Iet Radar Sonar and Navigation最新文献

Degrees of freedom (DOF) serves as a critical metric for evaluating the design of sparse arrays. Developing novel sparse arrays with enhanced degrees of freedom and mathematically expressible structures constitutes a significant research direction in the field of direction of arrival (DOA) estimation. In this paper, an innovative V-shaped sparse sensor array is proposed through the strategic adjustment of sensor positions within the array which is called V-shaped fourth-order linear array (VFLA). Compared to existing V-shaped sparse arrays, the proposed configuration demonstrates superior degrees of freedom when exploiting the covariance of received signals for DOA estimation. Furthermore, relative to commonly used V-shaped arrays and other sparse arrays, the VFLA exhibits not only higher degrees of freedom but also a larger array aperture, thereby enhancing the accuracy of two-dimensional (2-D) DOA estimation. Finally, simulation experiments validate the outstanding performance of the VFLA in 2-D DOA estimation.

Recent developments in driving technology have led to the creation of advanced driver assistance systems and progress towards fully autonomous vehicles. Cars equipped with radar technology can simultaneously detect multiple vulnerable road users, assessing their distance, speed, and approach angle. For autonomous vehicles to be deemed safe for public roads, they must effectively identify and classify these users. This study employs time–frequency analysis and deep learning techniques to classify spectrograms derived from targets. The training and testing datasets were generated using frequency-modulated continuous-wave (FMCW) radar signals operating at 77 GHz. A five-layer convolutional neural network (CNN) was trained for this purpose. We investigated how different time window types and durations affect the Short-Time Fourier Transform calculation and the CNN classification accuracy for each scenario. As the length of the time window increases, frequency resolution improves, enabling better differentiation between closely spaced frequencies and enhancing classification accuracy. However, increased time window lengths lead to decreased time resolution, causing accuracy to plateau at 800; beyond this point, accuracy declines. We achieved an accuracy rate of 88.95% in classifying seven data classes, with improvements in specific classes compared to prior studies. The findings suggest that micro-Doppler-based convolutional neural networks can effectively classify vulnerable road users, contributing to collision avoidance efforts.

This article discusses the problem of tracking a single 3D extended target (or widely separated targets) with convex polytope shape when the target may only be partially visible. An extended target (as opposed to a point target) may generate multiple measurements in a single frame. With the advent of high-resolution sensors (such as LiDAR), the targets need to be considered as extended targets and their shape as well as kinematics need to be estimated. The extended target may only be partially visible (self-occlusion) and the measurements occur only from the visible parts of the target. In this work, different parts of a single extended target are assumed to be different targets constrained by the rigid body motion of the whole target, and the multitarget tracking framework is used to handle the tracking. The target shape is described using a convex hull represented by its vertices and a Delaunay triangulation. The point target PMHT is modified to develop an extended target PMHT (ET-PMHT) joint association and filtering by assuming that the face triangulations are separate targets. Face management is incorporated into the algorithm to delete erroneous faces and the algorithm is able to add new faces to refine the shape estimate. The framework can handle self-occlusion (partial visibility) by associating measurements only to the visible parts of the target. The algorithm's performance is compared with the 3D Gaussian Process under various scenarios, and RMSE of the centre, velocity and IoU metrics are used to quantify the performance. The proposed algorithm is able to outperform the 3D Gaussian Process in the centre RMSE metric by about 40% while achieving an IoU of 0.6 (on average) even when the target is only partially visible.

Image forgery detection aims to identify tampered content and localise manipulated regions within images. With the rise of advanced editing tools, forgeries pose serious challenges across media, law and scientific domains. Existing CNN-based models struggle to detect subtle manipulations that mimic natural image patterns. To address this challenge, we propose a dual-stream contrastive learning network (DSCL-Net) that jointly exploits spatial (pixel-level) and frequency (noise-level) cues. The architecture employs two ResNet-50 encoders: one processes the red–green–blue (RGB) image to capture semantic context, whereas the other processes a spatial rich model (SRM) filtered version to extract high-frequency forensic traces. A multi-scale attention fusion module enhances manipulation-sensitive features. The network includes three heads: a classification head for image-level prediction, a segmentation head for pixel-wise localisation, and a contrastive projection head to improve feature discrimination. We validate the effectiveness of our proposed model on two benchmark datasets. The proposed DSCL-Net surpasses previous state-of-the-art methods by achieving an image-level accuracy of 97.9% on the CASIA and 89.8% on IMD2020. At the pixel level, it attains an F1-score of 92.7% and an AUC of 91.2% on CASIA, and an F1-score of 86.6% with an AUC of 90.1% on IMD2020. Furthermore, LIME and SHAP have been employed to provide explainability at individual image level to comprehend the alignment of the predicted mask with the ground truth mask. The developed approach contributes to the development of safe technology for dealing with misinformation and fake news.

Marine environmental pollution, particularly from oil spills, has garnered significant attention due to its irreversible damage to marine ecosystems. Ship-borne high-frequency surface wave radar (HFSWR) holds promise for long-distance, wide-area marine environment monitoring, enabling real-time surveillance of oil pollution on the sea surface. This paper utilises two sets of ship-borne HFSWR to swiftly deploy and monitor oil spill areas through optimal deployment planning, specifically tailored for addressing oil spill incidents in designated sea surface regions. First, this paper outlines the deployment model for two sets of ship-borne HFSWR, which is based on quadrilateral monitoring areas and circular deployment regions for transmitting and receiving stations. Then, this paper presents a traversal algorithm that operates under the minimum resource parameter limit, followed by a fast algorithm derived from geometric relationships with delineating the scope of application. Theoretical and experimental results demonstrate that the proposed algorithm significantly reduces the computational complexity of the traversal algorithm while maintaining high accuracy.

Synthetic Aperture Radar (SAR) image classification using deep neural networks (DNNs) has demonstrated vulnerability to adversarial attacks, particularly black-box attacks, which rely solely on model output scores to craft effective perturbations. Despite their practical threat, defences against such attacks in SAR tasks remain underexplored. To bridge this gap, we propose a novel defence mechanism that introduces a pointwise modulation layer to enforce gradient orthogonality, thereby disrupting the gradient estimation process employed in black-box attacks. This method preserves high accuracy on clean data by maintaining logit consistency while significantly reducing attack success rates. Furthermore, the approach is computationally efficient and can be easily integrated into existing models. Extensive experiments demonstrate the effectiveness of the proposed method in enhancing the robustness of SAR classifiers against a range of black-box attack scenarios, without compromising their performance on clean data. This work contributes to the development of secure and reliable SAR-based machine learning systems for critical applications.

With the U-space revolution, drones are going to reshape both the physical space and the cyberspace of the future urban environment, also with the support of autonomy and artificial intelligence (AI). However, this revolution comes with the cost of new multi-domain risks, which may be traced back to cyber and physical threats within drone-based new entrants. A proper assessment and treatment of these risks is essential to achieve the safety and security objectives of U-space for the drone ecosystem. This will entail further research, especially for the analysis of drone intruders and for the mitigation of the related U-space impacts. This work proposes a concept for improving the U-space resilience through a novel AI-centric service, named DARS (drone attack resilience service), focused on managing unauthorised operations of intruder drones in the physical and cyber domains. DARS-related threat scenarios and risk-assessment capabilities are discussed, resorting also to modelling specific drone cyber-physical attacks. A detailed analysis of DARS AI-centric functional architecture is provided, with a survey of the potential approaches for intruder trajectory prediction and intent recognition, to be used for the next design stages. Lastly, the work provides a preliminary analysis of how the neutralisation functions could be implemented in DARS.

The SAR-ATR (Synthetic Aperture Radar - Automatic Target Recognition) system based on deep learning technology has been proven to have a target recognition vulnerability—adversarial examples, which has attracted widespread attention. However, existing adversarial sample attacks focus primarily on the image domain, neglecting the unique characteristics of SAR imaging and the challenges of transferring attacks to the physical domain. In response, we propose a physically realisable adversarial attack method based on radar imaging principles and the Attribute Scattering Centre Model (ASCM), which aims to translate perturbations from the digital image domain to modifications of physical electromagnetic parameters of radar. The ASCM method consists of three key components: (1) reconstructing the backscattered signal to physical scattering centres using ASCM, (2) establishing a minimal perturbation optimisation model under $��{\ell }_0�$-norm constraints to restrict perturbations to scattering centres, and (3) applying the Monte Carlo Method (MCM) to determine optimal adjustment points and amounts for scattering centre amplitude parameters. Experimental results demonstrate that the proposed method achieves the highest success rate of 96.25% for nontargeted attacks and 88.89% for targeted attacks, with the potential for extension to the physical domain to generate high-success-rate adversarial attack effects.

Accurate classification of radar signals remains a key challenge in automatic modulation classification (AMC), particularly in scenarios with limited training data and complex signal variations. To address this, we propose a novel hybrid neural architecture and incorporate a magnitude rescaling method for data augmentation. Specifically, our hybrid neural structure integrates a bidirectional long short-term memory (Bi-LSTM) network, a dynamic feature extraction module, and a transformer encoder in a cascaded structure. It effectively processes one-dimensional signals enhanced via the proposed random magnitude rescaling method. Experimental results demonstrate our approach achieves a competitive classification accuracy of 94.18% on the RML2016a data set and exhibits strong performance on a hardware-in-the-loop simulation dataset. The implementation of our radar signal modulation classification method, along with the related datasets, is available at: https://github.com/stu-cjlu-sp/rsrc-for-pub/tree/main/ASEFEAMC.

In this work, we generalise the popular Kalman filter and Linear Quadratic Gaussian controller for use on multi-sensor and multi-agent/-target radar systems. The state-space representation for the dynamical evolution of targets and the sensor measurements is developed here using tensors in place of vectors and matrices, producing a multilinear dynamical system. In this dynamical framework, the tensor forms of the Kalman filter and the Linear Quadratic Gaussian controller are developed, allowing the simultaneous processing of (i) the inputs of all sensors, producing the estimation of the state of all agents/targets and (ii) the determination of the optimal control actions of all agents/targets. These tools are applied to implement optimal parallel waveform design and tracking control for a multi-radar system acting on multiple agents. In the study case, examined numerically, the radars can (i) estimate the state of the agents in terms of range, angular displacement, radial and angular velocities and (ii) jointly determine the agents control inputs and the radars transmitted waveforms to minimise the control cost action and the energy of the transmitted signals.