Generation of SAR Images Using Deep Learning

Synthetic aperture radar (SAR) is an imaging radar that produces high-resolution images. Classifying targets that are detected within a SAR image find many applications in environmental monitoring and mapping. In this project, an approach is proposed to classify targets that are detected in a SAR image. In this approach, CNN is applied to the data set to automatically extract features. These feature vectors are then used to train a softmax classifier. Examinations on the general MSTAR informational collection demonstrates that exactness of 99% can be accomplished when classification is done on two kinds of targets, and a precision of 98% on seven sorts of targets.

AbstractÑAiming at solving the problem in target recognition of synthetic aperture radar (SAR) images, this paper proposes a novel method based on the combination of Deep Convolutional Neural Network (DCNN) and Support Vector Machine (SVM). The class separation information, which explicitly facilitates intra-class compactness and inter-class separability in the process of learning features, is added to cross-entropy cost function as a regularization term to enhance the feature extraction ability of DCNN. Then the improved DCNN is applied to learn the features of SAR image. Finally, SVM is utilized to map the features into output labels. Experiments are performed on SAR image data in Moving and Stationary Target Acquisition and Recognition (MSTAR) database. The experiment results prove the effectiveness of our method, achieving an average accuracy of 99.15% on ten types of targets.

Journal of Scientific Research

In recent years, the importance of SAR (synthetic aperture radar) image analysis is growing day by day due to the vast applications in the field of oceanography, military war, land observation, agriculture, disaster management and geography. These applications required accurate classification of highresolution SAR images. In the present study, two different machine learning techniques are applied on the SAR dataset to analyze the classification accuracy. The first classification technique is SVM (support vector machine) along with principal component analysis (PCA) for feature reduction. While, in the second technique, a novel CNN (convolutional neural network) has been proposed to perform the classification on SAR images. Essential experimental analysis has been carried out based on the MSTAR dataset. Experimental results of the current study showed that the proposed CNN model outperformed the SVM classifier with an accuracy of 98.69%

Remote Sensing, 2017

Tremendous progress has been made in object recognition with deep convolutional neural networks (CNNs), thanks to the availability of large-scale annotated dataset. With the ability of learning highly hierarchical image feature extractors, deep CNNs are also expected to solve the Synthetic Aperture Radar (SAR) target classification problems. However, the limited labeled SAR target data becomes a handicap to train a deep CNN. To solve this problem, we propose a transfer learning based method, making knowledge learned from sufficient unlabeled SAR scene images transferrable to labeled SAR target data. We design an assembled CNN architecture consisting of a classification pathway and a reconstruction pathway, together with a feedback bypass additionally. Instead of training a deep network with limited dataset from scratch, a large number of unlabeled SAR scene images are used to train the reconstruction pathway with stacked convolutional auto-encoders (SCAE) at first. Then, these pre-trained convolutional layers are reused to transfer knowledge to SAR target classification tasks, with feedback bypass introducing the reconstruction loss simultaneously. The experimental results demonstrate that transfer learning leads to a better performance in the case of scarce labeled training data and the additional feedback bypass with reconstruction loss helps to boost the capability of classification pathway.

Journal of Advances in Information Technology

Demand towards the recognition of a target with a specific spatial signature by using remotely sensed images, the process of discovering the location, pose, and class that belongs to a particular kind of object is referred to as identification of the target. The progression of applying software and hardware to identify or recognize a goal from images of Synthetic Aperture Radar (SAR) outside the scope or within human availability is known as Automatic Target Recognition (ATR). The archaic architecture of ATR for SAR consists of three stages: recognition, distinctive, classification, and recognition. As1the time progresses many Deep1convolutional Neural Networks (DCNN) have been proposed and used for ATR-SAR and have obtained a state of the art results in many computer vision tasks, additionally shows subsequent result along the time, but most of them sort target from target chips found from SAR imagery, and used as a third stage (classification) of ATR-SAR archaic architecture. Also due to limited training images in ATR-SAR, DCNN yielded over-fitting when1directly applied to ATR-SAR. On the other hand, to make full use of SAR imagery, this paper presents Multi-View DCNN (MV-DCNN) for an end to end ATR-SAR which uses multiple views of SAR images. MV-DCNN takes1several views of a similar target. The proposed MV-DCNN proposed to instruct by the Moving1and Stationary Target Acquisition and1Recognition (MSTAR) benchmark dataset and to output scores of 10 number of classes. Index Terms-automated target recognition, synthetic aperture radar, multi-view deep CNN, MSTAR I. INTRODUCTION During 1903, Christian Hulsmeyer of Germany was the first person who granted the license from detecting objects by using radio waves to identify the presence of metallic objects but his invention called "Telemobiloscope" [1]. After few years around 1920s development of radar for ship and radar detection was started [2]. Before World War II, specialists in nations, for example, France, Britain, Germany, and Japan worked confidentially on creating advances that prompted the current-day version of radar.

Optical Engineering, 2020

We explore the blending of model-based and deep learning approaches for target recognition in inverse synthetic aperture radar (ISAR) imagery. It evaluates five different approaches, namely, a model-based geometric hashing approach, a supervised deep learning approach, and three different blending models that fuse the model-based and deep learning approaches. The model-based approach extracts scattering centers as features and requires domain experts to identify and characterize important features of a target, which makes the training process very costly and hard to generalize when the image quality degrades in low signal-tointerference-plus-noise-ratio conditions. Next, a deep learning algorithm using a convolutional neural network is considered to extract the spatial features when raw ISAR data are used as input. This approach does not need an expert and only requires the labels of images for training. Finally, three model-based and deep learning approaches are blended together at the feature level and decision level to benefit from the advantages of both approaches, achieving a higher performance. The results show that the blending of the two approaches achieves a high performance while providing explainable inferences. The performance of the five different approaches is evaluated under varying conditions of occlusion, clutter, masking of the target, and adversarial attacks. It is empirically shown that the model-based and deep learning approaches are able to complement each other and can achieve better classification accuracy upon fusing the integrated approach.

IRJET, 2020

In remote sensing, Synthetic Aperture Radar (SAR) images are well known for their all time and weather capabilities. SAR is a coherent radar imaging technology which is capable of producing high resolution images of targets and landscapes. Radar imaging applications are also susceptible to radar design constraints on bandwidth, acquisition time, number of antennas which affects the image resolution, this issues make the processing and interpretation of SAR images very difficult for human interpreters and computer vision systems. Single image super resolution (SISR), a greatly challenging task of computer vision and machine learning, attempts to reconstruct a high-resolution (HR) image from corresponding low image resolution. While image Super resolution (SR) is an ill-posed inverse procedure, since there exists a multitude of solutions for any LR input. To tackle this inverse problem, plenty of image SR algorithms have been proposed. Deep learning methods such as CNNs and GANs are being used to perform super resolution with results competitive to the state of the art. In this project will investigate whether the Deep learning (GANs & CNNs) can be used to achieve super resolution in SAR image.

HAL (Le Centre pour la Communication Scientifique Directe), 2020

In this paper, we introduce a new generative adversarial network (GAN) with dual image-color discriminators, to predict Artificial-SAR colorized images from SAR ones (Sentinel-1). Based on the conventional architecture of GANs, we employ an additional color discriminator that evaluates the differences in brightness, contrast, and major colors between images, while the image discriminator compares texture and content. To achieve the required level of colorization in the generation process, we employ non-adversarial color loss dedicated for color comparison, unlike conventional approaches that use only L 1 loss. Moreover, to overcome the vanishing gradient problem in deep architecture, and ensure the flow of low-level information inside network layers, we add residual connections to our generator that follows the general shape of U-Net. The performance of the proposed model was evaluated quantitatively as well as qualitatively with the SEN1-2 dataset. Results show that the proposed model generates realistic colorized images with fewer artifacts compared to the state-of-the-art approaches. This model helps to maintain color steadiness as well as visual recognizability at less textured large continuous regions, such as plantation and water areas, when it's difficult to be distinguished in SAR images.

Magnetic Resonance in Medicine, 2019

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Journal of Scientific Research

Synthetic Aperture Radar (SAR) target classification is one of the largest branches of SAR image analysis. Despite the remarkable achievements of deep learning-based SAR target prediction algorithms, current object recognition algorithms are limited in terms of military applications. Acquisition and labeling of SAR target images are time-consuming and cumbersome. Obtaining adequate training data is also challenging in many cases. Deep learning-based models are always susceptible to overfitting because of insufficient training data. This limitation prevents them from being widely used to classify SAR targets. To overcome the problem of insufficient sampling and to learn more accurate representations for SAR image recognition, we propose a two way input of SAR images into a dual stream of DCNN. Concatenating two input SAR images representations was done using the restricted raw SAR data in order to extract the integral features from the 2 input SAR images representations for classific...