Digital Signal Processing最新文献_第6页

An efficient coherent integration method for RFPA radar signals in high-speed and multi-target scenarios 一种高速多目标场景下RFPA雷达信号的高效相干积分方法

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-19 DOI: 10.1016/j.dsp.2025.105838

Yijiang Chen, Guohua Wei, Jiahao Bai, Xu Wang

Although random frequency and pulse repetition interval (PRI) agile (RFPA) radar signals provide strong electronic counter-countermeasures (ECCM) capabilities, coherent integration under high-speed and multi-target scenarios remains a critical but insufficiently explored challenge. We propose an efficient coherent integration method termed RFPA-KT-NUFFT, which combines Keystone transform (KT) and nonuniform fast Fourier transform (NUFFT). The truncated-sinc-interpolation-based KT exploits low computational complexity to correct range migration and decouple the interdependencies between frequency and PRI. Next, a tailored phase compensation function is developed to harness the frequency-agile bandwidth, thereby enabling high range resolution and facilitating multi-target separation. Finally, NUFFT efficiently handles the nonuniform slow-time sampling for coherent integration. Monte Carlo experiments with a 500-sequence agile waveform library demonstrate that the proposed method significantly reduces runtime while maintaining integration performance comparable to existing optimal methods, offering a viable solution that achieves a favorable trade-off between performance and efficiency in high-speed and multi-target scenarios.

尽管随机频率和脉冲重复间隔（PRI）敏捷（RFPA）雷达信号提供了强大的电子对抗（ECCM）能力，但高速和多目标场景下的相干集成仍然是一个关键但尚未充分探索的挑战。提出了一种结合了Keystone变换和非均匀快速傅里叶变换的高效相干积分方法RFPA-KT-NUFFT。基于截断自插补的KT利用较低的计算复杂度来校正距离迁移并解耦频率和PRI之间的相互依赖性。接下来，开发了定制的相位补偿函数来利用频率敏捷带宽，从而实现高距离分辨率并促进多目标分离。最后，NUFFT有效地处理了相干积分的非均匀慢时间采样。基于500序列敏捷波形库的蒙特卡罗实验表明，该方法显著减少了运行时间，同时保持了与现有最优方法相当的集成性能，提供了一种可行的解决方案，在高速和多目标场景下实现了性能和效率之间的良好权衡。

{"title":"An efficient coherent integration method for RFPA radar signals in high-speed and multi-target scenarios","authors":"Yijiang Chen, Guohua Wei, Jiahao Bai, Xu Wang","doi":"10.1016/j.dsp.2025.105838","DOIUrl":"10.1016/j.dsp.2025.105838","url":null,"abstract":"<div><div>Although random frequency and pulse repetition interval (PRI) agile (RFPA) radar signals provide strong electronic counter-countermeasures (ECCM) capabilities, coherent integration under high-speed and multi-target scenarios remains a critical but insufficiently explored challenge. We propose an efficient coherent integration method termed RFPA-KT-NUFFT, which combines Keystone transform (KT) and nonuniform fast Fourier transform (NUFFT). The truncated-sinc-interpolation-based KT exploits low computational complexity to correct range migration and decouple the interdependencies between frequency and PRI. Next, a tailored phase compensation function is developed to harness the frequency-agile bandwidth, thereby enabling high range resolution and facilitating multi-target separation. Finally, NUFFT efficiently handles the nonuniform slow-time sampling for coherent integration. Monte Carlo experiments with a 500-sequence agile waveform library demonstrate that the proposed method significantly reduces runtime while maintaining integration performance comparable to existing optimal methods, offering a viable solution that achieves a favorable trade-off between performance and efficiency in high-speed and multi-target scenarios.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105838"},"PeriodicalIF":3.0,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842762","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

UFDT-YOLO: Robust small object detection from UAV perspectives in foggy environments udt - yolo：雾天环境下无人机视角的鲁棒小目标检测

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-19 DOI: 10.1016/j.dsp.2025.105826

Ming Zhao, Jin Chen, Jianyu Zheng, Yanchao Li, Canhui Bao

Small object detection from UAV perspectives is crucial for autonomous environmental perception, especially under adverse weather such as fog. However, image degradation and the small size of targets often cause insufficient feature representation and background interference, reducing detection performance. Although recent studies have advanced image dehazing or small object detection, few address both challenges simultaneously. To tackle these issues, we propose UFDT-YOLO, a detection framework designed for small object detection in adverse weather. The model employs an end-to-end multi-task optimization strategy, jointly enhancing image restoration and detection. Based on YOLOv12, our network integrates an image restoration module to improve feature extraction robustness and introduces two novel modules for enriched representation. First, the Feature Attention Module strengthens weak features, improving sensitivity to small objects. Second, the C3 Module with Dynamic Transformer adaptively focuses on critical regions and captures essential contextual information, effectively suppressing fog-induced interference. Experiments on the synthetic UAV imagery dataset HazyDet and the real-world foggy UAV imagery dataset RDDTS validate the effectiveness of UFDT-YOLO for small object detection under adverse weather conditions.

从无人机角度进行小目标检测对于自主环境感知至关重要，特别是在恶劣天气（如雾）下。然而，图像的退化和目标的小尺寸往往导致特征表示不足和背景干扰，降低了检测性能。虽然最近的研究有先进的图像去雾或小目标检测，很少同时解决这两个挑战。为了解决这些问题，我们提出了一种用于恶劣天气下小目标检测的检测框架udt - yolo。该模型采用端到端多任务优化策略，共同增强图像恢复和检测能力。在YOLOv12的基础上，我们的网络集成了一个图像恢复模块来提高特征提取的鲁棒性，并引入了两个新的模块来丰富表征。首先，特征注意模块增强弱特征，提高对小物体的灵敏度。其次，具有动态变压器的C3模块自适应地关注关键区域并捕获必要的上下文信息，有效地抑制雾引起的干扰。在合成无人机图像数据集HazyDet和真实雾天无人机图像数据集RDDTS上的实验验证了udt - yolo在恶劣天气条件下小目标检测的有效性。

{"title":"UFDT-YOLO: Robust small object detection from UAV perspectives in foggy environments","authors":"Ming Zhao, Jin Chen, Jianyu Zheng, Yanchao Li, Canhui Bao","doi":"10.1016/j.dsp.2025.105826","DOIUrl":"10.1016/j.dsp.2025.105826","url":null,"abstract":"<div><div>Small object detection from UAV perspectives is crucial for autonomous environmental perception, especially under adverse weather such as fog. However, image degradation and the small size of targets often cause insufficient feature representation and background interference, reducing detection performance. Although recent studies have advanced image dehazing or small object detection, few address both challenges simultaneously. To tackle these issues, we propose UFDT-YOLO, a detection framework designed for small object detection in adverse weather. The model employs an end-to-end multi-task optimization strategy, jointly enhancing image restoration and detection. Based on YOLOv12, our network integrates an image restoration module to improve feature extraction robustness and introduces two novel modules for enriched representation. First, the Feature Attention Module strengthens weak features, improving sensitivity to small objects. Second, the C3 Module with Dynamic Transformer adaptively focuses on critical regions and captures essential contextual information, effectively suppressing fog-induced interference. Experiments on the synthetic UAV imagery dataset HazyDet and the real-world foggy UAV imagery dataset RDDTS validate the effectiveness of UFDT-YOLO for small object detection under adverse weather conditions.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105826"},"PeriodicalIF":3.0,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842699","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Federated sparse representation-based anomaly detection 基于联邦稀疏表示的异常检测

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-19 DOI: 10.1016/j.dsp.2025.105828

Michael Melek

Anomaly detection plays a vital role in industrial monitoring, IoT networks, and critical infrastructure protection. Sparse representation and dictionary learning have proved effective for this task, as they provide compact models of normal operation in which anomalies manifest as high reconstruction errors. However, traditional approaches assume centralized training, which is often infeasible due to privacy concerns, communication costs, and heterogeneous data distributions across clients. To address this gap, we propose two federated sparse representation frameworks for anomaly detection: a simple FedAvg-based and an improved method. Both approaches adapt K-SVD dictionary learning to the federated setting, enabling clients to collaboratively learn a global sparse model without sharing raw data. The improved method incorporates three mechanisms that enhance robustness under non-IID conditions, including intelligent atom assignment, contribution-aware weighted aggregation, and momentum-based updates to ensure stable convergence. Extensive experiments on synthetic and benchmark datasets demonstrate that the proposed framework outperforms centralized, local, and federated averaging baselines in terms of detection accuracy, stability, and scalability.

异常检测在工业监控、物联网网络和关键基础设施保护中发挥着至关重要的作用。稀疏表示和字典学习已被证明是有效的，因为它们提供了正常操作的紧凑模型，其中异常表现为高重建误差。然而，传统方法假定集中训练，由于隐私问题、通信成本和跨客户机的异构数据分布，这通常是不可行的。为了解决这一差距，我们提出了两种用于异常检测的联邦稀疏表示框架：一种简单的基于fedag的方法和一种改进的方法。这两种方法都使K-SVD字典学习适应联邦设置，使客户端能够协作学习全局稀疏模型，而无需共享原始数据。改进后的方法采用了智能原子分配、贡献感知加权聚合和基于动量的更新三种机制来增强非iid条件下的鲁棒性，以确保稳定的收敛。在合成和基准数据集上进行的大量实验表明，所提出的框架在检测精度、稳定性和可扩展性方面优于集中式、本地和联邦平均基线。

{"title":"Federated sparse representation-based anomaly detection","authors":"Michael Melek","doi":"10.1016/j.dsp.2025.105828","DOIUrl":"10.1016/j.dsp.2025.105828","url":null,"abstract":"<div><div>Anomaly detection plays a vital role in industrial monitoring, IoT networks, and critical infrastructure protection. Sparse representation and dictionary learning have proved effective for this task, as they provide compact models of normal operation in which anomalies manifest as high reconstruction errors. However, traditional approaches assume centralized training, which is often infeasible due to privacy concerns, communication costs, and heterogeneous data distributions across clients. To address this gap, we propose two federated sparse representation frameworks for anomaly detection: a simple FedAvg-based and an improved method. Both approaches adapt K-SVD dictionary learning to the federated setting, enabling clients to collaboratively learn a global sparse model without sharing raw data. The improved method incorporates three mechanisms that enhance robustness under non-IID conditions, including intelligent atom assignment, contribution-aware weighted aggregation, and momentum-based updates to ensure stable convergence. Extensive experiments on synthetic and benchmark datasets demonstrate that the proposed framework outperforms centralized, local, and federated averaging baselines in terms of detection accuracy, stability, and scalability.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"170 ","pages":"Article 105828"},"PeriodicalIF":3.0,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790620","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

M2NuFFT—A computationally efficient suboptimal power spectrum estimator for fast exploration of nonuniformly sampled time series m2nufft -一种计算效率高的次优功率谱估计器，用于非均匀采样时间序列的快速探测

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-18 DOI: 10.1016/j.dsp.2025.105834

Jie Cui, Benjamin H. Brinkmann, Gregory A. Worrell

Nonuniformly sampled signals are prevalent in real-world applications. However, estimating their power spectra from finite samples poses a significant challenge. The optimal solution—Generalized Prolate Spheroidal Sequence (GPSS) by solving the associated Generalized Eigenvalue Problem (GEP)—is computationally intensive and thus impractical for large datasets. This paper describes a fast, nonparametric method: Multiband-Multitaper Nonuniform Fast Fourier Transform (M²NuFFT), which substantially reduces computational burden while maintaining statistical efficiency. The algorithm partitions the signal frequency band into multiple sub-bands. Within each sub-band, optimal tapers are computed at a nominal analysis band and shifted to other analysis bands using the Nonuniform Fast Fourier Transform (NuFFT), avoiding repeated GEP computations. Spectral power within the analysis band is then estimated as the average power across the taper outputs. For the special case where the nominal band is centered at zero frequency, tapers can be approximated via cubic spline interpolation of Discrete Prolate Spheroidal Sequence (DPSS), eliminating GEP computation entirely. This reduces the complexity from O(N⁴) to as low as

O (N \log N + N \log (1 / ϵ))

. Statistical properties of the estimator, assessed using Bronez GPSS theory, reveal that the bias and variance bound of the M²NuFFT estimator are identical to those of the optimal estimator. Additionally, the degradation of bias bound indicates deviation from optimality. Finally, we propose an extension of Thomson F-test to test periodicity in nonuniform samples. The estimator’s performance is validated through simulation and real-world data, demonstrating its practical applicability. The Matlab code of the fast algorithm is available on GitHub [1].

非均匀采样信号在实际应用中很普遍。然而，从有限的样本中估计它们的功率谱是一个重大的挑战。通过求解相关的广义特征值问题（GEP）得到的最优解——广义长球面序列（GPSS）——计算量大，因此对于大数据集来说不切实际。本文提出了一种快速的非参数方法：多频带多锥度非均匀快速傅里叶变换（M2NuFFT），它在保持统计效率的同时大大减少了计算量。该算法将信号频带划分为多个子频带。在每个子带内，在标称分析带上计算最优锥度，并使用非均匀快速傅里叶变换（NuFFT）将其转移到其他分析带，避免重复的GEP计算。然后将分析频带内的频谱功率估计为整个锥度输出的平均功率。对于名义频带以零频率为中心的特殊情况，可以通过离散延长球面序列（DPSS）的三次样条插值来近似锥度，完全消除了GEP计算。这将复杂度从O（N4）降低到O（NlogN+Nlog(1/ λ)）。利用Bronez GPSS理论评估估计量的统计性质，表明M2NuFFT估计量的偏差和方差界与最优估计量的偏差和方差界相同。此外，偏差界的退化表明偏离最优性。最后，我们提出了汤姆逊f检验的推广，用于检验非均匀样本的周期性。通过仿真和实际数据验证了该估计器的性能，证明了该估计器的实用性。快速算法的Matlab代码可在GitHub[1]上获得。

{"title":"M2NuFFT—A computationally efficient suboptimal power spectrum estimator for fast exploration of nonuniformly sampled time series","authors":"Jie Cui, Benjamin H. Brinkmann, Gregory A. Worrell","doi":"10.1016/j.dsp.2025.105834","DOIUrl":"10.1016/j.dsp.2025.105834","url":null,"abstract":"<div><div>Nonuniformly sampled signals are prevalent in real-world applications. However, estimating their power spectra from finite samples poses a significant challenge. The optimal solution—Generalized Prolate Spheroidal Sequence (GPSS) by solving the associated Generalized Eigenvalue Problem (GEP)—is computationally intensive and thus impractical for large datasets. This paper describes a fast, nonparametric method: Multiband-Multitaper Nonuniform Fast Fourier Transform (M<sup>2</sup>NuFFT), which substantially reduces computational burden while maintaining statistical efficiency. The algorithm partitions the signal frequency band into multiple sub-bands. Within each sub-band, optimal tapers are computed at a nominal analysis band and shifted to other analysis bands using the Nonuniform Fast Fourier Transform (NuFFT), avoiding repeated GEP computations. Spectral power within the analysis band is then estimated as the average power across the taper outputs. For the special case where the nominal band is centered at zero frequency, tapers can be approximated via cubic spline interpolation of Discrete Prolate Spheroidal Sequence (DPSS), eliminating GEP computation entirely. This reduces the complexity from <em>O</em>(<em>N</em><sup>4</sup>) to as low as <span><math><mrow><mi>O</mi><mo>(</mo><mi>N</mi><mi>log</mi><mi>N</mi><mo>+</mo><mi>N</mi><mi>log</mi><mo>(</mo><mn>1</mn><mo>/</mo><mi>ϵ</mi><mo>)</mo><mo>)</mo></mrow></math></span>. Statistical properties of the estimator, assessed using Bronez GPSS theory, reveal that the bias and variance bound of the M<sup>2</sup>NuFFT estimator are identical to those of the optimal estimator. Additionally, the degradation of bias bound indicates deviation from optimality. Finally, we propose an extension of Thomson <em>F</em>-test to test periodicity in nonuniform samples. The estimator’s performance is validated through simulation and real-world data, demonstrating its practical applicability. The <span>Matlab</span> code of the fast algorithm is available on GitHub [1].</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105834"},"PeriodicalIF":3.0,"publicationDate":"2025-12-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842698","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Gridless coherent polarization-DOA estimation using intermediate supervision with angle-guided 基于中间监督的无网格相干偏振方位估计

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-18 DOI: 10.1016/j.dsp.2025.105835

Zihan Wu , Jun Wang , Zhiquan Zhou

Existing gridless direction of arrival (DOA) estimation methods based on deep learning have been extensively investigated. However, there is limited research on gridless coherent polarization-DOA estimation. In the present study, a method was proposed for gridless coherent polarization-DOA estimation incorporating intermediate supervision with angle guidance. Initially, a multi-attention mechanism is employed within a transformer model to extract multi-dimensional features from the spatial-polarization domain of covariance matrix data. Secondly, a designed angle-guided attention mechanism (AGAM) and intermediate supervision head (ISH) are used to enhance the model’s coarse-grained understanding of incoming signals within the spatial and polarization areas during the training phase. Finally, a designed gridless angle prediction head (GAPH) leverages this coarse-grained cognitive capability to achieve fine-grained gridless coherent polarization-DOA estimation during both training and testing phases. Simulation results indicate that the proposed algorithm outperformed existing data-driven algorithms in estimation performance and demonstrates superior accuracy compared to model-driven algorithms, particularly in low signal-to-noise ratio (SNR) scenarios.

现有的基于深度学习的无网格到达方向估计方法得到了广泛的研究。然而，关于无网格相干偏振方位估计的研究还很有限。本文提出了一种结合中间监督和角度制导的无网格相干偏振方位估计方法。首先，在变压器模型中采用多注意机制从协方差矩阵数据的空间极化域中提取多维特征。其次，利用设计的角度引导注意机制（AGAM）和中间监督头（ISH）增强模型在训练阶段对空间和极化区域内输入信号的粗粒度理解；最后，设计了一种无网格角度预测头（GAPH），利用这种粗粒度的认知能力，在训练和测试阶段实现了细粒度的无网格相干偏振方位估计。仿真结果表明，该算法在估计性能上优于现有的数据驱动算法，并且与模型驱动算法相比具有更高的精度，特别是在低信噪比（SNR）场景下。

{"title":"Gridless coherent polarization-DOA estimation using intermediate supervision with angle-guided","authors":"Zihan Wu , Jun Wang , Zhiquan Zhou","doi":"10.1016/j.dsp.2025.105835","DOIUrl":"10.1016/j.dsp.2025.105835","url":null,"abstract":"<div><div>Existing gridless direction of arrival (DOA) estimation methods based on deep learning have been extensively investigated. However, there is limited research on gridless coherent polarization-DOA estimation. In the present study, a method was proposed for gridless coherent polarization-DOA estimation incorporating intermediate supervision with angle guidance. Initially, a multi-attention mechanism is employed within a transformer model to extract multi-dimensional features from the spatial-polarization domain of covariance matrix data. Secondly, a designed angle-guided attention mechanism (AGAM) and intermediate supervision head (ISH) are used to enhance the model’s coarse-grained understanding of incoming signals within the spatial and polarization areas during the training phase. Finally, a designed gridless angle prediction head (GAPH) leverages this coarse-grained cognitive capability to achieve fine-grained gridless coherent polarization-DOA estimation during both training and testing phases. Simulation results indicate that the proposed algorithm outperformed existing data-driven algorithms in estimation performance and demonstrates superior accuracy compared to model-driven algorithms, particularly in low signal-to-noise ratio (SNR) scenarios.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105835"},"PeriodicalIF":3.0,"publicationDate":"2025-12-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842760","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Accurate DFT-based DOA estimation of multi sources in uniform linear array 基于dft的均匀线阵多源精确方位估计

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-17 DOI: 10.1016/j.dsp.2025.105829

Kai Wang, Chuanen Yu, Feiyang Zhong, Zichuan Yu

Accurate estimation of the direction-of-arrival (DOA) in noisy environments is essential, particularly when multiple closely spaced sources generate dense spatial signals. To address this issue, this paper proposes a novel zero-padding DFT-based DOA estimation method for multi-source scenarios under additive white Gaussian noise in a uniform linear array. The proposed algorithm supports up to

H = ⌊ N / 3 ⌋

sources using N sensors. By constructing a linear system of equations from 3H zero-padding DFT bins, our method accurately recovers the true DOAs. Crucially, the approach incorporates the influence of all signal components, enabling unbiased and precise azimuth estimation for each source. Simulation results demonstrate that the proposed algorithm achieves superior noise robustness and higher estimation accuracy compared to existing methods.

在噪声环境中准确估计到达方向（DOA）是必不可少的，特别是当多个紧密间隔的源产生密集的空间信号时。针对这一问题，本文提出了一种基于零填充dft的均匀线性阵列加性高斯白噪声下多源DOA估计方法。该算法支持最多H=⌊N/3⌋的传感器源。该方法通过构造一个由3H填充零的DFT箱组成的线性方程组，准确地恢复了真实的doa。至关重要的是，该方法结合了所有信号分量的影响，能够对每个源进行无偏和精确的方位角估计。仿真结果表明，与现有方法相比，该算法具有较好的噪声鲁棒性和较高的估计精度。

引用次数: 0

LMM-UNet: A lightweight UNet with multi-dilated coordinate enhanced and multi-level mixer for skin lesions segmentation LMM-UNet：一种轻型UNet，具有多重扩展坐标增强和多层次混合，用于皮肤病变分割

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-17 DOI: 10.1016/j.dsp.2025.105833

Feiyu Liu , Siyuan Huang , Fan Zhang , Yepeng Liu

High-precision segmentation of skin lesions is crucial for early detection of skin cancer and improving treatment outcomes. The UNet architecture has made significant progress in the segmentation of skin lesions. However, due to the inherent limitations of convolutional operations, this architecture performs poorly in capturing long-range dependencies between pixels and multi-scale global contextual information. To address this, we propose a lightweight high-precision skin lesion segmentation model, LMM-UNet. The multi-dilated coordinate enhanced (MDCE) module that captures both global and local semantic information by aggregating features along two spatial directions. This design not only strengthens the model’s perception of lesion morphology but also reduces computational complexity. Additionally, we designed a multi-level mixer (MLM) module that hierarchically integrates multi-scale encoder features under the guidance of semantic masks from the decoder. It adopts a shared MLP to learn channel weights, capturing long-term fine-grained features in two spatial directions to enhance detail at the edges and corners of the lesion area, thereby improving segmentation performance. Experimental results on two public datasets, ISIC 2017, ISIC 2018, demonstrate that LMM-UNet achieves superior segmentation performance for skin lesions. And its wide applicability was validated on the GLaS dataset and JSRT dataset. Moreover, LMM-UNet maintains a low parameter count of only 120k.

皮肤病变的高精度分割对于皮肤癌的早期发现和改善治疗效果至关重要。UNet体系结构在皮肤损伤的分割方面取得了重大进展。然而，由于卷积运算的固有局限性，这种架构在捕获像素和多尺度全局上下文信息之间的长期依赖关系方面表现不佳。为了解决这个问题，我们提出了一种轻量级的高精度皮肤病变分割模型LMM-UNet。多扩展坐标增强（MDCE）模块通过沿两个空间方向聚合特征来捕获全局和局部语义信息。这种设计不仅增强了模型对病变形态的感知，而且降低了计算复杂度。此外，我们设计了一个多级混频器（MLM）模块，该模块在解码器的语义掩码的指导下分层集成多尺度编码器特征。采用共享MLP学习信道权值，在两个空间方向上捕获长时间的细粒度特征，增强病灶区域边缘和角落的细节，从而提高分割性能。在ISIC 2017、ISIC 2018两个公共数据集上的实验结果表明，LMM-UNet对皮肤病变具有较好的分割性能。并在GLaS数据集和JSRT数据集上验证了其广泛的适用性。此外，LMM-UNet的参数计数很低，只有120k。

{"title":"LMM-UNet: A lightweight UNet with multi-dilated coordinate enhanced and multi-level mixer for skin lesions segmentation","authors":"Feiyu Liu , Siyuan Huang , Fan Zhang , Yepeng Liu","doi":"10.1016/j.dsp.2025.105833","DOIUrl":"10.1016/j.dsp.2025.105833","url":null,"abstract":"<div><div>High-precision segmentation of skin lesions is crucial for early detection of skin cancer and improving treatment outcomes. The UNet architecture has made significant progress in the segmentation of skin lesions. However, due to the inherent limitations of convolutional operations, this architecture performs poorly in capturing long-range dependencies between pixels and multi-scale global contextual information. To address this, we propose a lightweight high-precision skin lesion segmentation model, LMM-UNet. The multi-dilated coordinate enhanced (MDCE) module that captures both global and local semantic information by aggregating features along two spatial directions. This design not only strengthens the model’s perception of lesion morphology but also reduces computational complexity. Additionally, we designed a multi-level mixer (MLM) module that hierarchically integrates multi-scale encoder features under the guidance of semantic masks from the decoder. It adopts a shared MLP to learn channel weights, capturing long-term fine-grained features in two spatial directions to enhance detail at the edges and corners of the lesion area, thereby improving segmentation performance. Experimental results on two public datasets, ISIC 2017, ISIC 2018, demonstrate that LMM-UNet achieves superior segmentation performance for skin lesions. And its wide applicability was validated on the GLaS dataset and JSRT dataset. Moreover, LMM-UNet maintains a low parameter count of only 120k.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105833"},"PeriodicalIF":3.0,"publicationDate":"2025-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842761","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 bionic spiking recurrent neural network with sparse connections and Dale’s principle for image and speech recognition 一个具有稀疏连接的仿生尖峰循环神经网络和戴尔的图像和语音识别原理

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-17 DOI: 10.1016/j.dsp.2025.105808

Yun Zhang , Lang Xue , Xiaoling Luo , Ping Li , Hong Qu

Both recurrent neural networks (RNNs) and spiking neural networks (SNNs) are adept at processing temporal data, yet they present a fundamental trade-off: RNNs achieve higher performance but lack biological plausibility, whereas SNNs offer greater biological interpretability at the cost of computational efficiency and learning performance. To bridge this gap, we introduce a novel Bionic Spiking Recurrent Neural Network (BSRNN), a unified framework that integrates multiple bio-inspired mechanisms to enhance biological realism without compromising competitive performance. The BSRNN incorporates: (1) a novel weight-control rule that adheres to Dale’s Principle, clearly distinguishing excitatory and inhibitory neurons to improve biological interpretability; (2) sparse connectivity, inspired by biological networks to enhance processing efficiency; and (3) spiking neurons with spike-frequency adaptation to capture key neural dynamics. Experimental results show that the BSRNN delivers competitive performance on benchmark datasets including MNIST, N-MNIST, Fashion-MNIST, and TIDIGITS. Furthermore, the BSRNN model is 1 to 3 orders of magnitude more computationally efficient than LSTM and RNN models through theoretical analysis.

递归神经网络（rnn）和峰值神经网络（snn）都擅长处理时间数据，但它们存在一个基本的权衡：rnn实现了更高的性能，但缺乏生物合理性，而snn以计算效率和学习性能为代价提供了更高的生物可解释性。为了弥补这一差距，我们引入了一种新的仿生脉冲递归神经网络（BSRNN），这是一个统一的框架，集成了多种生物启发机制，在不影响竞争表现的情况下增强生物真实感。BSRNN包含：(1)一种新的体重控制规则，遵循戴尔原理，明确区分兴奋性和抑制性神经元，以提高生物学可解释性；(2)稀疏连接，受生物网络启发，提高处理效率；(3)具有尖峰频率适应性的尖峰神经元捕捉关键的神经动力学。实验结果表明，BSRNN在包括MNIST、N-MNIST、Fashion-MNIST和TIDIGITS在内的基准数据集上具有竞争力的性能。通过理论分析，BSRNN模型的计算效率比LSTM和RNN模型高1 ~ 3个数量级。

{"title":"A bionic spiking recurrent neural network with sparse connections and Dale’s principle for image and speech recognition","authors":"Yun Zhang , Lang Xue , Xiaoling Luo , Ping Li , Hong Qu","doi":"10.1016/j.dsp.2025.105808","DOIUrl":"10.1016/j.dsp.2025.105808","url":null,"abstract":"<div><div>Both recurrent neural networks (RNNs) and spiking neural networks (SNNs) are adept at processing temporal data, yet they present a fundamental trade-off: RNNs achieve higher performance but lack biological plausibility, whereas SNNs offer greater biological interpretability at the cost of computational efficiency and learning performance. To bridge this gap, we introduce a novel Bionic Spiking Recurrent Neural Network (BSRNN), a unified framework that integrates multiple bio-inspired mechanisms to enhance biological realism without compromising competitive performance. The BSRNN incorporates: (1) a novel weight-control rule that adheres to Dale’s Principle, clearly distinguishing excitatory and inhibitory neurons to improve biological interpretability; (2) sparse connectivity, inspired by biological networks to enhance processing efficiency; and (3) spiking neurons with spike-frequency adaptation to capture key neural dynamics. Experimental results show that the BSRNN delivers competitive performance on benchmark datasets including MNIST, N-MNIST, Fashion-MNIST, and TIDIGITS. Furthermore, the BSRNN model is 1 to 3 orders of magnitude more computationally efficient than LSTM and RNN models through theoretical analysis.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105808"},"PeriodicalIF":3.0,"publicationDate":"2025-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842754","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Context and intent enhanced target tracking 上下文和意图增强了目标跟踪

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-17 DOI: 10.1016/j.dsp.2025.105836

Lubos Vaci , Marco Mari , Lauro Snidaro , Gian Luca Foresti

Exploitation of contextual knowledge has recently emerged as a promising approach to increase the performance of Information Fusion systems. Despite pioneering efforts in context assisted target tracking, the realm is still in its infancy, as the frameworks combining context reasoning with target tracking are not abundant. We here postulate that, in addition to physical constraints, such as the road network, knowledge of common patterns that targets pursue can significantly improve tracking accuracy and continuity. In the presented approach, we address the problem of tracking ground targets in complex urban environments, which generally poses a challenge to modern airborne surveillance systems. A target’s actions are modeled as a Markov chain with relevant context defining transition and emission probabilities. Target’s kinematics are estimated by the Interacting Multiple Models (IMM) filter that estimates the mode transition probability matrix (TPM) at each recursion step. The TPM posterior is computed by a Quasi-Bayesian estimator conditioned on the prior and the likelihood originating from target’s measurements and the context. Through extensive simulations, we demonstrate that incorporating contextual information into TPM estimation significantly improves the filtering performance compared to both the IMM filter with a fixed TPM and adaptive TPM estimation without considering contextual information.

利用上下文知识最近成为提高信息融合系统性能的一种很有前途的方法。尽管在上下文辅助目标跟踪方面做出了开创性的努力，但由于将上下文推理与目标跟踪相结合的框架并不丰富，因此该领域仍处于起步阶段。我们在此假设，除了物理约束，如道路网络，目标追求的共同模式的知识可以显著提高跟踪的准确性和连续性。在提出的方法中，我们解决了在复杂的城市环境中跟踪地面目标的问题，这通常对现代机载监视系统构成挑战。目标的行为被建模为具有相关上下文定义转移和发射概率的马尔可夫链。通过多模型交互（IMM）滤波器估计目标的运动学，该滤波器估计每个递归步骤的模式转移概率矩阵（TPM）。TPM后验由拟贝叶斯估计量计算，该估计量的条件是先验和来自目标测量和上下文的似然。通过大量的模拟，我们证明，与具有固定TPM的IMM滤波器和不考虑上下文信息的自适应TPM估计相比，将上下文信息纳入TPM估计显着提高了过滤性能。

{"title":"Context and intent enhanced target tracking","authors":"Lubos Vaci , Marco Mari , Lauro Snidaro , Gian Luca Foresti","doi":"10.1016/j.dsp.2025.105836","DOIUrl":"10.1016/j.dsp.2025.105836","url":null,"abstract":"<div><div>Exploitation of contextual knowledge has recently emerged as a promising approach to increase the performance of Information Fusion systems. Despite pioneering efforts in context assisted target tracking, the realm is still in its infancy, as the frameworks combining context reasoning with target tracking are not abundant. We here postulate that, in addition to physical constraints, such as the road network, knowledge of common patterns that targets pursue can significantly improve tracking accuracy and continuity. In the presented approach, we address the problem of tracking ground targets in complex urban environments, which generally poses a challenge to modern airborne surveillance systems. A target’s actions are modeled as a Markov chain with relevant context defining transition and emission probabilities. Target’s kinematics are estimated by the Interacting Multiple Models (IMM) filter that estimates the mode transition probability matrix (TPM) at each recursion step. The TPM posterior is computed by a Quasi-Bayesian estimator conditioned on the prior and the likelihood originating from target’s measurements and the context. Through extensive simulations, we demonstrate that incorporating contextual information into TPM estimation significantly improves the filtering performance compared to both the IMM filter with a fixed TPM and adaptive TPM estimation without considering contextual information.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105836"},"PeriodicalIF":3.0,"publicationDate":"2025-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842758","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Overcoming radar sparsity and cross-view misalignment: A sparse-to-sparse fusion paradigm for robust 3D object detection 克服雷达稀疏性和交叉视不对准：用于鲁棒3D目标检测的稀疏到稀疏融合范式

IF 3 3区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

Digital Signal Processing

Pub Date : 2025-12-16 DOI: 10.1016/j.dsp.2025.105831

Rui Wan , Weigang Meng , Tianyun Zhao , Wei Lu

Millimeter-wave radar-camera fusion provides a cost-effective alternative to LiDAR for 3D perception in autonomous driving. However, its potential is constrained by two limitations: (1) Previous bird’s-eye view fusion methods struggle to accurately align and fuse cross-modal features, while the fusion strategy incurs substantial computational redundancy from background processing. (2) The extreme sparsity of radar points (typically  < 5 % LiDAR density) hinders robust geometric measurement. To address these challenges, we propose a radar-camera fusion 3D detection framework that redefines cross-modal interaction by transitioning from dense fusion to sparse-to-sparse paradigm. This transformation is initiated by generating spatially-aware 3D object queries from images and radar sweeps-leveraging image-derived seed points with radar depth to anchor queries to objects via a perspective-guided object query generator. Moreover, we introduce adaptive radar pillar diffusion within foreground regions to mitigate radar sparsity, allowing object queries to capture geometric information from diffused pillars. Additionally, to maximize image semantic clues, we further refine object boxes through image keypoint feature aggregation using a keypoint-aware object refinement module. Our framework not only circumvents traditional fusion bottlenecks but also achieves real-time inference at 23.4 FPS. Evaluated on nuScenes dataset, it demonstrates competitive detection performance (65.0 % NDS, 57.8 % mAP) and tracking precision (58.3 % AMOTA and 0.687m AMOTP). By overcoming sparsity constraints and improving cross-modal fusion, this work establishes a new paradigm for robust perception systems.

毫米波雷达-相机融合技术为自动驾驶领域的3D感知提供了一种经济有效的替代激光雷达技术。然而，其潜力受到两个方面的限制：(1)以往的鸟瞰图融合方法难以准确对齐和融合跨模态特征，而融合策略由于后台处理而产生大量的计算冗余。(2)雷达点的极端稀疏性（通常为 <； 5% LiDAR密度）阻碍了稳健的几何测量。为了应对这些挑战，我们提出了一个雷达-相机融合3D检测框架，该框架通过从密集融合过渡到稀疏到稀疏范式来重新定义跨模态交互。这种转换是通过从图像和雷达扫描中生成空间感知的3D对象查询来启动的——利用具有雷达深度的图像衍生种子点，通过透视引导的对象查询生成器将查询锚定到对象。此外，我们在前景区域内引入自适应雷达柱扩散来减轻雷达稀疏性，允许对象查询从扩散柱中捕获几何信息。此外，为了最大化图像语义线索，我们使用关键点感知对象细化模块，通过图像关键点特征聚合进一步细化对象框。我们的框架不仅绕过了传统的融合瓶颈，而且在23.4 FPS的速度下实现了实时推理。在nuScenes数据集上进行了评估，结果表明该方法具有较好的检测性能（65.0% NDS, 57.8% mAP）和跟踪精度（58.3% AMOTA和0.687m AMOTP）。通过克服稀疏性约束和改进跨模态融合，本研究为鲁棒感知系统建立了一个新的范式。

{"title":"Overcoming radar sparsity and cross-view misalignment: A sparse-to-sparse fusion paradigm for robust 3D object detection","authors":"Rui Wan , Weigang Meng , Tianyun Zhao , Wei Lu","doi":"10.1016/j.dsp.2025.105831","DOIUrl":"10.1016/j.dsp.2025.105831","url":null,"abstract":"<div><div>Millimeter-wave radar-camera fusion provides a cost-effective alternative to LiDAR for 3D perception in autonomous driving. However, its potential is constrained by two limitations: (1) Previous bird’s-eye view fusion methods struggle to accurately align and fuse cross-modal features, while the fusion strategy incurs substantial computational redundancy from background processing. (2) The extreme sparsity of radar points (typically  < 5 % LiDAR density) hinders robust geometric measurement. To address these challenges, we propose a radar-camera fusion 3D detection framework that redefines cross-modal interaction by transitioning from dense fusion to sparse-to-sparse paradigm. This transformation is initiated by generating spatially-aware 3D object queries from images and radar sweeps-leveraging image-derived seed points with radar depth to anchor queries to objects via a perspective-guided object query generator. Moreover, we introduce adaptive radar pillar diffusion within foreground regions to mitigate radar sparsity, allowing object queries to capture geometric information from diffused pillars. Additionally, to maximize image semantic clues, we further refine object boxes through image keypoint feature aggregation using a keypoint-aware object refinement module. Our framework not only circumvents traditional fusion bottlenecks but also achieves real-time inference at 23.4 FPS. Evaluated on nuScenes dataset, it demonstrates competitive detection performance (65.0 % NDS, 57.8 % mAP) and tracking precision (58.3 % AMOTA and 0.687m AMOTP). By overcoming sparsity constraints and improving cross-modal fusion, this work establishes a new paradigm for robust perception systems.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105831"},"PeriodicalIF":3.0,"publicationDate":"2025-12-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842757","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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