CNN-LRP: Understanding Convolutional Neural Networks Performance for Target Recognition in SAR Images.

Target recognition is one of the most challenging tasks in synthetic aperture radar (SAR) image processing since it is highly affected by a series of pre-processing techniques which usually require sophisticated manipulation for different data and consume huge calculation resources. To alleviate this limitation, numerous deep-learning based target recognition methods are proposed, particularly combined with convolutional neural network (CNN) due to its strong capability of data abstraction and end-to-end structure. In this case, although complex pre-processing can be avoided, the inner mechanism of CNN is still unclear. Such a "black box" only tells a result but not what CNN learned from the input data, thus it is difficult for researchers to further analyze the causes of errors. Layer-wise relevance propagation (LRP) is a prevalent pixel-level rearrangement algorithm to visualize neural networks' inner mechanism. LRP is usually applied in sparse auto-encoder with only fully-connected layers rather than CNN, but such network structure usually obtains much lower recognition accuracy than CNN. In this paper, we propose a novel LRP algorithm particularly designed for understanding CNN's performance on SAR image target recognition. We provide a concise form of the correlation between output of a layer and weights of the next layer in CNNs. The proposed method can provide positive and negative contributions in input SAR images for CNN's classification, viewed as a clear visual understanding of CNN's recognition mechanism. Numerous experimental results demonstrate the proposed method outperforms common LRP.

Ultra-High Resolution Imaging Method for Distributed Small Satellite Spotlight MIMO-SAR Based on Sub-Aperture Image Fusion.

Small satellite synthetic aperture radar (SAR) has become a new development direction of spaceborne SAR due to its advantages of flexible launch, short development cycle, and low cost. However, there are fewer researches on distributed small satellite multiple input multiple output (MIMO) SAR. This paper proposes an ultra-high resolution imaging method for the distributed small satellite spotlight MIMO-SAR, which applies the sub-aperture division technique and the sub-aperture image coherent fusion algorithm to MIMO-SAR. After deblurring the sub-aperture signal, the large bandwidth signal is obtained by using an improved time domain bandwidth synthesis (TBS) method, and then the ultra-high resolution image is obtained by using a sub-aperture image coherent fusion algorithm. Simulation results validate the feasibility and effectiveness of the proposed approach.

Realizing Target Detection in SAR Images Based on Multiscale Superpixel Fusion.

For target detection in complex scenes of synthetic aperture radar (SAR) images, the false alarms in the land areas are hard to eliminate, especially for the ones near the coastline. Focusing on the problem, an algorithm based on the fusion of multiscale superpixel segmentations is proposed in this paper. Firstly, the SAR images are partitioned by using different scales of superpixel segmentation. For the superpixels in each scale, the land-sea segmentation is achieved by judging their statistical properties. Then, the land-sea segmentation results obtained in each scale are combined with the result of the constant false alarm rate (CFAR) detector to eliminate the false alarms located on the land areas of the SAR image. In the end, to enhance the robustness of the proposed algorithm, the detection results obtained in different scales are fused together to realize the final target detection. Experimental results on real SAR images have verified the effectiveness of the proposed algorithm.

InSAR Signal and Data Processing.

We present here the recent advances in exploring new techniques related to interferometric synthetic aperture radar (InSAR) signal and data processing and applications.

A Novel Reconstruction Method of K-Distributed Sea Clutter with Spatial-Temporal Correlation.

The reconstruction of sea clutter plays an important role in target detection and recognition in a maritime environment. Reproducing the temporal and spatial correlations of real data simultaneously is always a problem in the reconstruction of sea clutter due to the complex coupling between them. In this paper, the spatial-temporal correlated proportional method (STCPM), based on a compound model, is proposed to reconstruct K-distributed sea clutter with correlation characteristics obtained from the real data. The texture component with spatial-temporal correlation is generated by the proportional method and the speckle component with temporal correlation is generated by matrix transformation. Compared with previous methods, the biggest innovation of the STCPM is that it can more accurately generate K-distributed sea clutter with both temporal and spatial correlations. The comparison of the reconstructed and real data demonstrates that the method can reproduce the characteristics of real sea clutter well.

A Target Identification Method for the Millimeter Wave Seeker via Correlation Matching and Beam Pointing.

Target identification is a challenging task under land backgrounds for the millimeter wave (MMW) seeker, especially under complex backgrounds. Focusing on the problem, an effective method combining correlation matching and beam pointing is proposed in this paper. In the beginning, seeker scanning for target detection is conducted in two rounds, and target information of the detected targets is stored for correlation matching. Point or body feature judgment is implemented by using high resolution range profile (HRRP). Then, the error distribution zone is constructed with the beam pointing as the origin. In the end, we identify the target by searching the one which lies in the closest error distribution from the beam pointing center. The effectiveness of the proposed method is verified by using mooring test-fly and real flight data.

High-Rise Building 3D Reconstruction with the Wrapped Interferometric Phase.

The great development of high-resolution SAR system gives more opportunities to observe building structures in detail, especially the advanced interferometric SAR (InSAR), which techniques attract more attention on exploiting useful information on urban infrastructures. Considering that the high-rise buildings in urban areas are quite common in big cities, it is of great importance to retrieve the three-dimension (3D) information of the urban high-rise buildings in urban remote sensing applications. In this paper, the 3D reconstruction of high-rise buildings using the wrapped InSAR phase image was studied, referring to the geometric modulation in very high resolution (VHR) SAR images, such as serious layover cause by high-rise buildings. Under the assumption of a rectangular shape, the high-rise buildings were detected and building façades were extracted based on the local frequency analysis of the layover fringe patterns. Then 3D information of buildings were finally extracted according to the detected façade geometry. Except for testing on a small urban area from the TanDEM-X data, the experiment carried on the single-pass InSAR wrapped phase in the wide urban scene, which was collected by the Chinese airborne N-SAR system, also demonstrated the possibility and applicability of the approach.

A Novel MIMO⁻SAR Solution Based on Azimuth Phase Coding Waveforms and Digital Beamforming.

In multiple-input multiple-output synthetic aperture radar (MIMO⁻SAR) signal processing, a reliable separation of multiple transmitted waveforms is one of the most important and challenging issues, for the unseparated signal will degrade the performance of most MIMO⁻SAR applications. As a solution to this problem, a novel APC⁻MIMO⁻SAR system is proposed based on the azimuth phase coding (APC) technique to transmit multiple waveforms simultaneously. Although the echo aliasing occurs in the time domain and Doppler domain, the echoes can be separated well without performance degradation by implementing the azimuth digital beamforming (DBF) technique, comparing to the performance of the orthogonal waveforms. The proposed MIMO⁻SAR solution based on the APC waveforms indicates the feasibility and the spatial diversity of the MIMO⁻SAR system. It forms a longer baseline in elevation, which gives the potential to expand the application of MIMO⁻SAR in elevation, such as improving the performance of multibaseline InSAR and three-dimensional SAR imaging. Simulated results on both a point target and distributed targets validate the effectiveness of the echo separation and reconstruction method with the azimuth DBF. The feasibility and advantage of the proposed MIMO⁻SAR solution based on the APC waveforms are demonstrated by comparing with the imaging result of the up- and down-chirp waveforms.

Micro-Doppler Feature Extraction of Inverse Synthetic Aperture Imaging Laser Radar Using Singular-Spectrum Analysis.

Different from microwave radar, laser radar could be more sensitive to the micro-Doppler (m-D) effect due to its wave length. This limits the application of conventional methods, such as time⁻frequency based approach, since the processing needs a receiver with much higher sampling frequency than microwave radar. In this paper, a micro-Doppler feature extraction algorithm is proposed for the inverse synthetic aperture imaging laser radar (ISAIL). Singular-spectrum analysis (SSA) is employed for separation and reconstruction of the micro-Doppler and rigid body signal. Clear ISAIL image is obtained by minimum entropy criteria after echo signal decomposition. After theoretical derivation, the computation efficiency and ability of the proposed method is proved by the results of simulation and real data of An-26.

SAR Target Configuration Recognition via Product Sparse Representation.

Sparse representation (SR) has been verified to be an effective tool for pattern recognition. Considering the multiplicative speckle noise in synthetic aperture radar (SAR) images, a product sparse representation (PSR) algorithm is proposed to achieve SAR target configuration recognition. To extract the essential characteristics of SAR images, the product model is utilized to describe SAR images. The advantages of sparse representation and the product model are combined to realize a more accurate sparse representation of the SAR image. Moreover, in order to weaken the influences of the speckle noise on recognition, the speckle noise of SAR images is modeled by the Gamma distribution, and the sparse vector of the SAR image is obtained from q statistical standpoint. Experiments are conducted on the moving and stationary target acquisition and recognition (MSTAR) database. The experimental results validate the effectiveness and robustness of the proposed algorithm, which can achieve higher recognition rates than some of the state-of-the-art algorithms under different circumstances.

A Channel Phase Error Correction Method Based on Joint Quality Function of GF-3 SAR Dual-Channel Images.

Multichannel SAR is an effective approach to solving the contradiction between high azimuth resolution and wide swath. The goal of this paper is to obtain a new and effective method for estimating and compensating the interchannel phase error of the Chinese GF-3 Synthetic aperture radar (SAR). A channel phase error correction method based on the optimal value of the image domain quality function is proposed. In this method, the phase error is initially compensated using the correlation function method. In the fine correction of dual-channel phase error, a heuristic search algorithm is used to estimate the residual phase by searching the extremum of the quality function. After phase compensation in the image domain, the azimuth ambiguities caused by the remaining phase are eliminated. The proposed image domain processing method provides a new idea for channel phase error correction. The measured data of high-resolution GF-3 dual-channel ultrafine imaging mode verifies the validity of this method.

A Real-Time Imaging Algorithm Based on Sub-Aperture CS-Dechirp for GF3-SAR Data.

Conventional synthetic aperture radar (SAR) imaging algorithms usually require a period of time to process data that is longer than the time it takes to record one synthetic aperture or that corresponding to an adequate azimuth resolution. That is to say, the real-time processing system is idle during the long data recording time and the utilization of computational resources is low. To deal with this problem, a real-time imaging algorithm based on sub-aperture chirp scaling dechirp (CS-dechirp) is proposed in this paper. With CS-dechirp, the sub-aperture data could be processed to form an image with relatively low resolution. Subsequently, a few low-resolution images are generated as longer azimuth data are recorded. At the stage of full-resolution image generation, a coherent combination method for the low-resolution complex-value images is developed. As the low-resolution complex-value images are coherently combined one by one, the resolution is gradually improved and the full-resolution image is finally obtained. The results of a simulation and real data from the GF3-SAR validate the effectiveness of the proposed algorithm.

Real-time implementation of inverse synthetic aperture radar imaging using field programmable gate array and digital signal processors.

Inverse synthetic aperture radar (ISAR) imaging plays an important role in modern radar systems. It is a powerful tool to surveil the air and space targets under all-weather and day-and-night conditions. With increasing demands for space surveillance and target recognition, real-time ISAR imaging is of great urgency. Based on field programmable gate arrays and digital signal processors, a new design of real-time processing of ISAR imaging is presented in this paper. First, a thorough description of the imaging algorithm is presented. The algorithm includes pulse compression, envelope alignment, phase adjustment, and cross range focusing. For the pulse compression, two different methods are adopted to process the "Dechirp" data and the directly sampled wideband data, respectively. Second, a hardware design of real-time imaging processing is given. Each step of the algorithm is mapped onto the hardware, and the computing resources and latency are analyzed. Finally, the realization is verified by both the simulated and measured data, and the quality of imaging result is evaluated qualitatively.

Terahertz Image Detection with the Improved Faster Region-Based Convolutional Neural Network.

In recent years, terahertz imaging systems and techniques have been developed and have gradually become a leading frontier field. With the advantages of low radiation and clothing-penetrable, terahertz imaging technology has been widely used for the detection of concealed weapons or other contraband carried on personnel at airports and other secure locations. This paper aims to detect these concealed items with deep learning method for its well detection performance and real-time detection speed. Based on the analysis of the characteristics of terahertz images, an effective detection system is proposed in this paper. First, a lots of terahertz images are collected and labeled as the standard data format. Secondly, this paper establishes the terahertz classification dataset and proposes a classification method based on transfer learning. Then considering the special distribution of terahertz image, an improved faster region-based convolutional neural network (Faster R-CNN) method based on threshold segmentation is proposed for detecting human body and other objects independently. Finally, experimental results demonstrate the effectiveness and efficiency of the proposed method for terahertz image detection.

Moving target detection for frequency agility radar by sparse reconstruction.

Frequency agility radar, with randomly varied carrier frequency from pulse to pulse, exhibits superior performance compared to the conventional fixed carrier frequency pulse-Doppler radar against the electromagnetic interference. A novel moving target detection (MTD) method is proposed for the estimation of the target's velocity of frequency agility radar based on pulses within a coherent processing interval by using sparse reconstruction. Hardware implementation of orthogonal matching pursuit algorithm is executed on Xilinx Virtex-7 Field Programmable Gata Array (FPGA) to perform sparse optimization. Finally, a series of experiments are performed to evaluate the performance of proposed MTD method for frequency agility radar systems.

3D Geometry and Motion Estimations of Maneuvering Targets for Interferometric ISAR With Sparse Aperture.

In the current scenario of high-resolution inverse synthetic aperture radar (ISAR) imaging, the non-cooperative targets may have strong maneuverability, which tends to cause time-variant Doppler modulation and imaging plane in the echoed data. Furthermore, it is still a challenge to realize ISAR imaging of maneuvering targets from sparse aperture (SA) data. In this paper, we focus on the problem of 3D geometry and motion estimations of maneuvering targets for interferometric ISAR (InISAR) with SA. For a target of uniformly accelerated rotation, the rotational modulation in echo is formulated as chirp sensing code under a chirp-Fourier dictionary to represent the maneuverability. In particular, a joint multi-channel imaging approach is developed to incorporate the multi-channel data and treat the multi-channel ISAR image formation as a joint-sparsity constraint optimization. Then, a modified orthogonal matching pursuit (OMP) algorithm is employed to solve the optimization problem to produce high-resolution range-Doppler (RD) images and chirp parameter estimation. The 3D target geometry and the motion estimations are followed by using the acquired RD images and chirp parameters. Herein, a joint estimation approach of 3D geometry and rotation motion is presented to realize outlier removing and error reduction. In comparison with independent single-channel processing, the proposed joint multi-channel imaging approach performs better in 2D imaging, 3D imaging, and motion estimation. Finally, experiments using both simulated and measured data are performed to confirm the effectiveness of the proposed algorithm.

Multiple-input multiple-output synthetic aperture ladar system for wide-range swath with high azimuth resolution.

For the trade-off between the high azimuth resolution and the wide-range swath in the single-input single-output synthetic aperture ladar (SAL) system, the range swath of the SAL system is restricted to a narrow range, this paper proposes a multiple-input multiple-output (MIMO) synthetic aperture ladar system. The MIMO system adopts a low pulse repetition frequency (PRF) to avoid a range ambiguity for the wide-range swath and in azimuth adopts the multi-channel method to achieve azimuth high resolution from the unambiguous azimuth wide-spectrum signal, processed through adaptive digital beam-forming technology. Simulations and analytical results are presented.

The design and implementation of a multi-waveform radar echo simulator.

Radar simulator is an effective tool for performance assessment of radar systems by accurately reproducing echo signals from complicated environment. This paper presents a design of fast multi-waveform radar echo generation based on deconvolution method. First, scene information is retrieved from outfield data based on improved conjugate gradient algorithm. Then, the new radar echoes are generated through convolution of new transmitted signal and restored scene information. A fast and area-efficient field programmable gate array realization is provided to meet the real-time requirement of radar echo simulation. Finally, a series of experiments are performed to evaluate the effectiveness of proposed radar simulation instrument.

Real-time implementation of frequency-modulated continuous-wave synthetic aperture radar imaging using field programmable gate array.

A new miniature linear frequency-modulated continuous-wave radar which mounted on an unmanned aerial vehicle is presented. It allows the accomplishment of high resolution synthetic aperture radar imaging in real-time. Only a Kintex-7 field programmable gate array from Xilinx is utilized for whole signal processing of sophisticated radar imaging algorithms. The proposed hardware architecture achieves remarkable improvement in integration, power consumption, volume, and computing performance over its predecessor designs. The realized design is verified by flight campaigns.

A robust channel-calibration algorithm for multi-channel in azimuth HRWS SAR imaging based on local maximum-likelihood weighted minimum entropy.

High-resolution and wide-swath (HRWS) synthetic aperture radar (SAR) is an essential tool for modern remote sensing. To effectively deal with the contradiction problem between high-resolution and low pulse repetition frequency and obtain an HRWS SAR image, a multi-channel in azimuth SAR system has been adopted in the literature. However, the performance of the Doppler ambiguity suppression via digital beam forming processing suffers the losses from the channel mismatch. In this paper, a robust channel-calibration algorithm based on weighted minimum entropy is proposed for the multi-channel in azimuth HRWS SAR imaging. The proposed algorithm is implemented by a two-step process. 1) The timing uncertainty in each channel and most of the range-invariant channel mismatches in amplitude and phase have been corrected in the pre-processing of the coarse-compensation. 2) After the pre-processing, there is only residual range-dependent channel mismatch in phase. Then, the retrieval of the range-dependent channel mismatch in phase is achieved by a local maximum-likelihood weighted minimum entropy algorithm. The simulated multi-channel in azimuth HRWS SAR data experiment is adopted to evaluate the performance of the proposed algorithm. Then, some real measured airborne multi-channel in azimuth HRWS Scan-SAR data is used to demonstrate the effectiveness of the proposed approach.