The majority of radiation field mixed source location methods that are resistant to array manifold errors demonstrate a robust inhibition effect on a single array manifold error and exhibit commendable source location estimation ability. However, multiple array manifold errors are frequently encountered in practical application environments, and the location performance of the aforementioned antiarray manifold error method is observed to decline rapidly. The objective of this paper is to present a study of the location method of a radiation field mixed source with amplitude and phase error and array position error. This paper proposes an auxiliary source and noise subspace power iteration (AS-NSPI) method, which considers the specific impact of array manifold error on radiated far-field sources. Two auxiliary far-field sources with different frequencies are employed to estimate the amplitude and phase error matrix and the array position error. Subsequently, the covariance matrix of the array received data is compensated for by the results of the error estimation. Thereafter, the noise subspace is calculated by noise subspace power iteration (NSPI), and the source location is estimated by space spectrum search. The simulation results demonstrate that the AS-NSPI can circumvent the impact of multiarray manifold errors on source location and exhibits superior estimation accuracy.
In thin-film transistor (TFT) array manufacturing, material inconsistencies and process imperfections frequently induce defects that severely compromise production yield. Traditional manual inspection remains labor-intensive and susceptible to subjective bias, thereby limiting manufacturing capacity. While convolutional neural networks (CNNs) have shown promise in automating anomaly detection, the application of vision transformers (ViTs) in this domain remains relatively unexplored. This study introduces Gaussian Process Kolmogorov–Arnold Transformer (GP-KAT), a structure-aware architecture that combines the nonlinear representation capacity of the KAT with GP-inspired locally periodic kernel modeling and variance-aware anomaly prediction. To address the inherent periodicity of TFT array imagery, we propose a GPKAN layer that incorporates a locally periodic Gaussian kernel to enhance structural feature extraction. Furthermore, a feature variance augmented head is designed to explicitly quantify predictive uncertainty, facilitating more reliable anomaly scoring under ambiguous conditions. Extensive experiments on a newly curated industrial dataset show that GP-KAT outperforms representative ViT-based baselines under a unified training protocol. By improving detection accuracy and providing informative uncertainty estimates under the tested conditions, the proposed approach offers a promising direction for TFT array inspection.
As an emerging coherent jamming technique, interrupted sampling repeater jamming (ISRJ) can effectively disrupt radars employing linear frequency modulated (LFM) signals through mainlobe deception. To counter this critical threat, theoretical analysis of ISRJ principles is conducted, and an analytical model applicable to intrapulse modulated radar waveforms is developed. From multiple perspectives, including signal processing, system operation modes, and system-wide cooperative detection, interference suppression methods incorporating biphase-coded/slope-agile waveform scheduling, slope diversity, and external information support are designed. By integrating these methods, a comprehensive anti-jamming strategy is formulated and validated through simulations and experiments. The proposed approach leverages radar’s preemptive advantage and systemic capabilities without modifying the existing signal processing architecture, offering strong interference suppression, real-time adaptability, independence from prior jamming parameters, and low implementation complexity.
This study proposes an L-shaped super augmented nested array (LSANA) structure for two-dimensional (2-D) direction of arrival (DOA) estimation. The proposed LSANA structure consists of two linear arrays located along the x-axis and y-axis, respectively. Each linear array is a SANA that achieves lower mutual coupling (MC) and higher degrees of freedom (DOF). The azimuth and elevation angles are estimated separately for each part of the array, and then the pairing of these angles is achieved using the cross-covariance matrices of the respective data sections, thus enabling 2-D DOA estimation. Finally, through simulation experiments, the proposed LSANA is compared with other sparse array structures. The results indicate that the proposed LSANA can achieve higher DOF with lower MC, while also providing improved estimation accuracy for DOA.
Pediatric wrist fractures are the most common pediatric traumatic injuries, while manual X-ray diagnosis has high missed diagnosis rates and strong subjectivity in emergency settings. Existing deep learning models face core clinical bottlenecks, including low recall for subtle fractures, poor robustness to low-quality images, and unbalanced accuracy and inference efficiency. We propose WFYOLO, an enhanced YOLO11-based algorithm for pediatric wrist fracture detection, with a multi-scale edge enhancement (MSE) module, lightweight slim-neck, dynamic head (DYHead), and class-weighted loss to address the above limitations. All experiments are conducted on the official standard split of the GRAZPEDWRI-DX benchmark. Results show that WFYOLO achieves 68.99% mAP@50 and 43.6% mAP@50-95, outperforming the YOLO11s baseline by 3.33% and 2.33%, and surpassing existing state-of-the-art models while maintaining 96.1 FPS real-time inference speed. WFYOLO has excellent clinical robustness and deployment potential for pediatric fracture auxiliary diagnosis.
Intelligent fault diagnosis (IFD) of rotating machinery is critical for ensuring industrial safety and reliability. However, existing deep learning-based IFD methods face three core challenges: suboptimal feature discrimination of single attention mechanisms, high computational cost limiting edge deployment, and class imbalance bias leading to misdiagnosis of rare faults. This study proposes an engineering-friendly IFD method integrating wavelet time-frequency analysis, convolutional block attention module (CBAM)-multihead self-attention (MHSA) hybrid attention, dimension-aligned lightweight distillation, and class-balanced weighted loss fusion. First, the continuous wavelet transform (CWT) converts raw vibration signals into time-frequency graphs for effective feature extraction. Second, the teacher model is designed by pretrained ResNet18, and the student model is designed by pretrained MobileNetV3-Small to extract feature map. Third, a CBAM-MHSA hybrid attention module is designed to complementarily capture local frequency enhancement and global temporal-spatial correlations, improving fault feature discrimination. Furthermore, a dimension-aligned teacher-student distillation framework is proposed to compress the model into a lightweight version, solving the high computational cost problem of traditional deep models. Finally, a class-balanced weighted loss fusion strategy is developed to mitigate class imbalance bias. Experimental results on two public datasets (SEU fault dataset and Case Western Reserve University [CWRU] bearing dataset) show that the proposed method achieves perfect classification (accuracy [Acc] = 1.0000, F1-macro = 1.0000) on both datasets, with 86.6% fewer parameters. It provides a practical solution for industrial IFD.
To reduce the impact of clutter on the target observation of surveillance radars, this article offers a new feature detection method that relies on modulation features throughout the radar azimuth scanning process to suppress clutter false alarms. In contrast to typical traditional time-domain detection methods, this work deviates from the standard method of measuring the signal-to-noise ratio for target identification by comparing the amplitudes of targets and the surrounding environment in the range direction. Rather, it makes use of the antenna azimuth direction, which is another dimension of the radar detecting space. This technique makes use of the obvious physical mechanism caused by the point-like distribution of targets and the planar distribution characteristics of clutter by building a feature vector based on the amplitude fluctuations of targets and clutter modulated by the antenna pattern during the radar scanning process. Based on this, this study suggests a joint detection architecture that combines traditional range detection with azimuth feature detection from the standpoint of engineering applications. Clutter false alarm suppression can be accomplished by incorporating an azimuth feature detection module into the conventional detection framework. The mechanism of the feature detection technique is validated in this study using simulation data, and its efficacy in decreasing clutter false alarms in radar target recognition is further confirmed by testing using real measured data from ground surveillance radar.
Costas arrays, known for their ideal autocorrelation properties, play a vital role in radar, sonar, and communication systems. Traditionally, they are obtained through exhaustive search or algebraic construction. This paper investigates the correlation properties of permutation arrays and examines the structural relationships and cross-correlation between an original Costas array and those derived from flipping and rotation. Through rigorous mathematical analysis, we establish a series of theorems that precisely characterize these relationships. This framework enhances the understanding of Costas array construction and establishes a solid theoretical foundation for the research and practical applications of Costas arrays.
This paper presents a novel method for designing low-complexity, multiplier-free filters tailored to multistage decimation architectures. The proposed decimation filters are constructed from building blocks Pm,r(z), which are self-reciprocal polynomials of degree 2m + 1, where m is an element of Z+ and r are design parameters. When r is selected within the admissible ranges identified in this work, these polynomials exhibit all their zeros on the unit circle (i.e., unimodular roots), making them well suited for multiplier-free, intrinsically stable, and highly selective decimation filtering. After establishing key properties of the proposed building blocks that enable computationally efficient implementations, a mixed-integer linear programming (MILP) optimization framework is introduced to determine a minimal-complexity set of such blocks whose cascade satisfies prescribed filter specifications. Practical guidelines for implementing the resulting multistage decimation filters are provided. Finally, the proposed approach is evaluated through comparisons with existing techniques in the literature.
On-site monitoring of iron oxide (FeO) content is crucial for ensuring product quality in the steel sintering process. However, the harsh sintering environment, characterized by nonperiodicity, high dust levels, heavy smoke, and high-temperature steam, severely degrades the quality of images captured at the sintering machine tail, thereby impacting the accuracy of FeO content prediction. Existing methods for keyframe extraction and image enhancement often rely on benchmark clear images or fail to address the unique challenges of the sintering environment, such as nonperiodic operational uncertainties and low-light conditions. To overcome these limitations, this paper proposes a novel temporal statistics and heuristic priors-based perceptual enhancement (THPE) framework. The proposed framework integrates an innovative dual-analysis strategy for keyframe extraction and perceptual enhancement. Gaussian statistical modeling and time-series analysis are employed to accurately extract key information from videos, enabling robust handling of the nonperiodic and uncertain nature of the sintering process and ensuring reliable information capture. Furthermore, for some harsh sintering environments, the heuristic priors based perceptual enhancement approach is introduced, which incorporates the frequency-domain adjustment model (FDAM), atmospheric scattering model (ASM), and contrast-based model (CBM). This design enables effective low-light noise modeling and image enhancement without reliance on benchmark images. Experimental results demonstrate that the proposed framework can effectively improve the accuracy of keyframe extraction, enhance perceptual quality, and improve the prediction accuracy of FeO in sintering.
To address the issue that the traditional direct position determination (DPD) method for coprime arrays (CAs) has limited performance in scenarios with low signal-to-noise ratio (SNR) and small quantity of observation data, a progressive optimization method based on the averaging of redundant data from virtual array elements is proposed. This method is built on the coprime difference co-array positioning model. It first employs the redundant data averaging technique of virtual array elements for noise reduction to enhance signal robustness, then selectively introduces virtual array hole interpolation (VAHI) technique according to actual scenario requirements to expand the effective aperture, and combines with Toeplitz matrix reconstruction technique to restore the full-rank property of the covariance matrix, thereby providing a data foundation for high-precision positioning. Finally, it utilizes a subspace data fusion algorithm to achieve accurate estimation of the radiation source position. Theoretical analysis and numerical simulations demonstrate that the proposed method improves the equivalent SNR through the utilization of redundant information, and its positioning performance is superior to that of existing methods. Particularly in cases of low SNR and limited observation resources, the advantages are significant, with the localization error reduced by more than 20%. This provides a solution for passive positioning in complex electromagnetic environments.
The burst-like and high-amplitude characteristics of impulsive noise, which markedly differ from those of Gaussian noise, render methods based on the Gaussian assumption unable to accurately characterize signals under impulsive noise. Moreover, when dealing with multicomponent signal, existing impulsive noise suppression methods inevitably introduce cross-term interference. To address these issues, this paper proposes an impulsive noise suppression method based on the torque clustering (TC) algorithm, and thus establishes an accurate representation of multicomponent linear frequency modulation (LFM) signal under impulsive noise. First, the theoretical analysis is conducted to elucidate the inherent limitations of existing noise suppression methods that inevitably cross-term introduction. A novel impulsive noise suppression technique based on TC is developed, fundamentally eliminating cross-term interference. Subsequently, two signal representation methods for multicomponent LFM signal under impulsive noise are proposed, namely, TC-fractional Fourier transform (TC-FRFT) and TC-synchrosqueezing transform (TC-SST). These methods enable accurate characterization of multicomponent LFM signal under impulsive noise, and facilitate precise extraction of signal features. Finally, a mathematical model for parameter estimation of multicomponent LFM signal is established using TC-FRFT, enabling high-precision estimation of center frequency and chirp rate in the presence of impulsive noise. Simulation results show that the proposed method can effectively suppress impulsive noise, avoid cross-term interference in existing methods, and achieve accurate characterization of multicomponent LFM signal under impulsive noise. Furthermore, the proposed TC-FRFT outperforms existing parameter estimation methods in terms of stability, accuracy, and robustness against noise.
Accurate electroencephalography (EEG) signals classification is essential for diagnosing brain disorders such as epilepsy. Whereas deep learning models such as convolution neural networks (CNNs) and long short-term memory (LSTM) improved EEG classification performance over traditional methods, existing attention mechanisms such as Additive, Luong, and Multihead struggle to capture EEG's complex temporal dependencies. This study proposes scaled custom attention (SCA); a mechanism for temporal dependency modeling during EEG signals classification. Unlike traditional QKV-based similarity scoring attention mechanisms, which applies semantic cross-token weighting, SCA replaces these operations with a direct feature-weighting strategy tailored to the temporal structure of EEG signals and incorporates a scaling mechanism to improve computational stability. To validate our approach, experiments were conducted using TUH EEG Epilepsy Corpus (TUEP) where SCA achieved an improved classification performance (accuracy: 98.07%, F1-Score: 98.06%), marginally higher than additive (97.60%, 97.61%), multihead (97.66%, 97.66%), and Luong (97.68%, 97.66%) baseline attention mechanisms when integrated to the LConvNet EEG classification model. Additionally, SCA achieves a balanced performance profile, with competitive inference time of 2.83 vs. 1.32-3.89 for baselines, parameter efficiency (58.5 params/sample vs., 58.5-63.7), and a comparable generalization, with an average training-validation difference (Avg) of 0.0191, making it a promising enhancement for EEG-based DL models. Subsequently, based on further performance comparison analyses using state-of-the-art (SOTA) EEG classification models; including EEGNet, DeepConvNet, and ShallowConvNet, the proposed LConvNet + SCA model demonstrates superior performance.
This paper proposes a Kalman-gain-driven neural Kalman filtering (KF) defense framework, termed KFDBP, for secure state estimation in cyber-physical systems (CPSs) under denial-of-service (DoS), spoofing, and replay attacks. Unlike end-to-end neural filtering approaches such as KalmanNet that directly learn state estimators or implicitly approximate the Kalman gain using deep recurrent architectures, the proposed method employs a lightweight back-propagation (BP) neural network to adaptively regulate the Kalman gain online, while strictly preserving the classical Kalman filter prediction-correction recursion. By formulating an innovation-oriented Kalman gain learning objective, KFDBP explicitly addresses attack-induced observation uncertainty and non-Gaussian measurement corruption without requiring prior knowledge of attack timing, attack type, or attack probability during online estimation. The bounded gain regulation mechanism enhances estimation stability and interpretability, which are critical for safety-sensitive CPS applications, while significantly reducing computational complexity compared with deep neural network-based filters. Extensive Monte Carlo simulations under single and hybrid attack scenarios demonstrate that KFDBP consistently achieves lower estimation error and improved robustness than the conventional Kalman filter and KalmanNet under different attack probabilities, making it suitable for real-time and resource-constrained CPS applications.
A robust parallel covariance intersection (PCI) fusion predictor for autoregressive (AR) systems facing mixed uncertainties is introduced. These mixed uncertainties comprise uncertain noise variances and missing measurements, which are prevalent and pose significant challenges in practical applications. The design methodology of the PCI fusion predictor is composed of three key steps. First, there is model conversion. By leveraging the state-space signal predictor and the fictitious approach, the original AR system is transformed into a multimodel system. Next comes the design of both local and PCI fusion predictors. Based on the minimax robust estimation principle and the PCI fusion algorithm, these predictors are developed. Finally, the robustness and robust accuracies of these predictors are verified. The matrix conversion method is employed to prove their robustness. To validate the proposed results, a simulation example is carried out. The simulation results clearly demonstrate the correctness and effectiveness of the developed predictors.
The implementation of Kalman filter (KF) in tracking high-dimensional, strongly correlated graph structured data is often complex and unstable. Meanwhile, in practical applications, the system may be subject to interference from non-Gaussian noise and various cyberattacks. First, the Student’s t-kernel-based graph signals maximum correntropy unscented KF (ST-GMCUKF) is proposed for hybrid attacks and non-Gaussian noise. The considered hybrid cyberattacks include denial of service (DoS) attacks and deception attacks. Then, the method integrates graph Fourier transform (GFT) to diagonalize Kalman gains for vertex-wise independent updates, reducing cumulative errors. Furthermore, by replacing Gaussian kernel with Student’s t-kernel in the maximum correntropy criterion (MCC), it enhances robustness against heavy-tailed noise and outliers. Finally, the estimation performance of the ST-GMCUKF algorithm is evaluated using a univariate nonstationary growth model. Simulation results demonstrate its superior capability in tracking high-dimensional graph signals systems under non-Gaussian noise and hybrid cyberattacks.
In the field of signal processing, modulation signals, including phase shift keying (PSK) and quadrature amplitude modulation (QAM), can significantly enhance the signal-to-noise ratio (SNR) through aliasing transmission following clustering and sorting. This article presents two novel approaches to compressed time difference of arrival (TDOA) estimation, leveraging amplitude-phase clustering signals. A carefully designed compression matrix is constructed based on the unique amplitude and phase characteristics of the signals. The study then analyzes the Cramer-Rao lower bound (CRLB) under full-sampling conditions. Finally, TDOA estimation is performed using the approximate maximum likelihood (AML) method. Simulation results demonstrate that the proposed compressed sampling TDOA estimation methods, based on amplitude-phase clustering, achieve accuracy within an order of magnitude of full-sampling performance. Additionally, this article explores the application of OFDM-QAM signals, which exhibit amplitude-phase convergence in the frequency domain, for time difference estimation in compressed sampling. A novel frequency-domain aliasing time difference estimation algorithm based on amplitude-phase convergence is proposed. Experimental results indicate that under high SNR conditions, the algorithm incurs only a minor SNR degradation of similar to 4 dB compared to time difference estimation in uncompressed transmission.
Hand gesture recognition using mmWave radar has emerged as a promising technology for human-computer interaction (HCI), smart home systems, and the Internet of Things (IoT). However, the practical application of this technology is often constrained by the high computational complexity and significant storage demands of contemporary deep neural networks, which impede their deployment on resource-limited embedded devices. To address this limitation, we present a novel approach that combines an improved MobileViT model with a knowledge distillation (KD) framework. The proposed method consists of three main stages. First, raw radar signals are captured and restructured into a three-dimensional format (Chirps x Samples x Frames, a 3D tensor) and processed to generate range-time maps (RTMs) and Doppler-time maps (DTMs). Second, an improved MobileViT network is designed, incorporating fewer redundant blocks, a lower input resolution, and a dual-branch input structure to effectively fuse features from the RTM and DTM. This enhanced architecture serves as a robust teacher model, excelling at extracting both local and global spatiotemporal features for accurate gesture recognition. Finally, KD is applied to transfer knowledge from the teacher model to a compact student network, thereby achieving model compression. Experimental results demonstrate that the final distilled student model, evaluated on the test set, has only 0.018 M parameters-similar to 10% of the teacher model's size-while still achieving a high recognition accuracy of 99.16%. Consequently, the resulting model is highly compact and accurate, demonstrating its suitability for real-world embedded deployment.
Brain-computer interface (BCI) plays an important role in various fields, such as neuroscience, rehabilitation, and machine learning. The silent BCI, which can reconstruct inner speech from neural activity, holds great promise for aphasia patients. In this paper, we design an imagined Chinese speech experimental paradigm based on initials and finals and collect raw signals from eight healthy participants by using 64-channel scalp electroencephalograms. Linear predictive coding (LPC) and mel frequency cepstral coefficients (MFCC), which are classical algorithms in the field of speech recognition, are used to extract distinguishing features for speech classification and reconstruction. Besides, the phase-lock value (PLV) is introduced to enrich the feature information. We choose support vector machine (SVM), linear discriminant analysis (LDA), decision tree (DT), and LogitBoost (LB) for binary classification in several different cases. Two-channel selection (CS) based on Broca’s area and Wernicke’s area of the brain is also introduced in the paper. The highest imaginary speech decoding accuracy reaches 84.38%, which demonstrates the effectiveness of the feature engineering. In addition, the comparative analysis is conducted with deep learning methods specifically designed for small sample scenarios. This study offers a novel systematic approach for the research of Chinese speech imagination BCI.
