Showing papers presented at "IEEE Asia-Pacific Conference on Synthetic Aperture Radar in 2019"

Open Set SAR Target Recognition Using Class Boundary Extracting

Abstract: Typically, synthetic aperture radar (SAR) target recognition uses supervised learning techniques to identify target types. A problem is when identifying the new untrained classes and the false targets, they will be misjudged as known classes. In this paper, an open set recognition method is proposed to provide a rejection option for classifier so that the new untrained target type is identified. The main idea is to model closed category boundaries for the known classes and detect the new classes beyond the reasonable scope of known data. The edge exemplar selection method based on local statistical information is proposed to extract the class boundaries. And the boundaries are used to train the open set recognition/outlier detection model. Experimental results demonstrate that the proposed O_SAR outperforms the other state of the arts on the accuracy of false target rejection and muti-class open set recognition.

Few-Shot SAR ATR Based on Conv-BiLSTM Prototypical Networks

Abstract: In recent studies, deep neural network has been successfully applied to synthetic aperture radar (SAR) automatic target recognition (ATR). However, these algorithms require hundreds of training samples of each class of targets that need to be recognized. In order to recognize the target with only a few training samples, this paper proposes a new few-shot SAR ATR method based on Conv-BiLSTM Prototypical Networks (CBLPN). First, a Conv-BiLSTM network is trained to map SAR images into a new feature space where it is easier for classification. Then, a classifier based on Euclidean distance is utilized to obtain the recognition results. Experimental results on the moving and stationary target acquisition and recognition (MSTAR) benchmark dataset illustrate that the proposed method achieves an accuracy of over 90% on the classification of three classes of targets with only 5 training samples of each class.

Aircraft Detection in High-Resolution SAR Images Using Scattering Feature Information

Abstract: Accurate aircraft detection in high-resolution Synthetic Aperture Radar (SAR) images is of great significance. Aiming at the challenges of sparsity and variability for aircraft targets in SAR images, a detection algorithm based on Scattering Feature Information (SFI) enhancement and Feature Pyramid Network (FPN) is proposed. In the former stage, the SFI, being composed of Strong Scattering Point (SSP) and its corresponding scattering region distribution model, is extracted by Harris-Laplace detector and Gaussian Mixture Model (GMM). Specially, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is introduced to response to the sensitivity of the GMM to the initial values. In the detection stage, an algorithm based on FPN is applied for aircraft detection in high-resolution images. This structure combines the high-resolution information of the underlying features with the high-semantic information of the deep features, which facilitates accurate detection of the aircrafts in a scene. In addition, Logarithmic-normal Distribution based Subdivided Conversion (LDSC) is newly proposed for SAR image preprocessing. Experiments conducted on the GF-3 satellite image of 0.5 m resolution demonstrates the superiority and robustness of the proposed method.

Sea Clutter Suppression Based on Selective Reconstruction of Features

Abstract: This paper presents a sea clutter suppression (SCS) method for single-channel maritime radar based on a deep Convolutional Neural Network (SCS-CNN), which consists of an encoder and a decoder. First, the encoder is used to extract depth features of the original Range-Doppler spectrum obtained from sub-aperture echoes. Second, the decoder is used to selectively reconstruct the desired Range-Doppler spectrum which only contains the moving targets. Finally, the results of moving target detection are obtained by using the cell average constant false alarm rate detector. This method effectively suppresses the sea clutter and correctly indicates the moving targets under different Signal-to-Clutter-plus-Noise Ratios with low false alarm rate. Particularly, it has good feature extraction and reconstruction abilities for the moving targets whose Doppler is inside the mainlobe clutter.

Deep learning based target detection method with multi-features in SAR imagery

Abstract: In view of synthetic aperture radar (SAR) target detection, traditional methods are based on hand-crafted feature extraction and classifier. Besides, deep learning (DL) based methods are research hotspots in recent year. However, their shortcomings cannot be neglected, i.e. detection accuracy of traditional method needs to be improved and DL features are difficult to interpret. To overcome these problems, a target detection method with multi-features in SAR imagery is proposed in this paper. It consists of two parallel sub-channels. DL features and hand-crafted features are extracted in these channels, respectively. Here, convolutional neural network (CNN) model is applied to capture DL features of original SAR images. Deep neural network (NN) is used to further analyze hand-crafted features. Furthermore, two sub-channel features are concatenated together in the main channel. After several layers network processing, fused deep features are extracted. Finally, softmax classifier is applied to discriminate ship target. According to the experiments based on Sentinel-1 SAR data, we can find that the detection performance is improved by the proposed method.

A Imaging Passive Localization Method for Wideband Signal Based on SAR

Abstract: A imaging passive localization method for wideband signal is proposed in this paper. By introducing synthetic aperture radar (SAR) imaging method, the location of the signal emitter is directly given in SAR image. The key of this method is to focus unknown wideband data in range and azimuth domain. To focus the data in range domain without signal parameter, a new pulse compress method is proposed by constructing reference signal from raw data. To focus the data in azimuth domain without knowing range, a range-searching azimuth focus method is proposed by constructing azimuth focus functions with different range. Simulation result validate the effectiveness of the proposed method.

Bistatic and Monostatic InSAR Results with the MetaSensing Airborne SAR System

Abstract: This paper presents the interferometric results obtained from the data acquired during two MetaSensing campaigns. The first results are related to the Bistatic SAR campaign at L-band in Belgium. The data were acquired with the objective to monitor agriculture and crop growth as part of the ESA BelSAR project. The second results are related to single-pass monostatic InSAR acquisitions at C-band in Norway. The objective is to derive the interferometric height for further use in the carbon stock estimation context.

Improved F-K Migration Based on Interpolation Method for GPR Imaging

Abstract: Migration technology is originated from seismic imaging field and has been widely used in ground penetrating radar signal processing. In this paper, the theories of four migration algorithms are studied and an interpolation method in time-frequency domain is proposed to enhance the image resolution of F-K algorithm. Based on the parameters of computational complexity, resolution, sidelobe level and dip-angle range, numerical simulations are carried out to investigate and compare the target reconstruction performance of these algorithms. By comparing the migration results of point-like target, distributed target and tilt target, some valuable conclusions about these algorithms are obtained, which will provide a guidance for selecting specific migration algorithm in practical engineering applications.

Two Dimensional Resolution Equalization Method in Image Domain for Airborne Multi-static SAR

Abstract: The resolution of bistatic SAR system is constrained by the signal bandwidth, distance, velocity and bistatic radar angle. The influence of topology structures and the strong coupling of echoes greatly limit the resolution of SAR images. And the problem of unbalanced two dimensional resolution brings difficulty and challenge to the target detection and recognition. Multi-static SAR can naturally form multiple pairs of bistatic SAR system and it fuses image information from multi-view by the design of topology structure to improve imaging resolution. To achieve optimal spatial resolution in all directions, the typical multi-static SAR geometric model is presented and the influence of topology structure on the two-dimensional resolution of bistatic and multi-static SAR is analyzed in this paper. Then, an equalization method to improve resolution for multi-static SAR is proposed. Finally, the simulation results validate the effectiveness of the proposed method.

Interference Mitigation for Synthetic Aperture Radar Using Deep Learning

Abstract: In this paper, we present a narrow-band interference (NBI) and wide-band interference (WBI) mitigation algorithm based on the deep residual network (ResNet). First, the short-time Fourier transform (STFT) is utilized to characterize the interference-corrupted echo in the time-frequency domain. Then, the interference detection model is built by the classical convolutional neural network (CNN) framework to identify whether the echoes exist interference signal component. Furthermore, the time-frequency feature of the target signal is extracted and reconstructed by utilizing the ResNet. Finally, the inverse time-frequency Fourier transform (ISTFT) is utilized to transform the time-frequency spectrum of the recovered signal into the time domain. The effectiveness of the interference mitigation algorithm is verified on the simulation and measured SAR data of the terrain observation by progressive scans (TOPS) mode. Moreover, the performance comparison with the notch filtering and eigensubspace filtering demonstrates the superiority of the proposed interference mitigation algorithm.

Super-resolution Doppler Beam Sharpening based on Online Tikhonov Regularization

Abstract: Doppler beam sharpening (DBS) is a crucial technique in radar imaging. However, the conventional DBS based on fast Fourier transform provides low azimuth resolution and high sidelobe level. Many of the current super-resolution DBS imaging methods are carried out to greatly improve the azimuth resolution. Nonetheless, these methods generally adopt a disposal of batch processing with high computational complexity, causing the insufficient of real-time capability. This paper proposes an online super-resolution DBS imaging approach based on Tikhonov regularization by adding the regularization term and utilizing matrix blocking. The current iterative estimation value of the scattering coefficient can be updated by the former iterative estimation value and the new data acquired from the current. The proposed method effectively improves the resolution with allocating total computation burden for each sampling interval, and possesses the particularly advantage of acquiring echo while processing imaging results, which is suitable for realtime continuous reconnaissance. Simulation results are given to demonstrate the effectiveness of the proposed method.

Ship Detection Based on RetinaNet-Plus for High-Resolution SAR Imagery

Abstract: Ship detection in high-resolution synthetic aperture radar (SAR) imagery is a fundamental and challenging problem due to the complex environments. In this paper, a RetinaNet-Plus method is presented for ship detection in high-resolution SAR imagery based on RetinaNet network modified. In this approach, instead of setting the score for neighboring region proposals to zero as in Non-Maximum Suppression (NMS), Soft-NMS decreases the detection scores as an increasing function of overlap to avoid loss of precision. In addition, focal loss is used to address the class imbalance and to increase the importance of the hard examples during training. The experiments on SAR ship SSDD dataset and TerraSAR-X image from Barcelona port, show that our method is more accurate than the existing algorithms and is effective for ship detection of high-resolution SAR imagery.

Using convolutional neural network (CNN) approach for ship detection in Sentinel-1 SAR imagery

Abstract: Accurate maritime ship surveillance and monitoring ensures compliance with port regulations and standards. The growing volume of waterborne traffic however, has made this goal difficult to achieve in applications like maritime traffic control, ship search and rescue, territorial regulation, and fishery management. Detection of ships is complicated, especially under unfavourable conditions, such as during night-time or on cloudy days. Synthetic aperture radar (SAR) provides high-resolution data that can overcome these limitations. Using machine-learning techniques to detect ships in a SAR based image can increase the accuracy of identification detection results as compared to traditional image-based object detection methods. Sentinel-1 SAR images from 2015 to 2018 were used in this exploratory study presenting an analysis of effective ship detection and ship count in a congested sea environment using a Convolutional Neural Networks (CNNs) method, the Faster R-CNN VGG16. Experiments using sixteen convolutional layers in the model yielded promising and significant detection, identification and ship count results for the port of Shanghai.

Low-Cost Matched Filter for 1-bit SAR Singnal Processing Based on Single-Frequency Threshold

Abstract: This paper presents a matched filter (MF) model for 1-bit SAR signal processing. Logic gates can be used instead of traditional multipliers to simplify the system. Moreover, the single-frequency threshold shifts the high-order harmonics introduced by the 1-bit sampling outside the imaging component, making it possible to reduce the sampling rate after MF. In this way, the convolutional down-sampling architecture is introduced to further cut down the computational load and reduce the storage pressure of the output high-precision data. The hardware implementation on FPGA verifies the validity of the proposed method and design.

An Inshore Ship Detection Method in SAR Images Based on Contextual Fluctuation Information

Abstract: During the inshore ship detection in SAR images, the high similarity between the harbor and the ship body on gray and texture features, resulting in low detection accuracy and high false alarm rate. In this paper, we propose a new approach using contextual fluctuation information to deal with this problem. Firstly, the maximum stability extremal region (MSER) method is used for global pre-screening to quickly obtain candidate targets. Next, the context slice of each candidate target is obtained, then each slice is devided into grids. Distinguish real ship targets from false alarms based on the fluctuations in pixel values within the divided grid. Experimental results based on satellite-borne SAR data illustrate that the proposed method obtains excellent detection performance and low false alarm rate.

Cognitive SAR Waveform Design Method Based on Joint Optimization Criteria

Abstract: Intelligence technology has been widely used in various radar applications in recent years. However, there are few studies on the combination of intelligence technology and synthetic aperture radar (SAR). In this paper, a cognitive SAR waveform design method based on joint optimization criteria is proposed. The proposed method constructs a criteria jointing the resolution of SAR transmitting waveform and output SCNR. Cognitive SAR waveform designed upon this criterion solves the problem of weak clutter suppression and poor cognitive ability of conventional SAR. Meanwhile, it also solves the problem that the resolution performance of waveform designed based on the traditional criterion of maximum output SCNR is too poor to be used in SAR for high resolution missions. Simulation results show that the waveform designed by this method has high SCNR and great resolution.

SAR/ISAR Imaing Using Commercial Millimeter-wave Radar

Abstract: Millimeter-wave (mmWave) radar has drawn much attention for both academic research and industrial applications. Different types of single-chip mMwave radar system has been developed driving by the increasing demand of automotive radar applications. MmWave radar is also favored by imaging radar since high range and cross-range resolutions can be easily achieved. However, high-resolution imaging is not the primary task of automotive radar, although it can be easily achieved by applying synthetic aperture technique. In this paper, we present short-range synthetic aperture radar (SAR) imaging experiments using commercial single-chip mmWave radar. Both SAR and ISAR experiments are conducted and the results show that cross-range resolution can be improved to several centimeters by adopting SAR technique to single-chip mmWave radar. Therefore, commercial mMwave radar provides an effective and low-cost approach for evaluating SAR/ISAR working mode and testing the processing algorithm.

Few-Shot Learning Neural Network for SAR Target Recognition

Abstract: Deep neural networks, especially convolutional neural networks are recently applied to synthetic aperture radar target recognition and achieved state-of-the-art results. Large amount of labeled data are needed during training period of deep neural network. However, labeling enough synthetic aperture radar data on novel classes is not feasible. In this paper, a new framework is presented by introducing triplet loss function to train a deep neural network with few labeled data. The proposed few-shot learning method is verified using Moving and Stationary Target Acquisition and Recognition data set. The results show that the proposed network has good recognition performance on limited labeled data.

A Maritime Target Detector Based on CNN and Embedded Device for GF-3 Images

Abstract: Recently, with the development of deep learning and the springing up of synthetic aperture radar (SAR) images, SAR maritime target detection based on convolutional neural network (CNN) has become a hot issue. However, most related work is realized on general purpose hardware like CPU or GPU, which is energy consuming, non-real-time and unable to be deployed on embedded devices. Aiming at this problem, this paper proposes a method to deploy a model of SAR maritime target detection network on an embedded device which employs custom artificial intelligence streaming architecture (CAISA). Moreover, the model is trained and tested on the Gaofen-3 (GF-3) spaceborne SAR images, which include six different kinds of maritime targets. Experiments based on the GF-3 dataset show the method is practicable and extensible.

Performance Analysis of Space-Borne Early Warning Radar for AMTI

Abstract: Space-borne early warning radar has great military and civil significance in ground, air, and space moving target detection in virtue of its wide-area coverage ability, where air moving target indication (AMTI) is an essential task for space-borne early warning radar. As for the AMTI mode, it is inevitably to handle the ground strong clutters due to the large surveillance area, and space-time adaptive processing (STAP) technique can be successfully applied to suppress the troubled clutter effectively. However, the clutter rejection performance of STAP will be restricted by many non-ideal factors, such as range ambiguity, channel number, and yaw angle induced by earth's rotation; thus it is necessary to analyze the influence of these practical non-ideal factors. In this paper, the multichannel signal model of clutter and air moving target in a space-borne radar (SBR) system are deduced. Then, we analysis the adverse influences of these non-ideal factors on STAP in detail. Finally, some simulation processing results are provided to validate the STAP performance with the consideration of these engineer factors, which can provide some useful guides of SBR system design with AMTI mode.

A new ship detection and classification method of spaceborne SAR images under complex scene

Abstract: Satellite remote sensing technology has always received wide attention for its developing performance of earth observation. Ship detection and classification based on spaceborne SAR images has been an attractive and intractable topic because the wide sea area is too complex to detect and classify all the objective ships. In this paper, a new ship detection and classification method for complex sea surface is presented. It adopts the visual saliency detection method based on spectral residual to obtain the locations of the regions of interest(ROIs) containing ships. And the morphology filter is employed to exclude a part of false alarm targets (FATs). Then, the types of the ships are classified based on convolution neural network (CNN). Finally, the locations and types of ships in large sea SAR images are acquired. Experimental results based on measured spaceborne SAR images have shown the effectiveness and accuracy of the proposed method.

Clutter suppression for multichannel wide area surveillance system via a combination of STAP and RPCA

Abstract: Wide area surveillance ground moving target indication (WAS-GMTI) is an important mode of multichannel airborne radar. Space-time adaptive processing (STAP) is widely used in WAS-GMTI clutter suppression. However, it will be difficult to estimate the clutter plus noise spectral density matrix accurately when the training samples are contaminated by moving targets. In this paper, target-free training samples obtained via robust principal component analysis (RPCA) algorithm are used for more accurate clutter plus noise spectral density matrix estimation. An eigenvector subspace projection method is used to calculate the filter vector. The simulation results show the advantages of the proposed method.

Inshore ship detection based on multi-aspect information in high-resolution SAR images

Abstract: A novel algorithm for inshore ship detection based on multi-aspect information in high-resolution Synthetic Aperture Radar (SAR) images is proposed. Based on the internal and external characteristics of inshore ship and harbor regions, multi-aspect information, including coastline information, context information, scattering mechanism, shape contour and deep feature information, are considered respectively to detect inshore ship targets. The algorithm is verified to be robust and efficient to exact the Region-of-Interest (ROI) of inshore ship, and achieve a good performance with detection rate 94.24%. Experiments demonstrate good performance with detection rate 94.24%. The results show that the method is simple and robust, which can effectively determine the Region-of-Interest (ROI) of inshore ship.

FPGA Implementation of SAR Imaging Processing System

Abstract: The miniaturized synthetic aperture radar (MiniSAR) signal processing system is designed and implemented in this paper, which is able to deal with chirped SAR signals based on FPGA. In this design, the Polar Format Algorithm (PFA) using the principle of chirp scaling (PCS) for range processing and Sinc interpolation for azimuth processing can achieve high precision results and increase speed significantly. Meanwhile, the phase gradient autofocus (PGA) and the geometric correction (GC) are applied to estimate and compensate for the residual phase error accurately and realize wavefront curvature correction caused by PFA. The system uses the Floating-Point IP cores and pipeline structure to achieve high-speed floating-point data computation, and uses a smart scheme to realize the transposition of matrix data demanded by the system algorithm with DDR3 SDRAM. The system is built on Virtex7-XC7VX690T evaluation board, and it takes 2.1s to obtain 4K*2K complex-image in single precision. Point target simulation has validated the presented methodology, and the real data processing results verify the reliability and stability of the proposed system.

A Fast BP Algorithm for Bistatic Spotlight SAR Based on Spectrum Compression

Abstract: For bistatic spotlight synthetic aperture radar (SAR) with parallel trajectory, this paper investigates a fast back-projection (BP) imaging algorithm based on spectrum compression. The long synthetic aperture is split into several short subapertures (SAs), and each of them constructs a SA image with coarse azimuth resolution in a unified rectangular coordinate system by BP integral. Then, by applying the spectrum compression technique introduced in the paper, the azimuth spectrum of the SA image can be compressed greatly and, as such, the images can be coherently accumulated after azimuth up-sampling and spectrum recovery. Since there is no need for data interpolation in the image fusion, the proposed algorithm can achieve high precision as the original BP algorithm does but with higher computational efficiency. Simulation experiments are presented to validate the effectiveness of the method.

Monitoring and analyzing surface subsidence based on PS-InSAR in Nanning region, China

Abstract: Surface subsidence is an environmental geological phenomenon in which the elevation of the crust surface is reduced due to the compression of underground loose soil layers under the influence of natural geological factors or human activities. It has the characteristics of high monitoring difficulty and long duration. Differential Interferometric Synthetic Aperture Radar (D-InSAR) was susceptible to signal decorrelation and atmospheric delay. It is difficult to obtain high-precision, high-temporal resolution urban land subsidence information. Therefore, this paper introduced Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) with high spatial-temporal resolution and high precision for urban surface subsidence monitoring and analysis research, taking Nanning region as an example. Based on the twenty Sentinel-1A TOPS SAR images, this paper completed the acquisition of surface subsidence information from July 2015 to February 2017 in Nanning region, and conducted subsidence spatial-temporal analysis in this region. The experimental results show that the uneven surface subsidence in Nanning region was obvious. The subsidence rate of most areas during the study period ranged from −31.5 to +17.5 mm/a. The ground in the central part of Xi Xiangtang district showed a trend of recovery, while the surface subsidence in Jinling town in the northwest of Xi Xiangtang district was as high as −30 mm/a, and most of the Jiangnan district is in a stable state. The surface subsidence was relatively large in Qingxiu district, Yongning district and northern Liangqing district, among which the largest cumulative subsidence in Qingxiu District is 61.5 mm. In addition, the surface subsidence in the study area was more affected by urban traffic track construction and real estate development significantly.

SAR and Optical Image Registration Method Based on Improved CycleGAN

Abstract: It is challenging to register synthetic aperture radar (SAR) and optical images due to large radiometric differences, geometric differences and strong speckle noise. To address this problem, a SAR and optical image registration method based on the improved CycleGAN is presented in this paper. The proposed method is composed of three steps. Firstly, the SAR image is translated into a pseudo-optical image by the improved CycleGAN which is obtained by modifying the generators, the discriminator and the loss function of the original CycleGAN. Secondly, the optical image and the pseudo-optical image are registered by the method suitable for optical image registration to get the affine transformation parameters. Finally, the original SAR image is aligned with the optical image according to these parameters. The results of experiments show that our proposal significantly improves the registration performance of optical and SAR image registration.

An ISAR Approach for Refocusing the Maritime Moving Targets With the GF-3 SAR Satellite

Abstract: Moving ship targets are always defocused in synthetic aperture radar (SAR) images under the influence of self-motion and sea waves. Inverse SAR (ISAR) technique is commonly used to focus non-cooperative targets by motion compensation. A novel refocusing method, improved rank one phase estimation (IROPE), is proposed in this paper. The ISAR processing is based on the proposed method, an iterative two-step convergence approach to obtain the phase error. The performance of the proposed method is tested by comparing it to other focusing algorithms using real spotlight complex image data of Gaofen-3 (GF-3) spaceborne SAR system. Results verify the effectiveness of the proposed method.

Dual-Frequency Interferometric Performance Simulation of UAV Dupa-SAR

Abstract: The dual-frequency SAR interferometry is an extension of the traditional SAR interferometry. Two interference phases with different ambiguity are used in two frequency bands to reduce the phase aliasing in steep regions, facilitate the phase unwrapping, reduce data noise and improve the accuracy of height estimation. In order to obtain high-precision interferometric elevation information in local areas flexible, SKL-SGIIT is developing the dual-frequency polarimetric advanced SAR system (Dupa-SAR) of unmanned aerial vehicle (UAV). In this paper, the parameters, and simulation results of InSAR performance of Dupa-SAR is presented.

Measurement of Tidal Flat Topography using Long-Baseline InSAR

Abstract: Tidal flats which are located between land and ocean are very vulnerable to the change of environment. These places are now facing many environmental challenges related to climate change and human-induced impacts. Topographic changes related to sedimentation or erosion in tidal flats are the most evident sign of the environmental changes. The tidal flats usually have small variations of topographies and experience ebb and flood tides every day. The conventional SAR interferometric techniques (such as repeat-pass InSAR) cannot be applied for generating digital elevation models (DEMs) in these tidal flats because of low coherence caused by ebb and flood tides. In this study, we developed a long-baseline single-pass airborne interferometric SAR (InSAR) system that has a small ambiguity height. We collected InSAR data in several tidal flats, the west coast of Korean peninsula, using our airborne InSAR system and TanDEM-X. The TanDEM-X is a space borne quasi-single pass interferometic SAR system. We also investigated the expected accuracy of DEM as function of baseline and coherence. In general, longer baseline is better for generating very sensitive and fine DEM that is absolutely needed for tidal flats, because longer baseline has smaller ambiguity height. But longer baseline can also cause low coherence due to baseline decorrelation. As the baseline increase, common spectral region is also decreased resulting in poorer coherence and poor vertical accuracy. Based on these investigations, we proposed optimal baselines for airborne InSAR system and space-borne SAR system. Finally, we compared the constructed topographies from our long-baseline airborne InSAR and TanDEM-X with GPS-RTK measurements.