Digital Signal Processing最新文献_第7页

Proportionate kernel risk sensitive loss adaptive filtering algorithms and their performance analysis for sparse system identification under non-Gaussian noise 非高斯噪声下稀疏系统识别的比例核风险敏感损失自适应滤波算法及其性能分析

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

Digital Signal Processing

Pub Date : 2025-12-13 DOI: 10.1016/j.dsp.2025.105801

Ben-Xue Su , Hui Li , Kun-De Yang , Lian-You Jing , Tian-He Liu

In practical communication systems, achieving accurate channel estimation under sparse non-Gaussian noise channel environments is a prerequisite for ensuring reliable signal transmission. However, most existing methods focus on system identification under Gaussian noise environments. Although the Kernel Risk-Sensitive Loss (KRSL) algorithm exhibits excellent steady-state performance in non-Gaussian impulsive noise, it fails to fully utilize prior channel information. To address this issue, we propose the proportionate KRSL (PKRSL) algorithm. This algorithm improves the recursive version of the KRSL algorithm by introducing a proportionate matrix. Concurrently, we develop the zero-attracting (ZA), reweighted zero-attracting (RZA), and l0-norm variants with proportionate matrix, named as the convex regularized PKRSL (CR-PKRSL) algorithm. While preserving robustness against non-Gaussian noise, the proposed algorithms enable more efficient integration of prior channel sparsity information and make better use of limited training sequences. This not only significantly accelerates the convergence speed but also reduces the estimation error of the algorithms. Theoretical analysis of the PKRSL algorithm is conducted from the perspective of first-order and second-order statistical characteristics of steady-state, and the selection ranges for the step size and convex penalty strength are provided. The results demonstrate that under sparse channel conditions and in a non-Gaussian noise environment, both the PKRSL and CR-PKRSL algorithms exhibit greater robustness and faster convergence speed compared with traditional algorithms. Finally, experimental validations confirm the consistency between the theoretically derived steady-state deviation and the simulation results, thus verifying the correctness of the theoretical analysis.

在实际通信系统中，在稀疏非高斯噪声信道环境下实现准确的信道估计是保证信号可靠传输的前提。然而，现有的方法大多集中在高斯噪声环境下的系统识别。虽然核风险敏感损失（KRSL）算法在非高斯脉冲噪声中表现出优异的稳态性能，但它不能充分利用先验信道信息。为了解决这个问题，我们提出了比例KRSL （PKRSL）算法。该算法通过引入比例矩阵改进了KRSL算法的递归版本。同时，我们开发了零吸引（ZA）、重加权零吸引（RZA）和带有比例矩阵的十范数变体，称为凸正则化PKRSL （CR-PKRSL）算法。在保持对非高斯噪声的鲁棒性的同时，该算法能够更有效地整合先验信道稀疏性信息，并更好地利用有限的训练序列。这不仅大大加快了算法的收敛速度，而且减小了算法的估计误差。从稳态一阶和二阶统计特性的角度对PKRSL算法进行了理论分析，给出了步长和凸罚强度的选择范围。结果表明，在稀疏信道条件和非高斯噪声环境下，与传统算法相比，PKRSL和CR-PKRSL算法具有更强的鲁棒性和更快的收敛速度。最后通过实验验证了理论推导的稳态偏差与仿真结果的一致性，从而验证了理论分析的正确性。

{"title":"Proportionate kernel risk sensitive loss adaptive filtering algorithms and their performance analysis for sparse system identification under non-Gaussian noise","authors":"Ben-Xue Su , Hui Li , Kun-De Yang , Lian-You Jing , Tian-He Liu","doi":"10.1016/j.dsp.2025.105801","DOIUrl":"10.1016/j.dsp.2025.105801","url":null,"abstract":"<div><div>In practical communication systems, achieving accurate channel estimation under sparse non-Gaussian noise channel environments is a prerequisite for ensuring reliable signal transmission. However, most existing methods focus on system identification under Gaussian noise environments. Although the Kernel Risk-Sensitive Loss (KRSL) algorithm exhibits excellent steady-state performance in non-Gaussian impulsive noise, it fails to fully utilize prior channel information. To address this issue, we propose the proportionate KRSL (PKRSL) algorithm. This algorithm improves the recursive version of the KRSL algorithm by introducing a proportionate matrix. Concurrently, we develop the zero-attracting (ZA), reweighted zero-attracting (RZA), and l0-norm variants with proportionate matrix, named as the convex regularized PKRSL (CR-PKRSL) algorithm. While preserving robustness against non-Gaussian noise, the proposed algorithms enable more efficient integration of prior channel sparsity information and make better use of limited training sequences. This not only significantly accelerates the convergence speed but also reduces the estimation error of the algorithms. Theoretical analysis of the PKRSL algorithm is conducted from the perspective of first-order and second-order statistical characteristics of steady-state, and the selection ranges for the step size and convex penalty strength are provided. The results demonstrate that under sparse channel conditions and in a non-Gaussian noise environment, both the PKRSL and CR-PKRSL algorithms exhibit greater robustness and faster convergence speed compared with traditional algorithms. Finally, experimental validations confirm the consistency between the theoretically derived steady-state deviation and the simulation results, thus verifying the correctness of the theoretical analysis.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105801"},"PeriodicalIF":3.0,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842753","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

Error mechanism analysis of MUSIC-based active angle measurement under retrodirective cross-eye jamming 反向对眼干扰下music主动角度测量误差机理分析

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

Digital Signal Processing

Pub Date : 2025-12-13 DOI: 10.1016/j.dsp.2025.105830

Jie Xue, Binbin Su, Yongcai Liu, Haonan Yang, Dong Li, Jin Meng

Spatial spectrum techniques such as Multiple Signal Classification (MUSIC), with their advantages of high-precision and super-resolution angle measurement, represent one of the core development directions for the new generation of active angle measurement technologies. However, the performance of spatial spectrum techniques in active angle measurement under cross-eye jamming remains unclear. This paper first establishes a mathematical model of retrodirective cross-eye jamming (RCJ) for the MUSIC active angle measurement system, derives expressions for the indicated angle and cross-eye gain to evaluate jamming effects. Then, an array factor ratio parameter is defined to quantify the nonlinear impact of radar transmit-receive channel inconsistencies on angular errors. Finally, the consistency of jamming responses between MUSIC and sum-difference beam angle measurement techniques is evaluated by comparing angular error under identical RCJ conditions. The results show that when the two RCJ signals in the radar receive channel tend to be of equal amplitude and opposite phase, an increase in cross-eye gain leads to a significant increase in angular errors. The inconsistency of radar transmit-receive channels modulates the amplitude ratio and phase difference of RCJ signals, thereby exerting a secondary influence on angular errors. This study provides an important theoretical basis for jamming technologies against radars employing spatial spectrum estimation.

多信号分类（MUSIC）等空间频谱技术以其高精度、超分辨率的角度测量优势，代表了新一代主动角度测量技术的核心发展方向之一。然而，空间频谱技术在交叉眼干扰下主动角度测量中的性能尚不清楚。本文首先建立了MUSIC有源测角系统的反向对眼干扰（RCJ）的数学模型，推导了指示角和对眼增益的表达式，以评价干扰效果。然后，定义阵列因子比参数，量化雷达收发信道不一致性对角误差的非线性影响。最后，通过比较相同rj条件下的角度误差，评价了MUSIC和和差波束角测量技术干扰响应的一致性。结果表明，当雷达接收通道中的两个RCJ信号趋于幅值相等、相位相反时，交眼增益的增加会导致角误差的显著增加。雷达收发信道的不一致性调制了RCJ信号的幅值比和相位差，从而对角误差产生二次影响。该研究为采用空间频谱估计的雷达干扰技术提供了重要的理论依据。

{"title":"Error mechanism analysis of MUSIC-based active angle measurement under retrodirective cross-eye jamming","authors":"Jie Xue, Binbin Su, Yongcai Liu, Haonan Yang, Dong Li, Jin Meng","doi":"10.1016/j.dsp.2025.105830","DOIUrl":"10.1016/j.dsp.2025.105830","url":null,"abstract":"<div><div>Spatial spectrum techniques such as Multiple Signal Classification (MUSIC), with their advantages of high-precision and super-resolution angle measurement, represent one of the core development directions for the new generation of active angle measurement technologies. However, the performance of spatial spectrum techniques in active angle measurement under cross-eye jamming remains unclear. This paper first establishes a mathematical model of retrodirective cross-eye jamming (RCJ) for the MUSIC active angle measurement system, derives expressions for the indicated angle and cross-eye gain to evaluate jamming effects. Then, an array factor ratio parameter is defined to quantify the nonlinear impact of radar transmit-receive channel inconsistencies on angular errors. Finally, the consistency of jamming responses between MUSIC and sum-difference beam angle measurement techniques is evaluated by comparing angular error under identical RCJ conditions. The results show that when the two RCJ signals in the radar receive channel tend to be of equal amplitude and opposite phase, an increase in cross-eye gain leads to a significant increase in angular errors. The inconsistency of radar transmit-receive channels modulates the amplitude ratio and phase difference of RCJ signals, thereby exerting a secondary influence on angular errors. This study provides an important theoretical basis for jamming technologies against radars employing spatial spectrum estimation.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105830"},"PeriodicalIF":3.0,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145799837","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

STFL-Net: A spatial-temporal feature learning network for breast tumor segmentation in multi-sequence MRI STFL-Net：一种用于多序列MRI乳腺肿瘤分割的时空特征学习网络

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

Digital Signal Processing

Pub Date : 2025-12-13 DOI: 10.1016/j.dsp.2025.105825

Xiangbin Luo , Zhitao Wei , Yanting Liang , Chinting Wong , Zeyan Xu , Kaili Liu , Chu Han , Zhen Zhang , Zaiyi Liu , Zhenwei Shi , Ying Wang

Multi-sequence MRI, including dynamic contrast-enhanced MRI (DCE-MRI) and T2-weighted (T2W) provide crucial morphological and temporal information for accurate breast tumor segmentation. However, researchers often only use multi-sequence MRI or single phase DCE-MRI to study breast tumors, and fail to combine the relevant information of the two. Most existing methods also only consider multi-sequence MRI, often ignoring the temporal correlation between different DCE phases, and fail to fully utilize the complementary information across sequences. To address these issues, we propose STFL-Net, a novel spatial-temporal feature fusion learning network that integrates multi-sequence and multi-phase MRI features to enhance segmentation performance. STFL-Net employs a multi-branch encoder for modality-specific feature extraction, a feature selection (FS) module for capturing texture-rich information using multi-scale convolutions and attention mechanisms, and two fusion modules (multi-sequence fusion (MSF) and temporal-spatial fusion (TSF)) to dynamically aggregate spatial and temporal semantics. Evaluated on a dataset of 600 high-risk breast cancer MRI scans, STFL-Net achieved a Dice Similarity Coefficient (DSC) of 80.18 %, Intersection over Union (IoU) of 70.05 %, and Sensitivity (SEN) of 84.42 %. Comparative experiments demonstrate the superior performance and strong generalization capability of STFL-Net in breast tumor segmentation tasks.

多序列MRI，包括动态对比增强MRI （DCE-MRI）和t2加权（T2W）为准确的乳腺肿瘤分割提供了重要的形态学和时间信息。然而，研究人员往往只使用多序列MRI或单相DCE-MRI来研究乳腺肿瘤，未能将两者的相关信息结合起来。现有的方法大多只考虑多序列MRI，往往忽略了不同DCE阶段之间的时间相关性，未能充分利用序列间的互补信息。为了解决这些问题，我们提出了一种新的时空特征融合学习网络STFL-Net，它集成了多序列和多阶段的MRI特征，以提高分割性能。STFL-Net使用多分支编码器进行模态特征提取，使用多尺度卷积和注意机制捕获纹理丰富信息的特征选择（FS）模块，以及两个融合模块（多序列融合（MSF）和时空融合（TSF））来动态聚合时空语义。在600个高危乳腺癌MRI扫描数据集上进行评估，STFL-Net的Dice Similarity Coefficient （DSC）为80.18%,Intersection over Union （IoU）为70.05%,Sensitivity （SEN）为84.42%。对比实验证明了STFL-Net在乳腺肿瘤分割任务中的优越性能和较强的泛化能力。

{"title":"STFL-Net: A spatial-temporal feature learning network for breast tumor segmentation in multi-sequence MRI","authors":"Xiangbin Luo , Zhitao Wei , Yanting Liang , Chinting Wong , Zeyan Xu , Kaili Liu , Chu Han , Zhen Zhang , Zaiyi Liu , Zhenwei Shi , Ying Wang","doi":"10.1016/j.dsp.2025.105825","DOIUrl":"10.1016/j.dsp.2025.105825","url":null,"abstract":"<div><div>Multi-sequence MRI, including dynamic contrast-enhanced MRI (DCE-MRI) and T2-weighted (T2W) provide crucial morphological and temporal information for accurate breast tumor segmentation. However, researchers often only use multi-sequence MRI or single phase DCE-MRI to study breast tumors, and fail to combine the relevant information of the two. Most existing methods also only consider multi-sequence MRI, often ignoring the temporal correlation between different DCE phases, and fail to fully utilize the complementary information across sequences. To address these issues, we propose STFL-Net, a novel spatial-temporal feature fusion learning network that integrates multi-sequence and multi-phase MRI features to enhance segmentation performance. STFL-Net employs a multi-branch encoder for modality-specific feature extraction, a feature selection (FS) module for capturing texture-rich information using multi-scale convolutions and attention mechanisms, and two fusion modules (multi-sequence fusion (MSF) and temporal-spatial fusion (TSF)) to dynamically aggregate spatial and temporal semantics. Evaluated on a dataset of 600 high-risk breast cancer MRI scans, STFL-Net achieved a Dice Similarity Coefficient (DSC) of 80.18 %, Intersection over Union (IoU) of 70.05 %, and Sensitivity (SEN) of 84.42 %. Comparative experiments demonstrate the superior performance and strong generalization capability of STFL-Net in breast tumor segmentation tasks.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105825"},"PeriodicalIF":3.0,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842751","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

Multi-layer image encryption framework using a chaotic system with a dynamic starting point 多层图像加密框架采用混沌系统的动态起始点

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

Digital Signal Processing

Pub Date : 2025-12-12 DOI: 10.1016/j.dsp.2025.105823

Yavar Mousavi , Ali Shokri , Yavar Khedmati Yengejeh , Hossein Kheiri

This paper presents a novel image encryption framework that significantly enhances chaotic encryption through two synergistic innovations. First, we introduce a new class of chaotic systems based on ill-conditioned matrix operations, exhibiting extreme sensitivity to initial conditions and transforming input images into highly disordered states. Second, we propose a dynamic initialization mechanism in which encryption parameters are uniquely generated for each session by combining cryptographic keys with features extracted from the plaintext image. This approach eliminates static vulnerabilities common in conventional systems. The framework also integrates an adaptive pixel shuffling process, applying variable circular shifts governed by the evolving state of the chaotic system to effectively disrupt statistical patterns. A comprehensive security analysis demonstrates that the proposed system achieves ideal encryption metrics, including high information entropy and resistance to statistical and differential attacks.

本文提出了一种新的图像加密框架，通过两个协同创新显著增强了混沌加密。首先，我们引入了一类新的基于病态矩阵运算的混沌系统，它对初始条件表现出极高的敏感性，并将输入图像转换为高度无序的状态。其次，我们提出了一种动态初始化机制，其中通过将加密密钥与从明文图像中提取的特征相结合，为每个会话唯一地生成加密参数。这种方法消除了传统系统中常见的静态漏洞。该框架还集成了自适应像素变换过程，应用由混沌系统的演化状态控制的可变圆位移来有效地破坏统计模式。全面的安全分析表明，该系统实现了理想的加密指标，包括高信息熵和抗统计和差分攻击。

引用次数: 0

DIMTrack: Dynamic Neuron-Based RGBX Tracking for Multimodal Visual Cross-Modal Interaction 基于神经元的动态RGBX多模态视觉跨模态交互跟踪

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

Digital Signal Processing

Pub Date : 2025-12-12 DOI: 10.1016/j.dsp.2025.105827

Yuhan Zhao , Kehan Cao , Yue Wu , Yaocong Hu , Liang Tao , Bingyou Liu

Multimodal sensors are invaluable in visual tracking tasks due to their distinct advantages in addressing various challenging scenarios. Although it would be ideal to utilize a common model for all modalities, practical applications often necessitate the use of a single modality due to data scarcity. This study first proposes a multimodal tracking method based on the Dynamic Interaction Module (DIM), which aims to enhance tracking performance by dynamically fusing the feature information from RGB and X modalities. During the mixed training process, the DIM module facilitates the matching of similarities between modalities and provides a unified training framework for diverse modalities. Additionally, the DIM module employs expert models (including Avg Expert and Mix Expert) to balance the feature representations of various modalities, thereby ensuring feature optimization and enhancing model performance. Through extensive experiments with paired modalities such as RGB-E, RGB-D, and RGB-T, we demonstrate that the proposed method outperforms the RGB-X tracker during the inference process. Dynamic neurons optimize the fusion of cross-modal features by selectively focusing on the most relevant modal features, further enhancing the model’s robustness and tracking accuracy. Experimental results indicate that the proposed method significantly enhances performance in multimodal tracking tasks and demonstrates exceptional effectiveness and flexibility in complex scenarios.

多模态传感器在视觉跟踪任务中具有不可估量的价值，因为它们在解决各种具有挑战性的场景方面具有独特的优势。虽然对所有模式使用一个通用模型是理想的，但由于数据稀缺，实际应用通常需要使用单一模式。本研究首先提出了一种基于动态交互模块（DIM）的多模态跟踪方法，通过动态融合RGB和X模态特征信息来提高跟踪性能。在混合训练过程中，DIM模块促进了模式之间相似性的匹配，为多种模式提供了统一的训练框架。此外，DIM模块采用专家模型（包括Avg expert和Mix expert）来平衡各种模态的特征表示，从而确保特征优化并提高模型性能。通过对RGB-E、RGB-D和RGB-T等配对模式的大量实验，我们证明了所提出的方法在推理过程中优于RGB-X跟踪器。动态神经元通过选择性地聚焦最相关的模态特征来优化跨模态特征的融合，进一步增强模型的鲁棒性和跟踪精度。实验结果表明，该方法显著提高了多模态跟踪任务的性能，在复杂场景下表现出优异的有效性和灵活性。

{"title":"DIMTrack: Dynamic Neuron-Based RGBX Tracking for Multimodal Visual Cross-Modal Interaction","authors":"Yuhan Zhao , Kehan Cao , Yue Wu , Yaocong Hu , Liang Tao , Bingyou Liu","doi":"10.1016/j.dsp.2025.105827","DOIUrl":"10.1016/j.dsp.2025.105827","url":null,"abstract":"<div><div>Multimodal sensors are invaluable in visual tracking tasks due to their distinct advantages in addressing various challenging scenarios. Although it would be ideal to utilize a common model for all modalities, practical applications often necessitate the use of a single modality due to data scarcity. This study first proposes a multimodal tracking method based on the Dynamic Interaction Module (DIM), which aims to enhance tracking performance by dynamically fusing the feature information from RGB and X modalities. During the mixed training process, the DIM module facilitates the matching of similarities between modalities and provides a unified training framework for diverse modalities. Additionally, the DIM module employs expert models (including Avg Expert and Mix Expert) to balance the feature representations of various modalities, thereby ensuring feature optimization and enhancing model performance. Through extensive experiments with paired modalities such as RGB-E, RGB-D, and RGB-T, we demonstrate that the proposed method outperforms the RGB-X tracker during the inference process. Dynamic neurons optimize the fusion of cross-modal features by selectively focusing on the most relevant modal features, further enhancing the model’s robustness and tracking accuracy. Experimental results indicate that the proposed method significantly enhances performance in multimodal tracking tasks and demonstrates exceptional effectiveness and flexibility in complex scenarios.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105827"},"PeriodicalIF":3.0,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842759","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

Semi-supervised teacher-student multi-modal network for object detection in foggy scenes 雾天场景中半监督师生多模态网络目标检测

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

Digital Signal Processing

Pub Date : 2025-12-12 DOI: 10.1016/j.dsp.2025.105824

Yuqi Yang, Ying Chen

Object detection in foggy weather suffers from serious visual feather degradation due to atmospheric particle scattering, which severely restrict the use of safety-critical applications such as autonomous driving. Vision Language Model (VLM) achieves great attention in assisting vision task, in which the text prompt is critical while it is usually unavailable in reality. To address the above issues, a Semi-supervised Teacher-Student Multi-modal (STSM) Network is proposed, which reduces the domain difference between synthetic and real data under conditions of missing or incomplete textual annotations through a domain adaptive training mechanism. The teacher model leverages complete visual-textual pairs to generate high-confidence pseudo-labels, and the student model learns robust feature representations from partially masked text inputs to simulate the absence of real-world text annotations, and in turn transfers parameters to the teacher model via exponential moving average (EMA). The Vision-Guided Text Completion Module (VGTC) is designed to maintain cross-modal semantic consistency in the process of domain adaptation and dynamically reconstruct missing text. The Spectral-Enhanced Multi-modal Interaction Module (SEMI) is introduced to amplify low-frequency visual semantic features in frequency domain space and improve the ability of domain invariant feature extraction. Experimental results show that the STSM exhibits outstanding performance both on synthetic and real foggy datasets, which achieves an accuracy of 51.2 % on the Foggy Cityscapes dataset, outperforming state-of-the-arts.

大雾天气下的目标检测由于大气粒子散射而遭受严重的视觉羽毛退化，这严重限制了自动驾驶等安全关键应用的使用。视觉语言模型（VLM）在辅助视觉任务中得到了广泛的关注，在视觉任务中文本提示是非常重要的，而在现实中文本提示通常是不可用的。针对上述问题，提出了一种半监督师生多模态网络（Semi-supervised Teacher-Student Multi-modal， STSM），该网络通过域自适应训练机制减少了文本标注缺失或不完整情况下合成数据与真实数据的域差异。教师模型利用完整的视觉文本对来生成高置信度的伪标签，学生模型从部分屏蔽的文本输入中学习鲁棒特征表示，以模拟现实世界文本注释的缺失，并通过指数移动平均（EMA）将参数传递给教师模型。视觉引导文本补全模块（VGTC）的设计是为了在领域自适应过程中保持跨模态语义一致性，并动态重建缺失文本。引入频谱增强多模态交互模块（SEMI），在频域空间放大低频视觉语义特征，提高域不变特征提取能力。实验结果表明，STSM在合成和真实雾天数据集上都表现出优异的性能，在雾天城市景观数据集上达到了51.2%的准确率，超过了目前的水平。

{"title":"Semi-supervised teacher-student multi-modal network for object detection in foggy scenes","authors":"Yuqi Yang, Ying Chen","doi":"10.1016/j.dsp.2025.105824","DOIUrl":"10.1016/j.dsp.2025.105824","url":null,"abstract":"<div><div>Object detection in foggy weather suffers from serious visual feather degradation due to atmospheric particle scattering, which severely restrict the use of safety-critical applications such as autonomous driving. Vision Language Model (VLM) achieves great attention in assisting vision task, in which the text prompt is critical while it is usually unavailable in reality. To address the above issues, a Semi-supervised Teacher-Student Multi-modal (STSM) Network is proposed, which reduces the domain difference between synthetic and real data under conditions of missing or incomplete textual annotations through a domain adaptive training mechanism. The teacher model leverages complete visual-textual pairs to generate high-confidence pseudo-labels, and the student model learns robust feature representations from partially masked text inputs to simulate the absence of real-world text annotations, and in turn transfers parameters to the teacher model via exponential moving average (EMA). The Vision-Guided Text Completion Module (VGTC) is designed to maintain cross-modal semantic consistency in the process of domain adaptation and dynamically reconstruct missing text. The Spectral-Enhanced Multi-modal Interaction Module (SEMI) is introduced to amplify low-frequency visual semantic features in frequency domain space and improve the ability of domain invariant feature extraction. Experimental results show that the STSM exhibits outstanding performance both on synthetic and real foggy datasets, which achieves an accuracy of 51.2 % on the Foggy Cityscapes dataset, outperforming state-of-the-arts.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"170 ","pages":"Article 105824"},"PeriodicalIF":3.0,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790619","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

HPGoG: Hetero-pooling integrated with graph-of-graphs to alleviate class imbalance for fault tracing of transformer substation devices HPGoG：结合图的异构池技术缓解类不平衡的变电站设备故障跟踪

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

Digital Signal Processing

Pub Date : 2025-12-10 DOI: 10.1016/j.dsp.2025.105822

Weizhuo Yao, Changjiang Ju, Genke Yang, Jian Chu

Fault tracing of devices in transformer substations has attracted significant attention. Existing methods primarily rely on manually defined logic or features. And other statistical approaches lack consideration of the connection topology between devices. To address these limitations, this paper designs heterogeneous graphs for fault events, namely event graphs. It integrates the connection topology of devices with alarm signal information. The alarm signal information is transmitted by device monitors during fault events. In addition, an automatic graph classification framework is established to do the fault tracing, using the constructed event graphs. Within this framework, we propose a Hetero-Pooling technique to exploit node label information inherent in heterogeneous graphs. Furthermore, we incorporate the Graph-of-Graphs technique to alleviate the class imbalance issue prevalent in fault tracing data. Experimental results demonstrate that the proposed framework obtains good performance on real-world transformer substation dataset. It also performed well on some commonly used heterogeneous graph classification datasets from other fields.

变电站设备的故障跟踪问题引起了人们的广泛关注。现有的方法主要依赖于手动定义的逻辑或特征。而其他统计方法缺乏对设备间连接拓扑的考虑。为了解决这些局限性，本文设计了故障事件的异构图，即事件图。将设备连接拓扑与告警信号信息集成在一起。在故障事件中，告警信号信息通过设备监视器传输。此外，利用构造的事件图，建立了自动图分类框架进行故障跟踪。在此框架下，我们提出了一种异构池技术来利用异构图中固有的节点标签信息。此外，我们还结合了图的技术来缓解故障跟踪数据中普遍存在的类不平衡问题。实验结果表明，该框架在实际变电站数据集上取得了良好的性能。在其他领域常用的异构图分类数据集上也表现良好。

{"title":"HPGoG: Hetero-pooling integrated with graph-of-graphs to alleviate class imbalance for fault tracing of transformer substation devices","authors":"Weizhuo Yao, Changjiang Ju, Genke Yang, Jian Chu","doi":"10.1016/j.dsp.2025.105822","DOIUrl":"10.1016/j.dsp.2025.105822","url":null,"abstract":"<div><div>Fault tracing of devices in transformer substations has attracted significant attention. Existing methods primarily rely on manually defined logic or features. And other statistical approaches lack consideration of the connection topology between devices. To address these limitations, this paper designs heterogeneous graphs for fault events, namely event graphs. It integrates the connection topology of devices with alarm signal information. The alarm signal information is transmitted by device monitors during fault events. In addition, an automatic graph classification framework is established to do the fault tracing, using the constructed event graphs. Within this framework, we propose a Hetero-Pooling technique to exploit node label information inherent in heterogeneous graphs. Furthermore, we incorporate the Graph-of-Graphs technique to alleviate the class imbalance issue prevalent in fault tracing data. Experimental results demonstrate that the proposed framework obtains good performance on real-world transformer substation dataset. It also performed well on some commonly used heterogeneous graph classification datasets from other fields.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"170 ","pages":"Article 105822"},"PeriodicalIF":3.0,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790624","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

Towards efficient heterogeneous multi-scale & multi-branch feature pyramid network for lightweight detection 面向轻量级检测的高效异构多尺度多分支特征金字塔网络

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

Digital Signal Processing

Pub Date : 2025-12-09 DOI: 10.1016/j.dsp.2025.105787

Xinyu Wang , Yude Wang , Haoyu Guo , Yuanpei Wang , Yidan Cui , Qinrou Ji , Houyi Yan

Aerial images face challenges such as low resolution, complex backgrounds, and low target contrast. Existing methods struggle to balance detection accuracy with lightweight models. To address this issue, we propose EHMNet with a multi-scale, multi-branch architecture. To enable lightweight deployment, we design an efficient heterogeneous multi-scale & multi-branch feature pyramid network (EHMFPN). This network includes various mechanisms and enhancement modules. It allows multi-scale convolutional kernels to capture multi-scale perceptual field information, which improves feature fusion across inter-layer modules. To further maintain detection accuracy, we introduced a multi-scale edge feature aggregation module (MEFA) combined with C3k2. This module extracts features that emphasize edge information from different scales and integrates these multi-scale features. For effective and feasible deployment, we reduce computational overhead using layer-adaptive magnitude-based pruning (LAMP). This method removes redundant network parameters while maintaining model performance. On the VisDrone2019-DET dataset, EHMNet improves the mAP50 by 2.6 % on the validation set and 1.2 % on the test set. At the same time, the computational complexity and number of parameters are reduced by 46.0 % and 70.9 %, respectively, compared to YOLO11-N. The validity of our method is verified through multiple experiments. This approach is expected to be deployed directly on UAVs for airborne small target detection tasks in the future.

航空图像面临着低分辨率、复杂背景和低目标对比度等挑战。现有的方法难以平衡检测精度和轻量级模型。为了解决这个问题，我们提出了一个多规模、多分支架构的EHMNet。为了实现轻量级部署，我们设计了一个高效的异构多尺度多分支特征金字塔网络（EHMFPN）。该网络包括各种机制和增强模块。它允许多尺度卷积核捕获多尺度感知场信息，提高了层间模块的特征融合。为了进一步保持检测精度，我们引入了结合C3k2的多尺度边缘特征聚合模块（MEFA）。该模块从不同尺度提取强调边缘信息的特征，并对这些多尺度特征进行整合。为了有效和可行的部署，我们使用层自适应的基于大小的修剪（LAMP）来减少计算开销。该方法在保持模型性能的同时去除冗余的网络参数。在VisDrone2019-DET数据集上，EHMNet在验证集上提高了2.6%的mAP50，在测试集上提高了1.2%。同时，与YOLO11-N相比，计算复杂度和参数数量分别降低了46.0%和70.9%。通过多次实验验证了该方法的有效性。这种方法有望在未来直接部署在无人机上用于机载小目标探测任务。

{"title":"Towards efficient heterogeneous multi-scale & multi-branch feature pyramid network for lightweight detection","authors":"Xinyu Wang , Yude Wang , Haoyu Guo , Yuanpei Wang , Yidan Cui , Qinrou Ji , Houyi Yan","doi":"10.1016/j.dsp.2025.105787","DOIUrl":"10.1016/j.dsp.2025.105787","url":null,"abstract":"<div><div>Aerial images face challenges such as low resolution, complex backgrounds, and low target contrast. Existing methods struggle to balance detection accuracy with lightweight models. To address this issue, we propose EHMNet with a multi-scale, multi-branch architecture. To enable lightweight deployment, we design an efficient heterogeneous multi-scale & multi-branch feature pyramid network (EHMFPN). This network includes various mechanisms and enhancement modules. It allows multi-scale convolutional kernels to capture multi-scale perceptual field information, which improves feature fusion across inter-layer modules. To further maintain detection accuracy, we introduced a multi-scale edge feature aggregation module (MEFA) combined with C3k2. This module extracts features that emphasize edge information from different scales and integrates these multi-scale features. For effective and feasible deployment, we reduce computational overhead using layer-adaptive magnitude-based pruning (LAMP). This method removes redundant network parameters while maintaining model performance. On the VisDrone2019-DET dataset, EHMNet improves the mAP50 by 2.6 % on the validation set and 1.2 % on the test set. At the same time, the computational complexity and number of parameters are reduced by 46.0 % and 70.9 %, respectively, compared to YOLO11-N. The validity of our method is verified through multiple experiments. This approach is expected to be deployed directly on UAVs for airborne small target detection tasks in the future.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"170 ","pages":"Article 105787"},"PeriodicalIF":3.0,"publicationDate":"2025-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790621","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

Electromagnetic signal recognition using multimodal tri-branch semantic fusion network in the UAV-assist integrated sensing and communication systems 基于多模态三分支语义融合网络的无人机辅助集成传感与通信系统电磁信号识别

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

Digital Signal Processing

Pub Date : 2025-12-08 DOI: 10.1016/j.dsp.2025.105820

Tiantian Wang , Nan Yan , Chaosan Yang , Zeliang An , Gongjing Zhang , Yuqing Xu

Driven by the proliferation of integrated sensing and communication (ISAC) systems, the accurate recognition of unauthorized unmanned aerial vehicle (UAV) signals in dynamic electromagnetic environments has emerged as a critical challenge for spectrum security and cognitive radio applications. Conventional automatic modulation recognition (AMR) frameworks suffer from significant performance degradation in low signal-to-noise ratio (SNR) regimes and exhibit limited adaptability to resource-constrained edge computing platforms. To address these limitations, we propose a novel Multimodal Tri-branch Fusion Network (MTF-Net) architecture that synergistically integrates time-frequency analysis with statistical feature learning. The framework systematically processes binarized time-frequency images (B-TFIs) and higher-order cumulant vectors through three collaboratively operating branches: (1) A primary temporal feature extractor employing dilated convolution-residual blocks (DCRBlocks) with hierarchical dilatation factors, incorporating channel attention mechanisms to dynamically emphasize discriminative temporal patterns; (2) Dual auxiliary branches based on Edge-Transformer modules (ETFormers), which achieve efficient spatial-structural learning through depthwise separable convolutions (DSC) while capturing long-range spectral dependencies via additive attention mechanisms with linear complexity; (3) A hierarchical fusion module implementing cross-branch feature recalibration through learnable parameter matrices. Extensive Monte Carlo experiments demonstrate that our MTF-Net significantly outperforms traditional methods in recognition accuracy for radar and communication signals under low SNR conditions, establishing a new benchmark for lightweight AMR solutions in ISAC systems.

在集成传感和通信（ISAC）系统普及的推动下，在动态电磁环境中准确识别未经授权的无人机（UAV）信号已成为频谱安全和认知无线电应用的关键挑战。传统的自动调制识别（AMR）框架在低信噪比（SNR）条件下存在显著的性能下降，并且对资源受限的边缘计算平台的适应性有限。为了解决这些限制，我们提出了一种新的多模态三分支融合网络（MTF-Net）架构，该架构将时频分析与统计特征学习协同集成。该框架通过三个协同操作分支系统地处理二值化时频图像（b - tfi）和高阶累积向量：(1)采用具有分层扩张因子的扩展卷积残差块（DCRBlocks）的初级时间特征提取器，结合通道注意机制，动态强调判别时间模式；(2)基于Edge-Transformer模块（ETFormers）的双辅助分支，通过深度可分离卷积（DSC）实现高效的空间结构学习，同时通过线性复杂性的加性注意机制捕获远程频谱依赖；(3)通过可学习参数矩阵实现跨分支特征再校准的分层融合模块。广泛的蒙特卡罗实验表明，我们的MTF-Net在低信噪比条件下对雷达和通信信号的识别精度显著优于传统方法，为ISAC系统中的轻量级AMR解决方案建立了新的基准。

{"title":"Electromagnetic signal recognition using multimodal tri-branch semantic fusion network in the UAV-assist integrated sensing and communication systems","authors":"Tiantian Wang , Nan Yan , Chaosan Yang , Zeliang An , Gongjing Zhang , Yuqing Xu","doi":"10.1016/j.dsp.2025.105820","DOIUrl":"10.1016/j.dsp.2025.105820","url":null,"abstract":"<div><div>Driven by the proliferation of integrated sensing and communication (ISAC) systems, the accurate recognition of unauthorized unmanned aerial vehicle (UAV) signals in dynamic electromagnetic environments has emerged as a critical challenge for spectrum security and cognitive radio applications. Conventional automatic modulation recognition (AMR) frameworks suffer from significant performance degradation in low signal-to-noise ratio (SNR) regimes and exhibit limited adaptability to resource-constrained edge computing platforms. To address these limitations, we propose a novel Multimodal Tri-branch Fusion Network (MTF-Net) architecture that synergistically integrates time-frequency analysis with statistical feature learning. The framework systematically processes binarized time-frequency images (B-TFIs) and higher-order cumulant vectors through three collaboratively operating branches: (1) A primary temporal feature extractor employing dilated convolution-residual blocks (DCRBlocks) with hierarchical dilatation factors, incorporating channel attention mechanisms to dynamically emphasize discriminative temporal patterns; (2) Dual auxiliary branches based on Edge-Transformer modules (ETFormers), which achieve efficient spatial-structural learning through depthwise separable convolutions (DSC) while capturing long-range spectral dependencies via additive attention mechanisms with linear complexity; (3) A hierarchical fusion module implementing cross-branch feature recalibration through learnable parameter matrices. Extensive Monte Carlo experiments demonstrate that our MTF-Net significantly outperforms traditional methods in recognition accuracy for radar and communication signals under low SNR conditions, establishing a new benchmark for lightweight AMR solutions in ISAC systems.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"171 ","pages":"Article 105820"},"PeriodicalIF":3.0,"publicationDate":"2025-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842756","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 covert scheme based on Rydberg atom reception under semantic communication system 语义通信系统下基于Rydberg原子接收的隐蔽方案

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

Digital Signal Processing

Pub Date : 2025-12-08 DOI: 10.1016/j.dsp.2025.105800

Qiang Yu, Ying Huang, Junjie Zeng, Jing Lei

As cryptographic technology develops, the risk of information security problems cannot be ignored. Traditional encryption methods expose the signal in a wireless channel, making it difficult to ensure long-term security. As a result, covert communication has become an important means of protecting information. Covert communication involves hiding the signal in the channel noise by reducing the power of the transmitted signal, ensuring that the eavesdropping party cannot detect the communication process. Since Bit Error Rate (BER) performance is closely related to the Signal-to-Noise Ratio (SNR) in traditional communication methods, achieving steganography poses a significant challenge, whereas the natural anti-distortion properties of semantic communication enable the recovery of semantic information in low SNR environments. Although reducing the signal power can achieve concealment, it is quite challenging to simultaneously detect and receive weak signals from the legitimate party. In this paper, we propose a covert scheme based on Rydberg Atom reception under semantic communication system, initially constructed by combining the Alice-Bob-Wille three-node model with the advantages of Rydberg atom system for low-power signal detection. Second, given the energy level conditions of the Rydberg atom system at Bob’s end, we plan a design criterion for the optimal transmit signal power at Alice’s end. Finally, under the condition of setting certain concealment threshold and semantic similarity threshold, we plan the control strategy for the optimal artificial noise power at the Alice’s end. The simulation results show that Alice’s power selection strategy under this scheme can achieve highly sensitive reception of weak signals, ensure the recovery of semantic information for both communicating parties, and conceal the communication process. Even if Willie also employs a Rydberg atom system for reception, achieving high-sensitivity demodulation of Alice’s transmitted signal remains challenging due to the difficulty in obtaining both the frequency of Alice’s signal and the four-level energy structure of Bob’s Rydberg atom system.

随着密码技术的发展，信息安全问题的风险不容忽视。传统的加密方法将信号暴露在无线信道中，难以保证长期的安全性。因此，秘密通信已成为信息保护的重要手段。隐蔽通信是通过降低传输信号的功率，将信号隐藏在信道噪声中，保证窃听方无法检测到通信过程。由于误码率（BER）性能与传统通信方法的信噪比（SNR）密切相关，因此实现隐写提出了重大挑战，而语义通信的天然抗失真特性使语义信息能够在低信噪比环境中恢复。虽然降低信号功率可以实现隐蔽，但同时检测和接收来自合法方的微弱信号是相当具有挑战性的。本文提出了一种语义通信系统下基于Rydberg原子接收的隐蔽方案，将Alice-Bob-Wille三节点模型与Rydberg原子系统的优点相结合，初步构建了用于低功耗信号检测的Rydberg原子接收方案。其次，考虑到Bob端Rydberg原子系统的能级条件，我们规划了Alice端最优发射信号功率的设计准则。最后，在设置一定的隐藏阈值和语义相似阈值的情况下，我们规划了Alice端最优人工噪声功率的控制策略。仿真结果表明，该方案下Alice的功率选择策略可以实现对微弱信号的高灵敏度接收，保证通信双方语义信息的恢复，并隐藏通信过程。即使Willie也采用Rydberg原子系统进行接收，由于很难同时获得Alice的信号频率和Bob的Rydberg原子系统的四能级能量结构，实现对Alice发射信号的高灵敏度解调仍然是一个挑战。

{"title":"A covert scheme based on Rydberg atom reception under semantic communication system","authors":"Qiang Yu, Ying Huang, Junjie Zeng, Jing Lei","doi":"10.1016/j.dsp.2025.105800","DOIUrl":"10.1016/j.dsp.2025.105800","url":null,"abstract":"<div><div>As cryptographic technology develops, the risk of information security problems cannot be ignored. Traditional encryption methods expose the signal in a wireless channel, making it difficult to ensure long-term security. As a result, covert communication has become an important means of protecting information. Covert communication involves hiding the signal in the channel noise by reducing the power of the transmitted signal, ensuring that the eavesdropping party cannot detect the communication process. Since Bit Error Rate (BER) performance is closely related to the Signal-to-Noise Ratio (SNR) in traditional communication methods, achieving steganography poses a significant challenge, whereas the natural anti-distortion properties of semantic communication enable the recovery of semantic information in low SNR environments. Although reducing the signal power can achieve concealment, it is quite challenging to simultaneously detect and receive weak signals from the legitimate party. In this paper, we propose a covert scheme based on Rydberg Atom reception under semantic communication system, initially constructed by combining the Alice-Bob-Wille three-node model with the advantages of Rydberg atom system for low-power signal detection. Second, given the energy level conditions of the Rydberg atom system at Bob’s end, we plan a design criterion for the optimal transmit signal power at Alice’s end. Finally, under the condition of setting certain concealment threshold and semantic similarity threshold, we plan the control strategy for the optimal artificial noise power at the Alice’s end. The simulation results show that Alice’s power selection strategy under this scheme can achieve highly sensitive reception of weak signals, ensure the recovery of semantic information for both communicating parties, and conceal the communication process. Even if Willie also employs a Rydberg atom system for reception, achieving high-sensitivity demodulation of Alice’s transmitted signal remains challenging due to the difficulty in obtaining both the frequency of Alice’s signal and the four-level energy structure of Bob’s Rydberg atom system.</div></div>","PeriodicalId":51011,"journal":{"name":"Digital Signal Processing","volume":"170 ","pages":"Article 105800"},"PeriodicalIF":3.0,"publicationDate":"2025-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790623","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