SR-ISTA-Net: Sparse Representation-Based Deep Learning Approach for SAR Imaging

Sparse-representation-based synthetic aperture radar (SAR) imaging technology has shown superior potential in the reconstruction of nonsparse scenes. However, many existing compressed sensing (CS) methods with sparse representation cannot obtain an optimal sparse basis and only apply to the sensing matrix obtained by exact observation, resulting in a low image quality occupying more storage space. To reduce the computational cost and improve the imaging performance of nonsparse scenes, we formulate a deep learning SAR imaging method based on sparse representation and approximated observation deduced from the chirp-scaling algorithm (CSA). First, we incorporate the CSA-derived approximated observation model and a nonlinear transform function within a sparse reconstruction framework. Second, an iterative shrinkage threshold algorithm is adopted to solve this framework, and the solving process is unfolded as a deep SAR imaging network. Third, a dual-path convolutional neural network (CNN) block is designed in the network to achieve the nonlinear transform, dramatically improving the sparse representation capability over conventional transform-domain-based CS methods. Last, we improve the CNN block to develop an enhanced version of the deep SAR imaging network, in which all the parameters are layer-varied and trained by supervised learning. The experiments demonstrate that our proposed two imaging networks outperform conventional CS-driven and deep-learning-based methods in terms of computing efficiency and reconstruction performance of nonsparse scenes.

Compressed sensing (CS) is currently utilized in the application of Synthetic Aperture Radar (SAR) imaging. Unfortunately, the CS SAR imaging performance degrades as SNR decreases. However, there exists a denoising convolutional neural network (DnCNN) based on 2D convolution (DnCNN-2D) in the field of image denoising. Traditional DnCNN-2D is not directly applicable to SAR imaging. Thus, a DnCNN based on 1D convolution (DnCNN-1D) for signal denoising is proposed and employed to denoise the SAR signal. The denoised SAR signal is subsequently utilized for CS SAR imaging. Each convolutional layer of DnCNN-1D has a fixed number and fixed size of filters and activation functions. The residual learning strategy and batch normalization are adopted in the DnCNN-1D. The segmental signal-to-noise ratio (SEGSNR) and the normalized mean square error (NMSE) are selected to evaluate signal denoising and SAR imaging performance, respectively. The experimental results indicate that imaging performance is significantly improved after denoising the SAR signal by the proposed method.

In this paper, alternative direction finding methods have been proposed using deep learning techniques. Firstly, Regeression and Classification models have created by using Convolutional Neural Networks (CNNs). In the second Convolutional Neural Networks and Recurrent Neural Networks (RNNs) have been utilized in the proposed methods. Despite having vast amount of direction finding studies, utilization of neural networks is scarce in literature and past works mostly only includes usage of CNNs. In this study, direction finding is performed by learning signals reaching multiple antenna arrays by networks. Created neural networks have been fed with different data formats and their performances against noisy and no-noise data have been shown. In addition, comparative analysis of the developed models were made in the similar Signal-to-Noise Ratio (SNR) range with the subspace based MUSIC algorithm, which is frequently used in direction finding.

As an active microwave imaging technology, synthetic aperture radar (SAR) has extensive application prospects in remote sensing, surveying, and mapping. How to reconstruct SAR images from the subsampled echo data has always been a challenging issue. Compressed sensing technology attempts to use the sparse characteristics of SAR images to recover the SAR images from the compressed samples. However, in practical applications, SAR images do not always meet the characteristic of sparsity. In this paper, we propose an imaging method from sub-Nyquist sampled data based on deep priors of SAR images. First, we use a generative flow network to model the deep prior information of the images based on the existing SAR image datasets. Then, the pre-trained network modeling deep prior information of images is embedded in a typical compressed sensing method, i.e., the Iterative Shrinkage-Thresholding Algorithm (ISTA), to replace the sparse shrinkage function of it. Meanwhile, in order to improve the accuracy and convergence speed of reconstruction, we used the subsampled echo data to fine-tune the iterative parameters of the embedded ISTA method. The results of the experiments show that through the use of the deep prior information of the images, this method can accurately reconstruct non-sparse SAR images from the subsampled echoes, even if only a few echo samples are available.

The study presents the results of the development and testing of deep learning neural network architectures, which demonstrate high accuracy rates in classifying neurophysiological data, in particular hemodynamic brain activation patterns obtained by functional near-infrared spectroscopy, during solving mathematical problems on spatial-numerical associations. The analyzed signal represents a multidimensional time series of oxyhemoglobin and deoxyhemoglobin dynamics. Taking the specificity of the fNIRS signal into account, a comparative analysis of 2 types of neural network architectures was carried out: (1) architectures based on recurrent neural networks: recurrent neural network with long short-term memory, recurrent neural network with long short-term memory with fully connected layers, bidirectional recurrent neural network with long short-term memory, convolutional recurrent neural network with long short-term memory; (2) architectures based on convolutional neural networks with 1D convolutions: convolutional neural network, fully convolutional neural network, residual neural network. Trained long short-term memory recurrent neural network architectures showed worse results in accuracy in comparison with 1D convolutional neural network architectures. Residual neural network (model_Resnet) demonstrated the highest accuracy rates in three experimental conditions more than 88% in detecting age-related differences in brain activation during spatial-numerical association tasks considering the individual characteristics of the respondents’ signal.

Synthetic aperture radar (SAR) image target detection and recognition (SAR-TDR) tasks have become research hot spots in the remote sensing application. These targets include ships, vehicles, aircraft, oil tanks, bridges, and so on. However, with the rapid development of SAR technology and increasingly complex electromagnetic environment, complex characteristics of SAR images bring severe challenges to the accurate SAR-TDR via traditional physical models or manual feature-extraction-based machine learning methods. In recent years, deep learning (DL), as a powerful automatic feature extraction algorithm, has been widely used in the computer vision domain. More specifically, DL has also been introduced into the SAR-TDR tasks and has effectively achieved good performance in terms of accuracy, real-time processing, etc. With the rapid development of DL, SAR image processing, and practical requirements of SAR-TDR in civilian and military domains, it is crucial to conduct a systematic survey on SAR-TDR in the past few years. In this survey article, we mainly conduct a systematic overview of DL-based SAR-TDR literature on two tasks, i.e., target recognition (e.g., ground vehicles, ships, and aircraft) and target detection (e.g., ships, aircraft, change detection, sea surface oil spills, and oil tanks). More specifically, our related works about these topics are also presented to verify the effectiveness of DL-based methods. First, several DL methods (e.g., convolutional neural networks, recurrent neural networks, autoencoders, and generative adversarial networks), commonly used in SAR-TDR, are briefly introduced. Then, a systematic review of DL-based SAR-TDR (including our related works) is presented. Finally, the current challenges and future possible research directions are deeply analyzed and discussed.

In synthetic aperture radar (SAR) images, inland water bodies generally appear as dark textures, but due to the interference of thermal noise, speckle noise, and azimuth blur, inland water bodies can be easily confused with weakly scattered areas such as roads, wetlands, and sand. Most of the existing denoising methods have been proposed for removing speckle noise. However, the presence of thermal noise and azimuth blur still makes it difficult to distinguish inland water bodies. The denoising method based on the Doppler spectrum can effectively remove thermal noise, speckle noise, and azimuth blur, but it reduces the resolution, which blurs the contour of the water body, and it is difficult to apply to the detection and processing of small inland water bodies. Fusing a noisy high-resolution image with a low-resolution image with less noise makes it possible to remove the speckle noise and thermal noise of the SAR image while maintaining the edge texture of the water body. Based on the idea of image fusion, sparse representation is used to fuse the original noisy SAR image (HR) and the low-precision image (LR) processed by the denoising algorithm to maximise the use of all the information in HR and LR. According to this idea, an enhancement algorithm for separability of inland water body based on sparse representation is proposed. The experimental results show that the proposed model significantly improves the separability of inland water bodies in SAR images.

Convolutional Neural Network (CNN) has shown great potential for structural information extraction, however, Synthetic Aperture Radar (SAR) image interpretation via CNN, is still a challenging task due to the pixel- to- pixel variation nature of SAR signal. In this paper, a pattern strengthened deep model for SAR image classification is presented. The main idea underlying this method is to strengthen (or sharpen) the salient patterns contained in SAR image while attenuate the random variation, so as to leverage the enormous potential of deep model. To achieve this goal, max pooling is elaborately designed to form a pattern strengthened deep model. The experimental results on two SAR datasets show that the proposed method increases the classification accuracy by 4.91% and 4.64% compared with the typical CNN, and indicate our innovative attempt is promising for SAR image classification in the case of limited training data.

ABSTRACTAutomatic Target Recognition (ATR) based on Synthetic Aperture Radar (SAR) images plays a key role in military applications. However, there are difficulties with this traditional recognition method. Principally, it is a challenge to design robust features and classifiers for different SAR images. Although Convolutional Neural Networks (CNNs) are very successful in many image classification tasks, building a deep network with limited labeled data remains a problem. The topologies of CNNs like the fully connected structure will lead to redundant parameters and the negligence of channel-wise information flow. A novel CNNs approach, called Group Squeeze Excitation Sparsely Connected Convolutional Networks (GSESCNNs), is therefore proposed as a solution. The group squeeze excitation performs dynamic channel-wise feature recalibration with less parameters than squeeze excitation. Sparsely connected convolutional networks are a more efficient way to operate the concatenation of feature maps from different layers. Experimental results on Moving and Stationary Target Acquisition and Recognition (MSTAR) SAR images, demonstrate that this approach achieves, at 99.79%, the best prediction accuracy, outperforming the most common skip connection models, such as Residual Networks and Densely Connected Convolutional Networks, as well as other methods reported in the MSTAR dataset.

Thanks to the availability of large-scale data, deep Convolutional Neural Networks (CNNs) have witnessed success in various applications of computer vision. However, the performance of CNNs on Synthetic Aperture Radar (SAR) image classification is unsatisfactory due to the lack of well-labeled SAR data, as well as the differences in imaging mechanisms between SAR images and optical images. Therefore, this paper addresses the problem of SAR image classification by employing the Generative Adversarial Network (GAN) to produce more labeled SAR data. We propose special GANs for generating SAR images to be used in the training process. First, we incorporate the quadratic operation into the GAN, extending the convolution to make the discriminator better represent the SAR data; second, the statistical characteristics of SAR images are integrated into the GAN to make its value function more reasonable; finally, two types of parallel connected GANs are designed, one of which we call PWGAN, combining the Deep Convolutional GAN (DCGAN) and Wasserstein GAN with Gradient Penalty (WGAN-GP) together in the structure, and the other, which we call CNN-PGAN, applying a pre-trained CNN as a discriminator to the parallel GAN. Both PWGAN and CNN-PGAN consist of a number of discriminators and generators according to the number of target categories. Experimental results on the TerraSAR-X single polarization dataset demonstrate the effectiveness of the proposed method.

This paper proposes a novel deep learning architecture involving combinations of Convolutional Neural Networks (CNN) layers and Recurrent neural networks (RNN) layers that can be used to perform segmentation and classification of 5 cardiac rhythms based on ECG recordings. The algorithm is developed in a sequence to sequence setting where the input is a sequence of five second ECG signal sliding windows and the output is a sequence of cardiac rhythm labels. The novel architecture processes as input both the spectrograms of the ECG signal as well as the heartbeats' signal waveform. Additionally, we are able to train the model in the presence of label noise. The model's performance and generalizability is verified on an external database different from the one we used to train. Experimental result shows this approach can achieve an average F1 scores of 0.89 (averaged across 5 classes). The proposed model also achieves comparable classification performance to existing state-of-the-art approach with considerably less number of training parameters.

The presented paper proposes a new speech command recognition model for novel engineering applications with limited resources. We built the proposed model with the help of a Convolutional Recurrent Neural Network (CRNN). The use of CRNN instead of Convolutional Neural Network (CNN) helps us to reduce the model parameters and memory requirement as per resource constraints. Furthermore, we insert transmute and curtailment layer between the layers of CRNN. By doing this we further reduce model parameters and float number of operations to half of the CRNN requirement. The proposed model is tested on Google’s speech command dataset. The obtained result shows that the proposed CRNN model requires 1/3 parameters as compared to the CNN model. The number of parameters of the CRNN model is further reduced by 45% and the float numbers of operations between 2% to 12 % in different recognition tasks. The recognition accuracy of the proposed model is 96% on Google’s speech command dataset, and on laboratory recording, its recognition accuracy is 89%.

Compressive Sensing(CS) has been widely utilized in Inverse synthetic aperture radar(ISAR) imaging since real ISAR data is easier to be non-completed, and CS-based methods can obtain high-quality imaging results using under-sampled data. However, traditional CS-based methods need pre-defined parameters, sparse transforms and iterative reconstruction processes. Optimal parameters as well as transforms are tough to be hand-crafted, and iterative reconstruction consumes plenty of time, which limit practical applications in ISAR imaging. Given that Convolution Neural Network(CNN) has great power to learn rapidly, we compose CNN with traditional Iterative Shrinkage-Thresholding Algorithm(ISTA) to propose CNN-ISTA(CIST)-based ISAR imaging method. CIST is capable of learning optimal parameters and transforms throughout the training (i.e. the optimization process is interpretable) instead of manually defined. Compared with traditional state-of-the-art CS imaging methods, the experimental results demonstrate that our proposed CIST-based imaging method is superior in both imaging quality and computational efficiency.

Synthetic aperture radar (SAR) images are corrupted by speckle noise due to random interference of electromagnetic waves. In this paper, we proposed a speckle reduction technique based on sparse representation and dictionary learning. Firstly, an adaptive dictionary was learned by performing KSVD algorithm through a large amount of training patches extracted from the noisy SAR image. Considering the inaccurate recovery of point targets which is brought by the inadequate number of training samples, we employed a point target enhancing scheme to highlight the important point targets in the SAR image. Some experiments were conducted on real SAR images, and the results shows that our proposed algorithm can effectively reduce the speckle noise as well as preserve details. Some comparisons are made to prove its superiority to the available algorithms.

Synthetic Aperture Radar (SAR) images are prone to be contaminated by noise, which makes it very difficult to perform target recognition in SAR images. Inspired by great success of very deep convolutional neural networks (CNNs), this paper proposes a robust feature extraction method for SAR image target classification by adaptively fusing effective features from different CNN layers. First, YOLOv4 network is fine-tuned to detect the targets from the respective MF SAR target images. Second, a very deep CNN is trained from scratch on the moving and stationary target acquisition and recognition (MSTAR) database by using small filters throughout the whole net to reduce the speckle noise. Besides, using small-size convolution filters decreases the number of parameters in each layer and, therefore, reduces computation cost as the CNN goes deeper. The resulting CNN model is capable of extracting very deep features from the target images without performing any noise filtering or pre-processing techniques. Third, our approach proposes to use the multi-canonical correlation analysis (MCCA) to adaptively learn CNN features from different layers such that the resulting representations are highly linearly correlated and therefore can achieve better classification accuracy even if a simple linear support vector machine is used. Experimental results on the MSTAR dataset demonstrate that the proposed method outperforms the state-of-the-art methods.