Deep Neural Networks (DNNs) based Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) systems have shown to be highly vulnerable adversarial perturbations that are deliberately designed yet almost imperceptible but can bias DNN inference when added targeted objects. This leads serious safety concerns applying DNNs high-stakes SAR ATR applications. Therefore, enhancing the robustness of is essential for modern real-world systems. Toward building more robust DNN-based models,...

Recent advances of deep neural networks (DNNs) highlight the success on synthetic aperture radar automatic target recognition (SAR ATR) with superiority effectiveness and efficiency. However, DNNs are known to be vulnerable adversarial examples, whose performance will dramatically reduced when imperceptible perturbation exists. In optical image processing, invisible perturbations typically embedded in way a full-scaled distribution purely digital setting. Whereas, it is not feasible achieve...

Synthetic Aperture Radar (SAR) stands as an indispensable sensor for Earth observation, owing to its unique capability all-day imaging. Nevertheless, in a data-driven era, the scarcity of large-scale datasets poses significant bottleneck advancing SAR automatic target recognition (ATR) technology. This paper introduces NUDT4MSTAR, dataset vehicle wild, including 40 types and wide array imaging conditions across 5 different scenes. NUDT4MSTAR represents leap forward scale, containing over...

The existence of adversarial examples causes serious security risks when deep neural networks are applied to synthetic aperture radar (SAR) target detection. In SAR image processing, the added small disturbances can cause model output incorrect predictions. Due multipath effect in propagation detection signals, there complex interactions between targets and their surroundings serving as supportive clues for manifested tight correlations pixels contextual information (where context refers...

It has been demonstrated that deep neural network (DNN)-based synthetic aperture radar (SAR) automatic target recognition (ATR) techniques are extremely susceptible to adversarial intrusions, is, malicious SAR images including deliberately generated perturbations imperceptible the human eye but can deflect DNN inference. Attack algorithms in previous studies based on direct access a ATR model such as gradients or training data generate examples for image, which is against non-cooperative...

Adversarial examples (AEs) bring increasing concern on the security of deep-learning-based synthetic aperture radar (SAR) target recognition systems. SAR AEs with perturbation constrained to vicinity have been recently in spotlight due physical realization prospects. However, current adversarial detection methods generally suffer severe performance degradation against region-constrained perturbation. To solve this problem, we treated as low-probability samples incompatible clean dataset....

The deep neural networks (DNNs) have freed the synthetic aperture radar automatic target recognition (SAR ATR) from expertise-based feature designing and demonstrated superiority over conventional solutions. There has been shown unique deficiency of ground vehicle benchmarks in shapes strong background correlation results DNNs overfitting clutter being non-robust to unfamiliar surroundings. However, gap between fixed model training varying application remains underexplored. This letter...

Electrical drawings, consisting of electrical symbol elements, connection lines and text annotations, describe the layout equipment. However, power system technicians often have to manually redraw them from image format digital files, which is challenging. Here we propose EDRS—a deep learning based automatic Drawings Recognition System for drawings. Specifically, first develop a Faster- RCNN model accomplish subtask symbols recognition. Then provide detection recognition recognize content...

This work investigated the nonlinear behaviors of disk resonator. Generally, nonlinearity occurs in resonators with small stiffness, such as cantilever, bridge, etc. However, it was observed that radial-contour-mode micromechanical resonator large stiffness and high quality factor (Q) also suffers when a bias voltage is applied. The shows linear response 10 V atmosphere room temperature. vibrates vacuum or at low temperature (below 110 K), even 6 V. effect closely related to change factor,...

Transfer-based targeted adversarial attacks against black-box deep neural networks (DNNs) have been proven to be significantly more challenging than untargeted ones. The impressive transferability of current SOTA, the generative methods, comes at cost requiring massive amounts additional data and time-consuming training for each label. This results in limited efficiency flexibility, hindering their deployment practical applications. In this paper, we offer a self-universal perspective that...

Synthetic aperture radar (SAR) image classification is a challenging problem due to the complex imaging mechanism as well random speckle noise, which affects interpretation. Recently, deep neural networks (DNNs) have been shown outperform previous state-of-the-art techniques in computer vision tasks owing their ability learn relevant features from data. However, fragility of these models has received far less academic attention remote sensing community, limits our understanding security...

With the rapid development of edge computing, clusters need to deal with a tremendous amount tasks, making some overloaded, which further translates into task completion lag. Previous works usually copy tasks from overloaded edges idle so as reduce queuing and computing delay. However, delay copied different cannot be predicted before replication decision is made, affects overall performance. In this paper, we propose an online algorithm based on predictions derived multi-armed bandit. Via...

With the increasing increments on Deep Learning (DL) in these years, attentions have been arose adversarial attack DL models for Synthetic Aperture Radar (SAR) target recognition. However, most of previous methods transfer exisiting from computer vision area which typically by generating background-located perturbation certain size. respect to physical realizability SAR recognition tasks, is only feasibly generated region. Therefore, this paper, we propose a Target Segmentation based...

The deep neural networks (DNNs) have freed the synthetic aperture radar automatic target recognition (SAR ATR) from expertise-based feature designing and demonstrated superiority over conventional solutions. There has been shown unique deficiency of ground vehicle benchmarks in shapes strong background correlation results DNNs overfitting clutter being non-robust to unfamiliar surroundings. However, gap between fixed model training varying application remains underexplored. Inspired by...

The stealthiness of frequency-hopping (FH) signals increases the difficulty reconnaissance in low SNR environment. In this paper, we propose a FH signal detection algorithm based on time-frequency matrix, which can make full use unique features matrix signals, even if are submerged high-power noise. addition, inspired by idea cancellation, utilize cancellation to remove fixed frequency interference. Simulation analysis shows that proposed has good anti-noise uncertainty and performance...

Latest developments in Deep Neural Networks (DNNs) emerge the ranking efficacy and efficiency of Synthetic Aperture Radar Automated Target Recognition (SAR ATR). However, it is difficult to collect SAR datasets various scenes due high cost non-cooperative property. The current DNN-based ATR algorithms are thus trapped into overfitting target acquisition environment robustly identify unknown background clutters. Based on Shapley value, which designed attribute contribution each input...

Deep neural network classifiers are susceptible to be deceived by adversarial examples made attackers, resulting in misclassification. Although the white-box attack performance of attacks is very high, transferability against unknown model always low, and enhancement often brings about an increase magnitude perturbations, concealment weakened. In this paper, we propose a new multi-level smoothing filter network, which embedded iteration algorithms enhance continuity adjacent pixels...

Deep Neural Networks (DNNs) based Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) systems have shown to be highly vulnerable adversarial perturbations that are deliberately designed yet almost imperceptible but can bias DNN inference when added targeted objects. This leads serious safety concerns applying DNNs high-stake SAR ATR applications. Therefore, enhancing the robustness of is essential for implementing modern real-world systems. Toward building more robust DNN-based...