Signal Processing最新文献

A high-resolution learning imaging method for THz-SAR moving targets based on AF-RPCA-Net 基于AF-RPCA-Net的THz-SAR运动目标高分辨率学习成像方法

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-13 DOI: 10.1016/j.sigpro.2026.110499

Xiaoyu Qin, Bin Deng, Hongqiang Wang

Terahertz-Synthetic Aperture Radar (THz-SAR) offers high frame rates and high resolution, making it particularly suitable for remote sensing applications, like dynamic monitoring of moving targets. However, due to the non-ideal motion of the airborne platform and the non-cooperative motion of targets, this phenomenon causes more severe defocusing compared with microwave band SAR. Traditional SAR imaging methods, if directly applied to image THz-SAR moving targets, often suffer from poor quality and low efficiency. To address this issue, this article proposes a moving target non-parametric learning imaging method based on the Deep Unfolding Network (DUN) framework. Firstly, an autofocusing module is derived based on the maximum imaging contrast and embedded within the Alternating Direction Method of Multipliers (ADMM) iterative solution process to achieve accurate compensation of azimuthal motion errors. Then, we introduce the concept of Robust Principal Component Analysis (RPCA) to achieve sparse recovery imaging of moving targets. Finally, based on the ADMM iterative solution process, we establish an imaging network, named AF-RPCA-Net, efficiently achieving model-data jointly driven moving target background separation and imaging. The proposed method is validated to be effective and efficient through experimental results derived from both simulated and measured data.

太赫兹合成孔径雷达（THz-SAR）提供高帧率和高分辨率，使其特别适用于遥感应用，如动态监测移动目标。然而，由于机载平台的非理想运动和目标的非协同运动，与微波波段SAR相比，这种现象造成了更严重的离焦。传统的SAR成像方法如果直接应用于太赫兹SAR运动目标成像，往往存在质量差、效率低的问题。为了解决这一问题，本文提出了一种基于深度展开网络（DUN）框架的运动目标非参数学习成像方法。首先，建立了基于最大成像对比度的自动对焦模块，并将其嵌入到交替方向乘法器（ADMM）迭代求解过程中，实现了方位运动误差的精确补偿；然后，引入鲁棒主成分分析（RPCA）的概念，实现运动目标稀疏恢复成像。最后，基于ADMM迭代求解过程，建立了AF-RPCA-Net成像网络，有效实现模型-数据联合驱动的运动目标背景分离与成像。仿真和实测数据的实验结果验证了该方法的有效性和有效性。

{"title":"A high-resolution learning imaging method for THz-SAR moving targets based on AF-RPCA-Net","authors":"Xiaoyu Qin, Bin Deng, Hongqiang Wang","doi":"10.1016/j.sigpro.2026.110499","DOIUrl":"10.1016/j.sigpro.2026.110499","url":null,"abstract":"<div><div>Terahertz-Synthetic Aperture Radar (THz-SAR) offers high frame rates and high resolution, making it particularly suitable for remote sensing applications, like dynamic monitoring of moving targets. However, due to the non-ideal motion of the airborne platform and the non-cooperative motion of targets, this phenomenon causes more severe defocusing compared with microwave band SAR. Traditional SAR imaging methods, if directly applied to image THz-SAR moving targets, often suffer from poor quality and low efficiency. To address this issue, this article proposes a moving target non-parametric learning imaging method based on the Deep Unfolding Network (DUN) framework. Firstly, an autofocusing module is derived based on the maximum imaging contrast and embedded within the Alternating Direction Method of Multipliers (ADMM) iterative solution process to achieve accurate compensation of azimuthal motion errors. Then, we introduce the concept of Robust Principal Component Analysis (RPCA) to achieve sparse recovery imaging of moving targets. Finally, based on the ADMM iterative solution process, we establish an imaging network, named AF-RPCA-Net, efficiently achieving model-data jointly driven moving target background separation and imaging. The proposed method is validated to be effective and efficient through experimental results derived from both simulated and measured data.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"243 ","pages":"Article 110499"},"PeriodicalIF":3.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978305","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

DEANet : Adaptive RGB-T salient object detection with two-dimensional entropy-guided dual-domain feature interaction 基于二维熵导双域特征交互的自适应RGB-T显著目标检测

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-09 DOI: 10.1016/j.sigpro.2026.110489

Zerui Zhu , Dongmei Liu , Huaxiang Zhang , Li Liu , Fengfei Jin

RGB-T salient object detection (RGB-T SOD) aims to accurately localize salient objects by integrating complementary cues from RGB and thermal images, yet existing methods often overlook critical frequency-domain information. Our frequency-domain analysis reveals modality inconsistencies in salient regions, highlighting the need for adaptive modality evaluation. To address this issue, we propose a two-dimensional information entropy-based weighting strategy that quantifies structural complexity and adaptively guides modality contribution. Building upon this strategy, we develop the Dual-Domain Entropy-Aware Network (DEANet), which incorporates a Progressive Dual-domain Fusion and Refinement (PDFR) design-a coherent two-stage progressive mechanism. Stage 1 performs entropy-guided spatial-frequency interaction to generate high-quality fused features, while Stage 2 leverages these fused features to enhance original modality representations and refine saliency through spatial-channel perception. This progressive dual-domain formulation enables robust multimodal fusion and more accurate saliency estimation under diverse imaging conditions. Extensive experiments on three public benchmarks demonstrate that DEANet consistently surpasses 17 state-of-the-art methods across multiple evaluation metrics.

RGB- t显著目标检测（RGB- t SOD）旨在通过整合来自RGB和热图像的互补线索来准确定位显著目标，但现有方法往往忽略了关键的频域信息。我们的频域分析揭示了显著区域的模态不一致，强调了自适应模态评估的必要性。为了解决这一问题，我们提出了一种基于二维信息熵的加权策略，该策略量化了结构复杂性并自适应地指导了模态的贡献。在此策略的基础上，我们开发了双域熵感知网络（DEANet），它结合了渐进式双域融合和细化（PDFR）设计-一种连贯的两阶段渐进机制。阶段1执行熵引导的空间频率交互以生成高质量的融合特征，而阶段2利用这些融合特征增强原始模态表征并通过空间通道感知改善显著性。这种渐进式双域公式能够实现鲁棒的多模态融合和在不同成像条件下更准确的显著性估计。在三个公共基准上进行的广泛实验表明，DEANet在多个评估指标上始终超过17种最先进的方法。

{"title":"DEANet : Adaptive RGB-T salient object detection with two-dimensional entropy-guided dual-domain feature interaction","authors":"Zerui Zhu , Dongmei Liu , Huaxiang Zhang , Li Liu , Fengfei Jin","doi":"10.1016/j.sigpro.2026.110489","DOIUrl":"10.1016/j.sigpro.2026.110489","url":null,"abstract":"<div><div>RGB-T salient object detection (RGB-T SOD) aims to accurately localize salient objects by integrating complementary cues from RGB and thermal images, yet existing methods often overlook critical frequency-domain information. Our frequency-domain analysis reveals modality inconsistencies in salient regions, highlighting the need for adaptive modality evaluation. To address this issue, we propose a two-dimensional information entropy-based weighting strategy that quantifies structural complexity and adaptively guides modality contribution. Building upon this strategy, we develop the Dual-Domain Entropy-Aware Network (DEANet), which incorporates a Progressive Dual-domain Fusion and Refinement (PDFR) design-a coherent two-stage progressive mechanism. Stage 1 performs entropy-guided spatial-frequency interaction to generate high-quality fused features, while Stage 2 leverages these fused features to enhance original modality representations and refine saliency through spatial-channel perception. This progressive dual-domain formulation enables robust multimodal fusion and more accurate saliency estimation under diverse imaging conditions. Extensive experiments on three public benchmarks demonstrate that DEANet consistently surpasses 17 state-of-the-art methods across multiple evaluation metrics.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"243 ","pages":"Article 110489"},"PeriodicalIF":3.6,"publicationDate":"2026-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978346","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

2-D DOA and polarization estimation using cylindrical coprime conformal array via cross-covariance tensor reconstruction 基于交叉协方差张量重构的柱质共形阵二维DOA和偏振估计

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-08 DOI: 10.1016/j.sigpro.2025.110485

Mingcheng Fu , Zhi Zheng , Ping Li , Wen-Qin Wang

In this article, we develop an efficient approach for two-dimensional (2-D) direction-of-arrival (DOA) and polarization estimation using the cylindrical coprime conformal array. Firstly, we derive the tensor-form coarray output of the cylindrical coprime conformal array and apply virtual array interpolation on the coarray output components. Subsequently, we construct a fourth-order cross-covariance tensor using the interpolated array outputs and recover a low-rank fourth-order augmented tensor by formulating a nuclear norm minimization problem. Using the reconstructed augmented tensor, we estimate the elevation and azimuth angles of sources separately through one-dimensional searching. With the estimated 2-D DOAs, we finally derive the closed-form expressions for the polarization parameter estimates. Compared with the previous techniques, the proposed algorithm can identify more sources and provide offer higher parameter estimation accuracy. Simulation results demonstrate the advantage of our algorithm over several existing techniques.

在本文中，我们开发了一种有效的二维（2-D）到达方向（DOA）和偏振估计的方法，使用圆柱质共形阵列。首先，我们推导出了柱素数共形阵列的张量形式的共阵输出，并对共阵输出分量进行了虚拟阵列插值。随后，我们利用插值数组输出构造一个四阶交叉协方差张量，并通过制定核范数最小化问题恢复一个低秩四阶增广张量。利用重构的增广张量，通过一维搜索分别估计光源的仰角和方位角。利用估计的二维DOAs，我们最后导出了偏振参数估计的封闭表达式。与以往的方法相比，该算法可以识别更多的信号源，并提供更高的参数估计精度。仿真结果表明，该算法优于现有的几种算法。

引用次数: 0

Robust learning under label noise via logit-based filtering and ranking-aware relabeling 标签噪声下的鲁棒学习，通过基于逻辑的过滤和等级感知的重标注

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-08 DOI: 10.1016/j.sigpro.2026.110498

Fanglong Wu , Min Yang , Peng Cheng , Zhisheng You

Label noise poses a significant challenge in supervised learning tasks such as image classification and face recognition, often steering models away from their optimal learning trajectory. To reduce the adverse impact of noisy annotations while effectively leveraging available training data, we propose a robust learning framework that exploits logit space distributions for noise identification, ranking-guided relabeling of closed-set noise, and noise-aware optimization. The key insight behind our approach is that clean non-target samples and noisy target-class samples that have not yet been memorized by the network tend to exhibit similar logit distribution patterns. Based on this observation, we design adaptive, class-specific decision boundaries for blind noise detection. For closed-set noise, we compute the margin between the top two logits from non-target classes as a confidence score and incorporate historical ranking statistics. A pseudo-label is assigned when either the logit margin or the historical average rank of the top-1 class satisfies predefined criteria. Finally, clean and relabeled samples are trained with different regularization strengths to improve robustness. Extensive experiments on three synthetic and four real-world noisy datasets, covering image classification and face recognition tasks, demonstrate the effectiveness and generality of the proposed method.

标签噪声对图像分类和人脸识别等监督学习任务构成了重大挑战，通常会使模型偏离最佳学习轨迹。为了减少噪声注释的不利影响，同时有效地利用可用的训练数据，我们提出了一个鲁棒的学习框架，该框架利用logit空间分布进行噪声识别、封闭集噪声的排序引导重新标记和噪声感知优化。我们的方法背后的关键见解是，干净的非目标样本和尚未被网络记忆的噪声目标类样本倾向于表现出相似的logit分布模式。基于这一观察，我们设计了自适应的、类特定的决策边界用于盲噪声检测。对于闭集噪声，我们计算非目标类的前两个logits之间的余量作为置信度分数，并结合历史排名统计。当前1类的logit裕度或历史平均排名满足预定义的标准时，分配伪标签。最后，用不同的正则化强度训练干净和重新标记的样本，以提高鲁棒性。在包含图像分类和人脸识别任务的3个合成数据集和4个真实噪声数据集上进行了大量实验，证明了该方法的有效性和通用性。

{"title":"Robust learning under label noise via logit-based filtering and ranking-aware relabeling","authors":"Fanglong Wu , Min Yang , Peng Cheng , Zhisheng You","doi":"10.1016/j.sigpro.2026.110498","DOIUrl":"10.1016/j.sigpro.2026.110498","url":null,"abstract":"<div><div>Label noise poses a significant challenge in supervised learning tasks such as image classification and face recognition, often steering models away from their optimal learning trajectory. To reduce the adverse impact of noisy annotations while effectively leveraging available training data, we propose a robust learning framework that exploits logit space distributions for noise identification, ranking-guided relabeling of closed-set noise, and noise-aware optimization. The key insight behind our approach is that clean non-target samples and noisy target-class samples that have not yet been memorized by the network tend to exhibit similar logit distribution patterns. Based on this observation, we design adaptive, class-specific decision boundaries for blind noise detection. For closed-set noise, we compute the margin between the top two logits from non-target classes as a confidence score and incorporate historical ranking statistics. A pseudo-label is assigned when either the logit margin or the historical average rank of the top-1 class satisfies predefined criteria. Finally, clean and relabeled samples are trained with different regularization strengths to improve robustness. Extensive experiments on three synthetic and four real-world noisy datasets, covering image classification and face recognition tasks, demonstrate the effectiveness and generality of the proposed method.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"243 ","pages":"Article 110498"},"PeriodicalIF":3.6,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978517","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Adaptive regularization parameter adjustment for total variation denoising 全变差去噪的自适应正则化参数调整

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-08 DOI: 10.1016/j.sigpro.2026.110494

Donghao Lv , Tianshun Li , Peihong Yang , Chao Zhang , Jianjun Li

Total variation denoising has been extensively used in the restoration of piecewise constant signals, which are highly valued in numerous practical applications. However, existing approaches often struggle with the choice of regularization parameter, potentially leading to suboptimal denoising performance. To address this issue, this paper presents an adaptive regularization parameter adjustment mechanism and incorporates it with total variation denoising algorithm. An optimization strategy based on the solution of differential equation is designed to determine the regularization parameter, enabling it to converge toward an optimal value automatically. This strategy is then integrated into the total variation denoising framework to dynamically adjust the regularization parameter during the denoising process. Simulations and experimental results confirm that the proposed method significantly enhances the denoising efficiency for piecewise constant signals.

全变差去噪在分段常数信号的复原中得到了广泛的应用，在许多实际应用中得到了高度重视。然而，现有的方法经常在正则化参数的选择上遇到困难，这可能导致去噪性能不理想。针对这一问题，本文提出了一种自适应正则化参数调整机制，并将其与全变分去噪算法相结合。设计了一种基于微分方程解的优化策略来确定正则化参数，使正则化参数自动收敛到最优值。然后将该策略集成到全变分去噪框架中，在去噪过程中动态调整正则化参数。仿真和实验结果表明，该方法显著提高了对分段常数信号的去噪效率。

引用次数: 0

Optimal design of stable allpass variable fractional delay filters using matrix-based algorithms 基于矩阵算法的稳定全通可变分数阶延迟滤波器优化设计

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-08 DOI: 10.1016/j.sigpro.2026.110496

Ruijie Zhao , Chunlu Lai

The optimal designs of allpass variable fractional delay (VFD) filters based on phase response approximation are investigated. The weighted least squares (WLS) design that allows for arbitrary nonnegative weighting functions is formulated in matrix form, and the optimality condition is then derived as a matrix equation. Two efficient algorithms that are derived from the conjugate gradient (CG) technique are proposed to solve the WLS problem. Subsequently, an iterative reweighted least squares (IRLS) algorithm is developed for the minimax design problem, which converts the original problem into a series of WLS subproblems and solves them successively using the proposed WLS algorithms. A transformation method using Chebyshev polynomials is presented to circumvent numerical problems in calculation. The filter coefficients are arranged as matrices, achieving significant computation and memory space savings. The associated computational complexity is evaluated. Moreover, by introducing a delay shift parameter in the desired response, design accuracy can be improved significantly. The stability of allpass VFD filters is analyzed, and stability conditions based on the delay shift parameter and phase error are established. Comparisons with existing methods are provided to show the efficiency and effectiveness of the proposed algorithms.

研究了基于相位响应近似的全通可变分数延迟（VFD）滤波器的优化设计。将允许任意非负权函数的加权最小二乘（WLS）设计以矩阵形式表述，并推导出最优性条件为矩阵方程。从共轭梯度（CG）技术出发，提出了两种求解WLS问题的有效算法。随后，针对极大极小设计问题，提出了一种迭代重加权最小二乘（IRLS）算法，该算法将原问题转化为一系列WLS子问题，并利用所提出的WLS算法依次求解。提出了一种利用切比雪夫多项式的变换方法，避免了计算中的数值问题。滤波器系数以矩阵形式排列，大大节省了计算和存储空间。计算相关的计算复杂度。此外，通过在期望响应中引入延迟移位参数，可以显著提高设计精度。分析了全通VFD滤波器的稳定性，建立了基于延时偏移参数和相位误差的稳定条件。通过与现有方法的比较，证明了所提算法的效率和有效性。

{"title":"Optimal design of stable allpass variable fractional delay filters using matrix-based algorithms","authors":"Ruijie Zhao , Chunlu Lai","doi":"10.1016/j.sigpro.2026.110496","DOIUrl":"10.1016/j.sigpro.2026.110496","url":null,"abstract":"<div><div>The optimal designs of allpass variable fractional delay (VFD) filters based on phase response approximation are investigated. The weighted least squares (WLS) design that allows for arbitrary nonnegative weighting functions is formulated in matrix form, and the optimality condition is then derived as a matrix equation. Two efficient algorithms that are derived from the conjugate gradient (CG) technique are proposed to solve the WLS problem. Subsequently, an iterative reweighted least squares (IRLS) algorithm is developed for the minimax design problem, which converts the original problem into a series of WLS subproblems and solves them successively using the proposed WLS algorithms. A transformation method using Chebyshev polynomials is presented to circumvent numerical problems in calculation. The filter coefficients are arranged as matrices, achieving significant computation and memory space savings. The associated computational complexity is evaluated. Moreover, by introducing a delay shift parameter in the desired response, design accuracy can be improved significantly. The stability of allpass VFD filters is analyzed, and stability conditions based on the delay shift parameter and phase error are established. Comparisons with existing methods are provided to show the efficiency and effectiveness of the proposed algorithms.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"243 ","pages":"Article 110496"},"PeriodicalIF":3.6,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978343","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Robust angle estimation in MIMO radar under impulsive noise via fast bayesian tensor decomposition with intra-dimension correlation 基于维内相关的快速贝叶斯张量分解的脉冲噪声下MIMO雷达鲁棒角度估计

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-08 DOI: 10.1016/j.sigpro.2026.110492

Jinli Chen , Yang Song , Hua Shao , Jiaqiang Li

Conventional angle estimation methods are highly sensitive to outliers, causing severe performance degradation under impulsive noise. Although existing tensor-based Bayesian approaches can alleviate the impact of outliers, strongly impulsive noise in multiple-input multiple-output (MIMO) radar often leads to outlier model mismatch, reducing robustness against outliers. To address this, we propose a fast Bayesian method for angle estimation under impulsive noise, which exploits the tensor intra-dimension correlations and incorporates the Vandermonde structure of factor matrices within a Bayesian tensor decomposition framework. Strong outliers in the array measurements are first removed via thresholding to mitigate model mismatch. A hierarchical probabilistic model based on canonical polyadic (CP) decomposition is then developed to capture the correlation structure and the Vandermonde structural prior. Model parameters are efficiently inferred via an expectation–maximization (EM) algorithm, which recovers missing entries caused by thresholding and suppresses residual outliers. Furthermore, a complexity-reduction method is developed to accelerate computation by employing a snapshot-wise stackable strategy and leveraging the sparsity of thresholded entries, enabling efficient estimation of factor matrices across multiple snapshots. Finally, DOAs and DODs are jointly estimated from the decomposed factor matrices. Simulations verify the outlier-robust performance of the proposed method in providing high-accuracy angle estimation under impulsive noise.

传统的角度估计方法对异常值非常敏感，在脉冲噪声下会导致性能严重下降。虽然现有的基于张量的贝叶斯方法可以减轻异常值的影响，但多输入多输出（MIMO）雷达中强烈的脉冲噪声经常导致异常值模型失配，降低了对异常值的鲁棒性。为了解决这个问题，我们提出了一种快速的贝叶斯方法用于脉冲噪声下的角度估计，该方法利用了张量的维内相关性，并在贝叶斯张量分解框架内结合了因子矩阵的Vandermonde结构。阵列测量中的强异常值首先通过阈值去除，以减轻模型不匹配。然后建立了基于正则多进（CP）分解的分层概率模型来捕获相关结构和Vandermonde结构先验。通过期望最大化（EM）算法有效地推断模型参数，该算法恢复阈值导致的缺失条目并抑制残差异常值。此外，开发了一种复杂性降低方法，通过采用快照可堆叠策略和利用阈值条目的稀疏性来加速计算，从而实现跨多个快照对因子矩阵的有效估计。最后，从分解的因子矩阵中联合估计doa和DODs。仿真结果验证了该方法在脉冲噪声下提供高精度角度估计的异常鲁棒性。

{"title":"Robust angle estimation in MIMO radar under impulsive noise via fast bayesian tensor decomposition with intra-dimension correlation","authors":"Jinli Chen , Yang Song , Hua Shao , Jiaqiang Li","doi":"10.1016/j.sigpro.2026.110492","DOIUrl":"10.1016/j.sigpro.2026.110492","url":null,"abstract":"<div><div>Conventional angle estimation methods are highly sensitive to outliers, causing severe performance degradation under impulsive noise. Although existing tensor-based Bayesian approaches can alleviate the impact of outliers, strongly impulsive noise in multiple-input multiple-output (MIMO) radar often leads to outlier model mismatch, reducing robustness against outliers. To address this, we propose a fast Bayesian method for angle estimation under impulsive noise, which exploits the tensor intra-dimension correlations and incorporates the Vandermonde structure of factor matrices within a Bayesian tensor decomposition framework. Strong outliers in the array measurements are first removed via thresholding to mitigate model mismatch. A hierarchical probabilistic model based on canonical polyadic (CP) decomposition is then developed to capture the correlation structure and the Vandermonde structural prior. Model parameters are efficiently inferred via an expectation–maximization (EM) algorithm, which recovers missing entries caused by thresholding and suppresses residual outliers. Furthermore, a complexity-reduction method is developed to accelerate computation by employing a snapshot-wise stackable strategy and leveraging the sparsity of thresholded entries, enabling efficient estimation of factor matrices across multiple snapshots. Finally, DOAs and DODs are jointly estimated from the decomposed factor matrices. Simulations verify the outlier-robust performance of the proposed method in providing high-accuracy angle estimation under impulsive noise.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"243 ","pages":"Article 110492"},"PeriodicalIF":3.6,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978345","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

From structure to detail: A conditional diffusion framework for extremely low-bitrate image compression 从结构到细节：用于极低比特率图像压缩的条件扩散框架

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-08 DOI: 10.1016/j.sigpro.2025.110480

Junhui Li, Yiyang Zou, Xingsong Hou, Yutao Zhang, Zhixuan Guo

Although existing diffusion-based methods produce visually rich textures at extremely low bitrates, they often sacrifice structural fidelity, resulting in significant deviations from the original image. To address this fundamental trade-off, we propose Fidelity-Perception Diffusion-based Image Compression (FPD-IC), a two-stage conditional diffusion framework that explicitly separates structure reconstruction and detail restoration. In Stage I, we use a VAE-based compressor to recover structurally faithful conditional images from highly compact bitstreams. In Stage II, a diffusion model, guided by the output from Stage I, generates visually rich details. This conditional approach allows the diffusion model to focus exclusively on perceptual enhancement while preserving the overall structure established in Stage I. Additionally, we introduce a lightweight Fidelity-Perception Tuner Module (FPTM) to combine the outputs of both stages, enabling controllable trade-offs between fidelity and perceptual quality. Extensive experiments on the Kodak and Tecnick datasets demonstrate the effectiveness and robustness of FPD-IC. On the Tecnick dataset, FPD-IC outperforms state-of-the-art diffusion-based methods by 2.24–3.57 dB in PSNR at bitrates below 0.06 bpp, while also achieving superior perceptual quality. Furthermore, FPD-IC shows strong robustness to input noise, consistently maintaining high fidelity and perceptual quality under Gaussian perturbations. The code will be released at https://github.com/mlkk518/FPD-IC.

尽管现有的基于扩散的方法以极低的比特率产生视觉上丰富的纹理，但它们往往会牺牲结构保真度，导致与原始图像的显著偏差。为了解决这一基本问题，我们提出了基于保真度感知扩散的图像压缩（FPD-IC），这是一个明确分离结构重建和细节恢复的两阶段条件扩散框架。在第一阶段，我们使用基于vae的压缩器从高度紧凑的比特流中恢复结构忠实的条件图像。在第二阶段，扩散模型，由第一阶段的输出引导，产生视觉上丰富的细节。这种有条件的方法允许扩散模型专注于感知增强，同时保留阶段i中建立的整体结构。此外，我们引入了一个轻量级的保真度-感知调谐模块（FPTM）来组合两个阶段的输出，从而在保真度和感知质量之间实现可控的权衡。在Kodak和Tecnick数据集上的大量实验证明了FPD-IC的有效性和鲁棒性。在Tecnick数据集上，在比特率低于0.06 bpp的情况下，FPD-IC的PSNR比最先进的基于扩散的方法高出2.24-3.57 dB，同时也实现了卓越的感知质量。此外，FPD-IC对输入噪声具有很强的鲁棒性，在高斯扰动下始终保持高保真度和感知质量。代码将在https://github.com/mlkk518/FPD-IC上发布。

{"title":"From structure to detail: A conditional diffusion framework for extremely low-bitrate image compression","authors":"Junhui Li, Yiyang Zou, Xingsong Hou, Yutao Zhang, Zhixuan Guo","doi":"10.1016/j.sigpro.2025.110480","DOIUrl":"10.1016/j.sigpro.2025.110480","url":null,"abstract":"<div><div>Although existing diffusion-based methods produce visually rich textures at extremely low bitrates, they often sacrifice structural fidelity, resulting in significant deviations from the original image. To address this fundamental trade-off, we propose Fidelity-Perception Diffusion-based Image Compression (FPD-IC), a two-stage conditional diffusion framework that explicitly separates structure reconstruction and detail restoration. In Stage I, we use a VAE-based compressor to recover structurally faithful conditional images from highly compact bitstreams. In Stage II, a diffusion model, guided by the output from Stage I, generates visually rich details. This conditional approach allows the diffusion model to focus exclusively on perceptual enhancement while preserving the overall structure established in Stage I. Additionally, we introduce a lightweight Fidelity-Perception Tuner Module (FPTM) to combine the outputs of both stages, enabling controllable trade-offs between fidelity and perceptual quality. Extensive experiments on the Kodak and Tecnick datasets demonstrate the effectiveness and robustness of FPD-IC. On the Tecnick dataset, FPD-IC outperforms state-of-the-art diffusion-based methods by 2.24–3.57 dB in PSNR at bitrates below 0.06 bpp, while also achieving superior perceptual quality. Furthermore, FPD-IC shows strong robustness to input noise, consistently maintaining high fidelity and perceptual quality under Gaussian perturbations. The code will be released at <span><span>https://github.com/mlkk518/FPD-IC</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"243 ","pages":"Article 110480"},"PeriodicalIF":3.6,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978344","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

MWNet: Image dehazing network based on multi-scale feature extraction and wavelet feature enhancement MWNet：基于多尺度特征提取和小波特征增强的图像去雾网络

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-07 DOI: 10.1016/j.sigpro.2026.110493

Haixin Jia , Han Wang , Yu Zhang , Guoying Zhang , Zhengfan Li , Hengchen Xu

Atmospheric haze degrades image quality, impairing downstream vision tasks like object detection and segmentation. While wavelet-based deep learning methods are effective by leveraging lossless downsampling and spectral discrepancies, they often suffer from limited multi-scale feature extraction, inadequate frequency-domain enhancement, and a lack of structural priors. To overcome these issues, we propose MWNet, a novel framework integrating structural constraints into a U-Net with wavelet transforms. Our approach introduces dense multi-scale blocks for robust feature extraction, a hierarchical attention mechanism for high-frequency detail enhancement, and a cross-enhancement module for frequency feature interaction. Extensive experiments conducted on four benchmark datasets (SOTS-Indoor, Haze4K, Dense-Haze, NH-Haze) have demonstrated consistent superiority, with MWNet achieving SOTA in quantitative results compared to existing advanced methods (Surpassing the second-best method with average improvements of 0.16 dB in PSNR and 0.0026 in SSIM.), while qualitative results demonstrate enhanced detail preservation and noise suppression. In addition, we conducted generalization tests on three other datasets (RTTS, REAL-NH, CM-Haze), fully verifying the good generalization performance of MWNet.

大气雾霾会降低图像质量，损害下游视觉任务，如物体检测和分割。虽然基于小波的深度学习方法通过利用无损下采样和频谱差异是有效的，但它们往往受到多尺度特征提取有限、频域增强不足和缺乏结构先验的影响。为了克服这些问题，我们提出了MWNet，一种将结构约束与小波变换集成到U-Net中的新框架。我们的方法引入了密集的多尺度块用于鲁棒特征提取，分层关注机制用于高频细节增强，交叉增强模块用于频率特征交互。在四个基准数据集（SOTS-Indoor、Haze4K、Dense-Haze、NH-Haze）上进行的大量实验显示出了一致的优势，与现有的先进方法相比，MWNet在定量结果上达到了SOTA (PSNR平均提高0.16 dB， SSIM平均提高0.0026)，而定性结果显示细节保存和噪声抑制得到了增强。此外，我们还对另外三个数据集（RTTS、REAL-NH、CM-Haze）进行了泛化测试，充分验证了MWNet良好的泛化性能。

{"title":"MWNet: Image dehazing network based on multi-scale feature extraction and wavelet feature enhancement","authors":"Haixin Jia , Han Wang , Yu Zhang , Guoying Zhang , Zhengfan Li , Hengchen Xu","doi":"10.1016/j.sigpro.2026.110493","DOIUrl":"10.1016/j.sigpro.2026.110493","url":null,"abstract":"<div><div>Atmospheric haze degrades image quality, impairing downstream vision tasks like object detection and segmentation. While wavelet-based deep learning methods are effective by leveraging lossless downsampling and spectral discrepancies, they often suffer from limited multi-scale feature extraction, inadequate frequency-domain enhancement, and a lack of structural priors. To overcome these issues, we propose MWNet, a novel framework integrating structural constraints into a U-Net with wavelet transforms. Our approach introduces dense multi-scale blocks for robust feature extraction, a hierarchical attention mechanism for high-frequency detail enhancement, and a cross-enhancement module for frequency feature interaction. Extensive experiments conducted on four benchmark datasets (SOTS-Indoor, Haze4K, Dense-Haze, NH-Haze) have demonstrated consistent superiority, with MWNet achieving SOTA in quantitative results compared to existing advanced methods (Surpassing the second-best method with average improvements of 0.16 dB in PSNR and 0.0026 in SSIM.), while qualitative results demonstrate enhanced detail preservation and noise suppression. In addition, we conducted generalization tests on three other datasets (RTTS, REAL-NH, CM-Haze), fully verifying the good generalization performance of MWNet.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"243 ","pages":"Article 110493"},"PeriodicalIF":3.6,"publicationDate":"2026-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978340","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A weighted coherent integration method for weak target detection based on active-passive radar 一种基于主-被动雷达的弱目标检测加权相干积分方法

IF 3.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Signal Processing

Pub Date : 2026-01-07 DOI: 10.1016/j.sigpro.2026.110495

Boyang Jia , Jianwei Zhao , Yaxing Yue , Xuepan Zhang , Jiayi Zhao , Sining Liu , Guisheng Liao

We propose a Weighted Coherent Integration (WCI) algorithm for weak target detection. This method aims to achieve maximum signal-to-noise ratio (SNR) gain by coherently accumulating echoes from active and passive radars. Two key challenges about range-Doppler misalignment and inter-channel differences are addressed by a two-stage framework. First, spatial alignment is achieved by partitioning the surveillance area, and Doppler resolutions are unified via adaptive integration time. Target fluctuations and phase errors between channels are then mitigated via inter-channel coherent integration with phase compensation. Simulation results validate that WCI outperforms conventional monostatic and multistatic non-coherent fusion methods, achieving superior detection capability in both low-SNR and multi-target scenarios.

提出了一种用于弱目标检测的加权相干积分（WCI）算法。该方法旨在通过相干积累主动式和无源雷达回波，实现最大信噪比增益。两阶段框架解决了距离-多普勒失调和信道间差异的两个关键挑战。首先，通过划分监视区域实现空间对准，并通过自适应积分时间统一多普勒分辨率；然后通过带相位补偿的信道间相干积分来减轻信道间的目标波动和相位误差。仿真结果验证了WCI优于传统的单静态和多静态非相干融合方法，在低信噪比和多目标场景下都具有出色的检测能力。

引用次数: 0