Pub Date : 2026-02-06DOI: 10.1109/LSP.2026.3661843
Deijany Rodriguez Linares;Oksana Moryakova;Håkan Johansson
This letter presents a novel approach for efficiently computing time-index powered weighted sums of the form $sum _{n=0}^{N-1} n^{K} v[n]$ using cascaded accumulators. Traditional direct computation requires $K{times }N$ general multiplications, which become prohibitive for large $N$, while alternative strategies based on lookup tables or signal reversal require storing entire data blocks. By exploiting accumulator properties, the proposed method eliminates the need for such storage and reduces the multiplicative cost to only $K{+}1$ constant multiplications, enabling efficient real-time implementation. The approach is particularly useful when such sums need to be efficiently computed in sample-by-sample processing systems.
{"title":"Efficient Computation of Time-Index Powered Weighted Sums Using Cascaded Accumulators","authors":"Deijany Rodriguez Linares;Oksana Moryakova;Håkan Johansson","doi":"10.1109/LSP.2026.3661843","DOIUrl":"https://doi.org/10.1109/LSP.2026.3661843","url":null,"abstract":"This letter presents a novel approach for efficiently computing time-index powered weighted sums of the form <inline-formula><tex-math>$sum _{n=0}^{N-1} n^{K} v[n]$</tex-math></inline-formula> using cascaded accumulators. Traditional direct computation requires <inline-formula><tex-math>$K{times }N$</tex-math></inline-formula> general multiplications, which become prohibitive for large <inline-formula><tex-math>$N$</tex-math></inline-formula>, while alternative strategies based on lookup tables or signal reversal require storing entire data blocks. By exploiting accumulator properties, the proposed method eliminates the need for such storage and reduces the multiplicative cost to only <inline-formula><tex-math>$K{+}1$</tex-math></inline-formula> constant multiplications, enabling efficient real-time implementation. The approach is particularly useful when such sums need to be efficiently computed in sample-by-sample processing systems.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"893-897"},"PeriodicalIF":3.9,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146223653","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}
Pub Date : 2026-02-06DOI: 10.1109/LSP.2026.3662609
Changkai Cai;Wei Meng
In order to cope with the track-to-track association (T2TA) problem under large sensor bias, and dense objects, this letter proposes a T2TA algorithm based on two alternating triangle and translation local track feature descriptors (TFDs). The proposed TFDs are constructed by pseudo-correspondence, triangle area and angle, translation vector, and Euclidean distance, and are combined via a transition algorithm that enables them alternating operation, to ensure the real-time performance of the feature extraction process. Finally, we establish the T2TA algorithm through a combination of global track feature, linear assignment algorithm, thin plate spline function, and simulated annealing algorithm. Experiments demonstrate the significant advantages of our proposed TFDs and T2TA algorithm compared with state-of-the-art algorithms under large sensor bias, and dense objects.
{"title":"Robust Track-to-Track Association Algorithm for Large Sensor Bias and Dense Objects","authors":"Changkai Cai;Wei Meng","doi":"10.1109/LSP.2026.3662609","DOIUrl":"https://doi.org/10.1109/LSP.2026.3662609","url":null,"abstract":"In order to cope with the track-to-track association (T2TA) problem under large sensor bias, and dense objects, this letter proposes a T2TA algorithm based on two alternating triangle and translation local track feature descriptors (TFDs). The proposed TFDs are constructed by pseudo-correspondence, triangle area and angle, translation vector, and Euclidean distance, and are combined via a transition algorithm that enables them alternating operation, to ensure the real-time performance of the feature extraction process. Finally, we establish the T2TA algorithm through a combination of global track feature, linear assignment algorithm, thin plate spline function, and simulated annealing algorithm. Experiments demonstrate the significant advantages of our proposed TFDs and T2TA algorithm compared with state-of-the-art algorithms under large sensor bias, and dense objects.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"898-902"},"PeriodicalIF":3.9,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146223676","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}
Object encoding is essential for enabling robots to efficiently perform tasks such as recognition and autonomous exploration. Existing object encoding approaches typically rely on appearance-based representations, leading to poor performance in environments with multiple visually similar objects, which can compromise downstream task accuracy. Inspired by human perception of objects through both appearance and structure, we introduce DGOE, a novel object encoding method, which adopts a dual-graph embedding scheme. Rather than treating object representation as a single-level appearance encoding problem, DGOE explicitly models object discriminability through a dual-level structural formulation, decomposing it into intrinsic inner-object structure and cross-object relational context. DGOE leverages graph structures as the carrier to integrate object intrinsic appearance and cross-object spatial features. This unified framework leads to more distinctive and reliable object representations. Additionally, we use multi-head self-attention to learn cross-object graph structures for extracting cross-object spatial features. Experimental evaluations on multiple publicly available datasets demonstrate the superior performance of our method in object-level matching tasks, underscoring its effectiveness and robustness.
{"title":"Beyond Appearance: Dual-Graph Object Encoding With Learnable Graph Structure","authors":"Cuiyun Fang;Fan Wang;Xirun Cheng;Wen Zhang;Zhenyu Gao;Yingwei Xia;Guoqing Deng;Chaofan Zhang","doi":"10.1109/LSP.2026.3660573","DOIUrl":"https://doi.org/10.1109/LSP.2026.3660573","url":null,"abstract":"Object encoding is essential for enabling robots to efficiently perform tasks such as recognition and autonomous exploration. Existing object encoding approaches typically rely on appearance-based representations, leading to poor performance in environments with multiple visually similar objects, which can compromise downstream task accuracy. Inspired by human perception of objects through both appearance and structure, we introduce DGOE, a novel object encoding method, which adopts a dual-graph embedding scheme. Rather than treating object representation as a single-level appearance encoding problem, DGOE explicitly models object discriminability through a dual-level structural formulation, decomposing it into intrinsic inner-object structure and cross-object relational context. DGOE leverages graph structures as the carrier to integrate object intrinsic appearance and cross-object spatial features. This unified framework leads to more distinctive and reliable object representations. Additionally, we use multi-head self-attention to learn cross-object graph structures for extracting cross-object spatial features. Experimental evaluations on multiple publicly available datasets demonstrate the superior performance of our method in object-level matching tasks, underscoring its effectiveness and robustness.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"1057-1061"},"PeriodicalIF":3.9,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362411","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}
Pub Date : 2026-01-29DOI: 10.1109/LSP.2026.3659030
Zhen Li;Huafeng He;Yongquan You;Xin Zhang;LiYuan Wang
To address the poor performance of conventional sea clutter suppression methods when target echoes share the same frequency as sea clutter, this letter proposes a suppression approach based on feature transformation. The autocorrelation matrix of sea clutter is first employed to construct a feature transformation matrix, through which the received radar data are normalized in the feature space. A differential operation is then applied to cancel clutter components, and the residual energy is combined with a cell-averaging constant false alarm rate (CA-CFAR) detector to achieve target detection in co-frequency scenarios. The proposed method requires no additional prior information and provides a new solution for detecting low-speed maritime targets. Simulation results demonstrate that the method significantly improves both signal-to-clutter ratio (SCR) and detection probability under various sea states and SCR conditions, with particularly notable advantages when the target frequency is close to the sea clutter spectral peak.
{"title":"A Target Detection Method Based on Feature Transformation for Sea Clutter Suppression","authors":"Zhen Li;Huafeng He;Yongquan You;Xin Zhang;LiYuan Wang","doi":"10.1109/LSP.2026.3659030","DOIUrl":"https://doi.org/10.1109/LSP.2026.3659030","url":null,"abstract":"To address the poor performance of conventional sea clutter suppression methods when target echoes share the same frequency as sea clutter, this letter proposes a suppression approach based on feature transformation. The autocorrelation matrix of sea clutter is first employed to construct a feature transformation matrix, through which the received radar data are normalized in the feature space. A differential operation is then applied to cancel clutter components, and the residual energy is combined with a cell-averaging constant false alarm rate (CA-CFAR) detector to achieve target detection in co-frequency scenarios. The proposed method requires no additional prior information and provides a new solution for detecting low-speed maritime targets. Simulation results demonstrate that the method significantly improves both signal-to-clutter ratio (SCR) and detection probability under various sea states and SCR conditions, with particularly notable advantages when the target frequency is close to the sea clutter spectral peak.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"1052-1056"},"PeriodicalIF":3.9,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362370","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}
RGB-T object detection demonstrates strong potential by leveraging the complementary strengths of visible (RGB) and thermal (T) modalities for applications in intelligent surveillance, autonomous driving, and search-and-rescue. However, most existing approaches rely on aligning RGB and thermal features followed by global feature fusion, which often fail to capture fine-grained spatial misalignments and generate reliable fusion weights. These limitations become particularly pronounced in tiny object detection scenarios, where feature representations are inherently sparse and highly sensitive to cross-modal spatial misalignment. To address these challenges, we propose the Modality-Aware Dynamic Fusion detection (MDFDet) network, an end-to-end framework that leverages Modality-Aware Queries to adaptively fuse visible and thermal features for tiny object detection in weakly aligned RGB-T images. At its core, the Modality-Decoupled Dynamic Fusion Decoder employs dual self-attention mechanisms coupled with cross-modal deformable attention, enabling alignment-free feature aggregation at the region level through dynamically decoupled queries. These queries originate from our Modality-Aware Paired Query Selection module, which adaptively balances multimodal contributions to select semantically rich, decoupled queries. Additionally, we introduce a Tiny Object Detection Layer that preserves critical low-level spatial details, addressing the feature degradation commonly observed in deep networks. Extensive experiments on the RGBT-Tiny benchmark demonstrate that MDFDet achieves state-of-the-art performance, outperforming existing methods by more than 3% in $text{AP}_{50}^{s}$. These results highlight the effectiveness of modality-aware dynamic decoupled fusion and its potential for real-world RGB-T tiny object detection tasks.
{"title":"Modality-Aware Dynamic Fusion for Weakly Aligned RGB-T Tiny Object Detection","authors":"Yuting Xie;Zhili Zhang;Yi Hou;Puzuo Wang;Hanxiao Zhang","doi":"10.1109/LSP.2026.3659814","DOIUrl":"https://doi.org/10.1109/LSP.2026.3659814","url":null,"abstract":"RGB-T object detection demonstrates strong potential by leveraging the complementary strengths of visible (RGB) and thermal (T) modalities for applications in intelligent surveillance, autonomous driving, and search-and-rescue. However, most existing approaches rely on aligning RGB and thermal features followed by global feature fusion, which often fail to capture fine-grained spatial misalignments and generate reliable fusion weights. These limitations become particularly pronounced in tiny object detection scenarios, where feature representations are inherently sparse and highly sensitive to cross-modal spatial misalignment. To address these challenges, we propose the Modality-Aware Dynamic Fusion detection (MDFDet) network, an end-to-end framework that leverages Modality-Aware Queries to adaptively fuse visible and thermal features for tiny object detection in weakly aligned RGB-T images. At its core, the Modality-Decoupled Dynamic Fusion Decoder employs dual self-attention mechanisms coupled with cross-modal deformable attention, enabling alignment-free feature aggregation at the region level through dynamically decoupled queries. These queries originate from our Modality-Aware Paired Query Selection module, which adaptively balances multimodal contributions to select semantically rich, decoupled queries. Additionally, we introduce a Tiny Object Detection Layer that preserves critical low-level spatial details, addressing the feature degradation commonly observed in deep networks. Extensive experiments on the RGBT-Tiny benchmark demonstrate that MDFDet achieves state-of-the-art performance, outperforming existing methods by more than 3% in <inline-formula><tex-math>$text{AP}_{50}^{s}$</tex-math></inline-formula>. These results highlight the effectiveness of modality-aware dynamic decoupled fusion and its potential for real-world RGB-T tiny object detection tasks.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"1023-1027"},"PeriodicalIF":3.9,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362286","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}
Pub Date : 2026-01-21DOI: 10.1109/LSP.2026.3656465
Li Cho;Shi-Wei Chang
The Papoulis–Gerchberg iterative reconstruction (PGIR) algorithm has been widely applied across diverse signal processing tasks, and reliable performance prediction during the system design stage is crucial for ensuring its effectiveness. However, predicting PGIR performance for arbitrary signal lengths and observation patterns has long been computationally intractable due to the combinatorial explosion of possible configurations. This letter addresses the problem by analyzing convergence conditions and modeling reconstruction error distributions in both noise-free and noisy scenarios. The derived closed-form probability laws enable accurate prediction for individual geometries, and the observed concentration of the operator's spectral radius with increasing signal length further allows performance characterization based only on loss and knowledge ratios. Tresulting probabilistic framework thus provides the first scalable tool for predicting PGIR performance, validated through case studies in multicarrier communication systems.
{"title":"On the Performance of Discrete Papoulis–Gerchberg Type Iterative Reconstruction","authors":"Li Cho;Shi-Wei Chang","doi":"10.1109/LSP.2026.3656465","DOIUrl":"https://doi.org/10.1109/LSP.2026.3656465","url":null,"abstract":"The Papoulis–Gerchberg iterative reconstruction (PGIR) algorithm has been widely applied across diverse signal processing tasks, and reliable performance prediction during the system design stage is crucial for ensuring its effectiveness. However, predicting PGIR performance for arbitrary signal lengths and observation patterns has long been computationally intractable due to the combinatorial explosion of possible configurations. This letter addresses the problem by analyzing convergence conditions and modeling reconstruction error distributions in both noise-free and noisy scenarios. The derived closed-form probability laws enable accurate prediction for individual geometries, and the observed concentration of the operator's spectral radius with increasing signal length further allows performance characterization based only on loss and knowledge ratios. Tresulting probabilistic framework thus provides the first scalable tool for predicting PGIR performance, validated through case studies in multicarrier communication systems.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"718-722"},"PeriodicalIF":3.9,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082161","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}
Pub Date : 2026-01-21DOI: 10.1109/LSP.2026.3656752
Xi Yu;Jiahao Zhang;Jundi Wang;Hao Wu;Wantian Wang;Jin Meng
This study proposes an adaptive reward function to improve the convergence speed and adaptability of radar intelligent anti-jamming models.The design is based on two key factors: interference suppression effectiveness and target integrity after suppression. The primary reward is the improvement in signal-to-interference-plus-noise ratio (SINR), a standard metric for anti-jamming performance. To better distinguish between strategies, three performance indicators—interference suppression ratio (ISR), target amplitude fidelity (TAF), and target detection integrity (TDI)—are used as threshold constraints. An adaptive threshold mechanism reduces outlier rewards, accelerating convergence and improving flexibility and robustness across interference environments. Experiments show that the proposed method converges faster than existing approaches: the simulation results show a 40%–45% reduction in convergence time, and the anechoic chamber tests show a 60% reduction. The method also performs well under main lobe, side lobe, and suppression interference scenarios.
{"title":"An Adaptive Threshold Reward Function for Radar Anti-Jamming Decision-Making","authors":"Xi Yu;Jiahao Zhang;Jundi Wang;Hao Wu;Wantian Wang;Jin Meng","doi":"10.1109/LSP.2026.3656752","DOIUrl":"https://doi.org/10.1109/LSP.2026.3656752","url":null,"abstract":"This study proposes an adaptive reward function to improve the convergence speed and adaptability of radar intelligent anti-jamming models.The design is based on two key factors: interference suppression effectiveness and target integrity after suppression. The primary reward is the improvement in signal-to-interference-plus-noise ratio (SINR), a standard metric for anti-jamming performance. To better distinguish between strategies, three performance indicators—interference suppression ratio (ISR), target amplitude fidelity (TAF), and target detection integrity (TDI)—are used as threshold constraints. An adaptive threshold mechanism reduces outlier rewards, accelerating convergence and improving flexibility and robustness across interference environments. Experiments show that the proposed method converges faster than existing approaches: the simulation results show a 40%–45% reduction in convergence time, and the anechoic chamber tests show a 60% reduction. The method also performs well under main lobe, side lobe, and suppression interference scenarios.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"883-887"},"PeriodicalIF":3.9,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146223545","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}
Pub Date : 2026-01-20DOI: 10.1109/LSP.2026.3656054
Minjian Chen;Liquan Shen;Qi Teng;Shiwei Wang;Feifeng Wang
Compressed point clouds are increasingly used in machine vision tasks, which rely on key semantic regions of the point cloud such as geometric details and structural boundaries. However, existing point cloud compression methods for machine vision lack explicit awareness of geometrically induced semantic boundaries, causing semantic ambiguity in certain boundary regions during compression, thereby degrading machine vision performance. To address this issue, we propose a Dual-domain Boundary Aware Point Cloud Geometry Compression (DBA-PCGC) method that explicitly preserves semantic geometric boundaries from complementary spatial and frequency perspectives, enabling beneficial for machine vision tasks. Specifically, a Structure Aware Transform Module (SATM) exploits Gram matrix traces on local graphs to capture structural variations and highlight high-variation boundary regions, while compactly encoding smooth areas. In parallel, a Frequency Aware Transform Module (FATM) applies Chebyshev high-pass filtering to enhance high-frequency components corresponding to semantic geometric boundaries and suppress redundant low-frequency content. Experimental results on point cloud machine vision tasks demonstrate that our method achieves superior performance compared with existing compression approaches.
{"title":"DBA-PCGC: Dual-Domain Boundary Aware for Task-Friendly Point Cloud Geometry Compression","authors":"Minjian Chen;Liquan Shen;Qi Teng;Shiwei Wang;Feifeng Wang","doi":"10.1109/LSP.2026.3656054","DOIUrl":"https://doi.org/10.1109/LSP.2026.3656054","url":null,"abstract":"Compressed point clouds are increasingly used in machine vision tasks, which rely on key semantic regions of the point cloud such as geometric details and structural boundaries. However, existing point cloud compression methods for machine vision lack explicit awareness of geometrically induced semantic boundaries, causing semantic ambiguity in certain boundary regions during compression, thereby degrading machine vision performance. To address this issue, we propose a Dual-domain Boundary Aware Point Cloud Geometry Compression (DBA-PCGC) method that explicitly preserves semantic geometric boundaries from complementary spatial and frequency perspectives, enabling beneficial for machine vision tasks. Specifically, a Structure Aware Transform Module (SATM) exploits Gram matrix traces on local graphs to capture structural variations and highlight high-variation boundary regions, while compactly encoding smooth areas. In parallel, a Frequency Aware Transform Module (FATM) applies Chebyshev high-pass filtering to enhance high-frequency components corresponding to semantic geometric boundaries and suppress redundant low-frequency content. Experimental results on point cloud machine vision tasks demonstrate that our method achieves superior performance compared with existing compression approaches.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"713-717"},"PeriodicalIF":3.9,"publicationDate":"2026-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082109","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}
Conventional clustering algorithms, such as $k$-means and its variants, often assume that data are linearly separable and that all samples contribute equally to the clustering process. However, real-world data usually lies on nonlinear manifolds and contains noisy or ambiguous samples, making such assumptions unrealistic. To address these challenges, we incorporate a Sample-Weighting mechanism into the Kernel Clustering model, which is based on the strategy of coupling Prototype Distance with Local Manifold information together (PDLM-SWKC). Specifically, PDLM-SWKC performs clustering in kernel space to capture nonlinear structures, while adaptively assigning sample weights according to both their proximity to cluster centers and their local manifold connectivity; Besides, the learned sample weights in turn guide graph affinity matrix learning to generate better topological relation matrix, achieving tight coupling between sample-weighted kernel clustering and topological manifold learning. This dual-driven weighting mechanism enhances the robustness and structural consistency, effectively emphasizing reliable samples and suppressing outliers. Extensive experiments on eight benchmark datasets demonstrate that PDLM-SWKC achieves superior performance compared with state-of-the-art clustering methods. Moreover, convergence and visualization analyses confirm its stability, interpretability, and strong capability in modeling complex nonlinear data distributions.
{"title":"Prototype Distance and Local Manifold Guided Sample-Weighted Kernel Clustering","authors":"Chang Wu;Pengxin Xu;Zhaohu Liu;Luyun Wang;Yong Peng","doi":"10.1109/LSP.2026.3655339","DOIUrl":"https://doi.org/10.1109/LSP.2026.3655339","url":null,"abstract":"Conventional clustering algorithms, such as <inline-formula><tex-math>$k$</tex-math></inline-formula>-means and its variants, often assume that data are linearly separable and that all samples contribute equally to the clustering process. However, real-world data usually lies on nonlinear manifolds and contains noisy or ambiguous samples, making such assumptions unrealistic. To address these challenges, we incorporate a Sample-Weighting mechanism into the Kernel Clustering model, which is based on the strategy of coupling Prototype Distance with Local Manifold information together (PDLM-SWKC). Specifically, PDLM-SWKC performs clustering in kernel space to capture nonlinear structures, while adaptively assigning sample weights according to both their proximity to cluster centers and their local manifold connectivity; Besides, the learned sample weights in turn guide graph affinity matrix learning to generate better topological relation matrix, achieving tight coupling between sample-weighted kernel clustering and topological manifold learning. This dual-driven weighting mechanism enhances the robustness and structural consistency, effectively emphasizing reliable samples and suppressing outliers. Extensive experiments on eight benchmark datasets demonstrate that PDLM-SWKC achieves superior performance compared with state-of-the-art clustering methods. Moreover, convergence and visualization analyses confirm its stability, interpretability, and strong capability in modeling complex nonlinear data distributions.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"658-662"},"PeriodicalIF":3.9,"publicationDate":"2026-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082086","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}
Pub Date : 2026-01-19DOI: 10.1109/LSP.2026.3655611
Shaoping Xu;Hanyang Hu;Wuyong Tao
Unsupervised methods like deep image prior (DIP) leverage network priors for denoising without labeled data but suffer from slow convergence and overfitting, while deep random projector (DRP) improves efficiency via fixed weights and a random seed yet remains limited by its fully random initialization. In this work, we propose deep fixed projector (DFP), an enhanced DRP-based framework featuring three synergistic improvements: (1) initializing the seed with the noisy image to align optimization with the clean image manifold, (2) using pre-trained clean-to-clean encoder-decoder weights to embed structural priors and accelerate convergence, and (3) introducing inter-inference consistency (IIC), a self-supervised regularization that enforces output stability under input perturbations to suppress noise and reduce overfitting. Experiments show DFP consistently surpasses DIP, DRP, and recent variants in PSNR, while achieving faster convergence and robust denoising quality.
{"title":"Deep Fixed Projector: Fast Projection Network for Image Denoising via Frozen Weights and Inter-Inference Consistency","authors":"Shaoping Xu;Hanyang Hu;Wuyong Tao","doi":"10.1109/LSP.2026.3655611","DOIUrl":"https://doi.org/10.1109/LSP.2026.3655611","url":null,"abstract":"Unsupervised methods like deep image prior (DIP) leverage network priors for denoising without labeled data but suffer from slow convergence and overfitting, while deep random projector (DRP) improves efficiency via fixed weights and a random seed yet remains limited by its fully random initialization. In this work, we propose deep fixed projector (DFP), an enhanced DRP-based framework featuring three synergistic improvements: (1) initializing the seed with the noisy image to align optimization with the clean image manifold, (2) using pre-trained clean-to-clean encoder-decoder weights to embed structural priors and accelerate convergence, and (3) introducing inter-inference consistency (IIC), a self-supervised regularization that enforces output stability under input perturbations to suppress noise and reduce overfitting. Experiments show DFP consistently surpasses DIP, DRP, and recent variants in PSNR, while achieving faster convergence and robust denoising quality.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"708-712"},"PeriodicalIF":3.9,"publicationDate":"2026-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082042","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：