaction, which is helpful for detecting multi-scale ships. Secondly, to effectively detect arbitrary-oriented and densely arranged ships in HR SAR images, we add an angular classification structure to the head network. This structure is conducive to accurately obtaining ships’ angle information without the problem of boundary discontinuity and complicated parameter regression. Meanwhile, in BiFA-YOLO, a random rotation mosaic data augmentation method is employed to suppress the impact of angle imbalance. Compared with other conventional data augmentation methods, the proposed method can better improve detection performance of arbitrary-oriented ships. Finally, we conduct extensive experiments on the SAR ship detection dataset (SSDD) and large-scene HR SAR images from GF-3 satellite to verify our method. The proposed method can reach the detection performance with precision = 94.85%, recall = 93.97%, average precision = 93.90%, and F1-score = 0.9441 on SSDD. The detection speed of our method is approximately 13.3 ms per 512 × 512 image. In addition, comparison experiments with other deep learning-based methods and verification experiments on large-scene HR SAR images demonstrate that our method shows strong robustness and adaptability.

In high-resolution synthetic aperture radar (SAR) images, the target size may be much larger than the resolution of SAR and the background clutter statistics are generally not Gaussian-distributed. But traditional constant false alarm rate (CFAR) detection algorithm, which based on the Gaussian-distributed model, performs only one test cell at a time and ignores the mutuality information with other cells in the target, so the performance of the CFAR detection algorithm can be substantially degraded. In the paper, an extend target detection method based on the generalized compound model is proposed. Unlike traditional CFAR detection method which distinguish test cell from the background only on the basis of energy contrast, the extend target detection is sensitive to both the contrast and the distribution of the target. We demonstrate the improved performance using the extend target detection method in the real high-resolution SAR images.

In this paper, a new automatic and adaptive aircraft target detection algorithm in high-resolution airport synthetic aperture radar (SAR) images is proposed. Firstly, region segmentation is used to detect the apron area in the images, which provides the potential area where aircrafts may exist and reduce the search range. Secondly, upon the apron area the pre-segmentation is taken to label the possible target points. Thirdly, the constant false alarm rate (CFAR) detector is improved to cope with multi-target detection situation. The clutter pixels in the sliding detection window will be removed automatically based on pre-segmentation result. As a result, more structural features of the targets are preserved. At last, in order to eliminate the detected false targets and solve the problem that the same target is divided into several disconnected areas, a new joint algorithm based on the area recognition factors and distance cluster is presented. The real airborne SAR image data of some airport is used to verify this target detection algorithm, and the result indicates that this algorithm can detect the aircraft target precisely and decrease the false alarm rate.

Convolutional neural networks (CNNs) have been applied to learn spatial features for high-resolution (HR) synthetic aperture radar (SAR) image classification. However, there has been little work on integrating the unique statistical distributions of SAR images which can reveal physical properties of terrain objects, into CNNs in a supervised feature learning framework. To address this problem, a novel end-to-end supervised classification method is proposed for HR SAR images by considering both spatial context and statistical features. First, to extract more effective spatial features from SAR images, a new deep spatial context encoder network (DSCEN) is proposed, which is a lightweight structure and can be effectively trained with a small number of samples. Meanwhile, to enhance the diversity of statistics, the nonstationary joint statistical model (NS-JSM) is adopted to form the global statistical features. Specifically, SAR images are transformed into the Gabor wavelet domain and the produced multi-subbands magnitudes and phases are modeled by the log-normal and uniform distribution. The covariance matrix is further utilized to capture the inter-scale and intra-scale nonstationary correlation between the statistical subbands and make the joint statistical features more compact and distinguishable. Considering complementary advantages, a feature fusion network (Fusion-Net) base on group compression and smooth normalization is constructed to embed the statistical features into the spatial features and optimize the fusion feature representation. As a result, our model can learn the discriminative features and improve the final classification performance. Experiments on four HR SAR images validate the superiority of the proposed method over other related algorithms.

Region map is the sparse representation of a high-resolution synthetic aperture radar (SAR) image on the middle-level semantic layer in its semantic space. Based on the semantic information of the region map, the high-resolution SAR image is divided into hybrid, structural, and homogeneous pixel subspaces. The segmentation of SAR images can be divided into these three subspaces segmentation, of which the segmentation of hybrid subspace has more challenge because of complex structures. There are often many extremely inhomogeneous areas in the hybrid pixel subspace. Are these nonconnected areas in the same or different classes? To solve this problem, a Bayesian learning model with the constraint of sketch characteristic and an initialization method is proposed to construct a structural vector that can reflect the essential features of each extremely inhomogeneous area. Then, the unsupervised segmentation of the hybrid pixel subspace can be realized by using the structural vectors of these areas in this paper. Theoretical analysis and experimental results show that the performance of the hybrid pixel subspace segmentation realized by the structural vectors based on the Bayesian learning model proposed in the paper is better than that only used by hand designing features.

To investigate the limits of building detection from very high-resolution (VHR) synthetic aperture radar (SAR) images, a new method, based on statistical and structural information fusion, is proposed in this paper. The proposed method contains two stages: First, using order statistics constant false alarm rate (OS-CFAR) and power ratio (PR) detectors, a set of detections are made. These detections have different statistical properties, compared to the other objects, and these properties are selected for discriminating buildings from clutters. Second, the morphological analysis is used for increasing the precision of the detection. In this stage, segments, which have the most similarities to buildings in terms of shape and size, are extracted via various structural elements (SEs). The final result is obtained by fusing the two sets of detections. The experimental results on the four real VHR SAR images show that the proposed method has a high detection rate (DR) and low false alarm rate (FAR).

A new threshold template algorithm is presented in this paper to improve the performance in ship detection in high-resolution synthetic aperture radar (SAR) images. The algorithm introduces SAR formation theories into detection. First, we divide the image into tiles within which the sea background can be treated as homogeneous one. Second, we use a threshold template method to detect candidate ship pixels in each tile. The data model and point spread function model together decide the parameters of the template. Third, group candidate pixels together and exclude false alarms. The template method exploits the relationship between neighbor pixels, which characterise high-resolution images. Experiments validate the algorithm with good performance.

Statistical modeling of Synthetic Aperture Radar (SAR) images is of great importance for speckle noise flltering, target detection and classiflcation, etc. Moreover, it can provide a comprehensive understanding of terrain electromagnetics scattering mechanism. Over the past three decades, many sophisticated models have been developed for SAR images, such as Rayleigh, Gamma, K and G etc. The G 0 distribution is a special form of the G model, which can model the speckle ∞uctuations of many classes of objects like homogeneous, heterogeneous and extremely heterogeneous ones, and is widely used in SAR images interpretation. However, as many improvements have been performed on SAR sensors, the traditional parameter estimation methods of the G 0 distribution may be not su-cient, notably in high resolution SAR images. They cannot arrive at a solution frequently when modeling regions in high resolution SAR images, especially the extremely homogeneous regions. In order to deal with this problem, this paper proposes an improved parameter estimation scheme of the G 0 distribution, which combines the classical moment estimation with the mellin transform. To quantitatively assess the fltting precision of the proposed method, we adopt the Kullback- Leibler (KL) distance, Kolmogorov-Smirnov (KS) test and Mean Square Error (MSE) as similarity measurements. The advantage of this proposed parameter estimation method becomes evident through the analysis of a variety of areas (ground vegetation, trees and buildings) in two high resolution SAR images.

The vortex electromagnetic (EM) wave carrying orbital angular momentum (OAM) offers a new degree of freedom for radar techniques, although to date its application to synthetic aperture radar (SAR) imaging has not yet been widely reported. In this letter, a SAR imaging model based on vortex EM waves is derived. Ku-band vehicle-borne vortex EM waves SAR experiment using phased uniform circular array antennas is reported for the first time. A detailed comparison with conventional plane EM waves SAR imaging is given in terms of signal-to-noise ratio and azimuth resolution. The experimental results demonstrate the effectiveness of the proposed imaging model, and indicate that vortex EM waves can be exploited in SAR imaging with higher azimuth resolution than plane EM waves. OAM, which is completely independent of frequency and polarization, is a new degree of freedom in SAR imaging. This work shows the new applications of vortex EM waves, which paves the way for the development of novel SAR systems.

In high-resolution (HR) and very-high resolution (VHR) synthetic aperture radar (SAR) images, focus is now on the patch-oriented image categorization in contrast to the pixel-based classification in low-resolution SAR images. SAR image categorization requires the generation of a compact feature descriptor that accurately defines the content of the image patch under consideration. In this paper we propose a parametric feature descriptor generated on the complex-valued SAR image within a transformation space. The fractional Fourier transform (FrFT), has been considered to transform the image pixels of the single-look complex (SLC) SAR images in order to obtain a simpler statistical response. The real and imaginary components of the complex-valued FrFT coefficients have been modelled with generalized Gaussian distribution (GGD). The proposed feature descriptor is compared with a Wavelet-decomposition-based parametric feature descriptor; and with the FrFT-based and Gabor-filter-bank-based non-parametric feature descriptors. Categorization accuracy enhancement is demonstrated over several categories comprising of natural topologies. The experimental database consists of 2000 image patches (of size 200 × 200 pixels) extracted from SLC HR TerraSAR-X scenes.

The traditional target detection methods for high-resolution synthetic aperture radar (SAR) images rely on the high intensity contrast between the targets and clutter, which are usually effective on the homogeneous background; however, they may lose effectiveness in the heterogeneous background. Compared with the clutter, the targets of interest usually have specific size and shape characteristics in high-resolution SAR images. Based on what mentioned above, we propose a target detection method based on the hierarchical saliency (HS) for high-resolution SAR images. The proposed method constructs the Bayesian saliency map to obtain the complete structures of both the targets of interest and strong clutter with superpixels as units firstly. Then, the morphological saliency map is constructed to retain the targets of interest while suppressing the strong clutter via the introduction of prior size information. The experimental results on the MiniSAR images show the effectiveness of the proposed target detection method.

Ocean vessels and aircrafts usually cause large oscillated responses and span a high dynamic range (HDR) in synthetic aperture radar (SAR) images. To visualize and further recognize these objects on common display devices, it is necessary to compress the HDR at the proper ratio and enhance the details at both low and high grey levels. To achieve this goal, an adaptive two-scale enhancement method is proposed. First, to enhance the details at low grey levels, an adaptive global HDR compression algorithm is designed based on the grey-level distribution feature. Then, with a simplified heterogeneity measurement based on the weighted standard deviation and entropy at a local scale, high-backscattering objects are determined and enhanced to highlight their peak details at high grey levels. The proposed method is validated using two different types of HDR SAR images, and the experimental results suggest that the proposed method visualizes the information at all grey levels and enhances the object’s local peaks. Compared with gamma mapping algorithms, which are widely used for SAR image display, the proposed method outperforms in visualizing the details of man-made objects in very high resolution SAR images.

Accurate optical and synthetic aperture radar (SAR) image registration is crucial to multisensor remote sensing applications. Though several algorithms have been proposed, the practical implementation that directly matches large and high-resolution optical and SAR images remains underexplored by the community. Equipped with the satellite positioning parameters, optical and SAR images can be roughly registered based on geographic coordinates. However, the relative positioning accuracy is still dozens, even hundreds of pixels due to the inaccuracies of sensor parameters and elevation. Consequently, we propose a robust registration algorithm, which consists of two stages, where the horizontal positioning errors can be reduced in the first stage, and then, we fine-tune the correspondences in the second stage. Specifically, we propose a novel template matching method based on the dilated convolutional feature (DCF) and epipolar-oriented phase correlation. DCF is constructed by a depthwise-separable dilated convolution with multichannel gradients, which are generated by the Sobel operator for optical images and a ratio of exponentially weighted averages (ROEWA) operator for SAR images. Due to the large reception field of dilated convolution, DCF can retain invariance even for large relative positioning errors. The epipolar-oriented rectangle template, which stretches along the epipolar line, is then proposed to capture more overlapping areas compared to square templates. Furthermore, the outlier removal is implemented in the coordinate system of optical images to avoid the effect of range compression in SAR images. Inliers are finally used to refine the rational polynomial coefficients (RPCs) based on the bundle adjustment technique. Experimental results on high-resolution optical and SAR image products of various scenarios demonstrate the effectiveness of the proposed registration framework. The relative positioning errors of the refined RPCs can be reduced from hundreds of pixels to the subpixel level.