Neurocomputing最新文献

{"title":"SPLAC: A single-step PTZ camera linear auto-calibration method","authors":"Feng Yang ,&nbsp;Hua Chen ,&nbsp;Xiaogang Zhang ,&nbsp;Shi Wang ,&nbsp;Lei Zhang","doi":"10.1016/j.neucom.2025.130377","DOIUrl":"10.1016/j.neucom.2025.130377","url":null,"abstract":"<div><div>Pan-tilt-zoom (PTZ) cameras require auto-calibration techniques to estimate intrinsic parameters. Existing methods often involve multi-step processes or higher-order nonlinear constraints, which increase computational complexity and potentially lead to error accumulation. To address these challenges, we propose SPLAC, a single-step PTZ camera linear auto-calibration method based on an imaging model. SPLAC introduces a novel linear constraint derived from geometric relationships of line pairs in the perspective projection process. By solving a single linear set of equations, the method estimates camera parameters directly from images and pan–tilt feedback, thereby simplifying the calibration process. Additionally, to mitigate the impact of feature point matching errors, which can reduce calibration accuracy or cause failure, we introduce a filtering method based on the rotation characteristics of PTZ cameras to eliminate mismatched feature points. Experiments conducted on simulated data, warped panoramic images, and real PTZ camera images demonstrate the feasibility, accuracy, and robustness of our proposed method.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"645 ","pages":"Article 130377"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144116546","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Population-Based Meta-Heuristic Optimization Algorithm Booster: An Evolutionary and Learning Competition Scheme","authors":"Jun Wang ,&nbsp;Junyu Dong ,&nbsp;Huiyu Zhou ,&nbsp;Xinghui Dong","doi":"10.1016/j.neucom.2025.130405","DOIUrl":"10.1016/j.neucom.2025.130405","url":null,"abstract":"<div><div>In a Population-Based Meta-Heuristic Optimization Algorithm (PMOA), individuals in the population will constantly generate new promising individuals, to form new populations. Although the population continuously changes, the variations in each individual are traceable in most algorithms. An individual in the population comes from the individual in the previous population. The direction of the evolution of populations can be identified on top of this historical inheritance relationship, which improves the efficiency of PMOAs and solves optimization problems more effectively. Since Recurrent Neural Networks (RNNs) are able to capture the temporal dependencies in sequences, we are motivated to propose a novel but simple Evolutionary and Learning Competition Scheme (ELCS), also referred to as the PMOA Booster, in which individuals keep changing for the better fitness based on the heuristic rules of the PMOA while an RNN is used to learn the process that each individual changes in order to guide the generation of promising individuals. The ELCS automatically selects the RNN or PMOA which generates more individuals with the better fitness. We test the proposed scheme using the benchmark of IEEE Congress on Evolutionary Computation 2022 competition (CEC 2022). The results show that this scheme is able to boost the performance of both the classical and state-of-the-art PMOAs and outperforms its counterparts. Also, the ELCS produces promising results in two real-world industrial scenarios. We believe that the effectiveness of the proposed ELCS is due to the adaptive competition between the RNN and the PMOA.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"643 ","pages":"Article 130405"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144072701","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Spiking Trans-YOLO: A range-adaptive energy-efficient bridge between YOLO and Transformer","authors":"Yushi Huo,&nbsp;Hongwei Ge,&nbsp;Guozhi Tang,&nbsp;Shengxuan Gao,&nbsp;Jiale Xu","doi":"10.1016/j.neucom.2025.130407","DOIUrl":"10.1016/j.neucom.2025.130407","url":null,"abstract":"<div><div>The remarkable success of Transformers in Artificial Neural Networks (ANNs) has driven growing interest in leveraging self-attention mechanisms and Transformer-based architectures for Spiking Neural Network (SNN) object detection. However, existing methods combining Transformers with YOLO lack reasonable bridging strategies, leading to significantly increased computational costs and limitations in local feature extraction. To address these challenges, we propose Spiking Trans-YOLO, introducing a Top-Attention Hybrid Feature Fusion module, which exclusively applies self-attention to high-level spiking features that are more stable and semantically meaningful. This prevents redundant computations caused by unstable low-level spikes and reduces energy consumption. Subsequently, we perform cross-scale feature fusion to compensate for Transformers’ shortcomings in local feature extraction. This approach efficiently bridges YOLO and Transformer architectures while preserving the low-power characteristics of SNNs. Additionally, The newly proposed Integer Leaky Integrate-and-Fire (I-LIF) neuron has demonstrated significant potential in SNNs by enabling integer-valued training and spike-driven inference, thereby reducing quantization errors. However, existing spiking self-attention mechanisms fail to incorporate proper scaling factors for I-LIF neurons, which may lead to gradient vanishing. To address this, we propose a Range-Adaptive Spiking Attention for intra-scale interactions. By dynamically adjusting scaling coefficients, RASA mitigates gradient vanishing issues associated with integer training, allowing I-LIF neurons to exploit the benefits of spike-based self-attention fully. The proposed method achieves 67.7% mAP@50 on the COCO dataset and 68.6% mAP@50 on the Gen1 dataset, outperforming state-of-the-art YOLO architectures and achieving superior energy efficiency compared to advanced Transformer-based architectures. Code: <span><span>https://github.com/s1110/Spiking-Trans-YOLO</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"645 ","pages":"Article 130407"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144099705","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Neural network-based fully actuated trajectory tracking control for space manipulator with system uncertainties","authors":"Jing Xu ,&nbsp;Kai Zhang ,&nbsp;Zhaoke Ning ,&nbsp;Zhihua Chen","doi":"10.1016/j.neucom.2025.130399","DOIUrl":"10.1016/j.neucom.2025.130399","url":null,"abstract":"<div><div>Space manipulator is regarded as one of the technologies with the greatest potential in various on-orbit service missions. By leveraging the fully actuated system approach (FASA) and neural network (NN), this paper explores the trajectory tracking control problem of space manipulator with the dynamical model error and external disturbance. The designed controller consists of three parts: the FASA-based part and two radial basis function NN (RBFNN) auxiliary inputs, the FASA-based part provides the desired eigenstructure and the global asymptotic stability for the nominal error system. It solves the problem that the trajectory tracking effect of space manipulator is poor under the working conditions of model error and external disturbance. This paper adopts a combination of offline training and online learning to comprehensively deal with uncertainties. By leveraging particle swarm optimization (PSO) method, we utilize abundant offline data to train the first RBFNN so as to approximate the inherent model error, and adopt an online adaptive learning strategy to update the weight of the second RBFNN. Based on Lyapunov’s direct method, stability analysis validates that the tracking error and the estimation error of RBFNN weight would finally converge to a small neighborhoods about the origin. Finally, simulation results with a planar free-flying space manipulator show that the proposed controller demonstrates more rapid convergence and more accurate tracking precision in contrast to the classical adaptive RBFNN controller.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"647 ","pages":"Article 130399"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144190420","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Sentiment-Enhanced Triplet Region Detection Framework for aspect sentiment triplet extraction","authors":"Xiaoyong Liu ,&nbsp;Jiahao Chen ,&nbsp;Chunlin Xu ,&nbsp;Zhiguo Du ,&nbsp;Weiqi Chen ,&nbsp;Huihui Li","doi":"10.1016/j.neucom.2025.130392","DOIUrl":"10.1016/j.neucom.2025.130392","url":null,"abstract":"<div><div>Aspect Sentiment Triplet Extraction (ASTE) is a crucial task in sentiment analysis that focuses on identifying aspect terms, corresponding opinion terms and the sentiments expressed, ultimately forming sentiment triplets. Recent tagging-based methods where triplets are represented by a two-dimensional matrix of word-pair relations, have achieved promising performance on ASTE. However, existing tagging-based methods usually treat multi-word aspect or opinion terms as a set of independent words and solely focus on the interactions between words, causing the model to potentially extract only parts of the aspect terms or opinion terms, thus producing incorrect triplets. To alleviate this limitation, we propose Sentiment-Enhanced Triplet Region Detection Framework (SE-TRDF) for ASTE. Instead of classifying the relationship between each word pair directly, SE-TRDF conducts ASTE by detecting a triplet region enclosed by aspect and opinion words firstly and then classifying the sentiment polarity of the detected region. With the goal of identifying the four vertices of the triplet region, SE-TRDF is able to obtain the correct aspect and opinion terms with multi-word. Moreover, to further enhance the capability of our SE-TRDF, a sentiment detection module (SDM) is devised which aims to identify the aspect and opinion terms within a sentence and helps to filter some wrong triplet regions detected. Extensive experimental results on four public benchmarks demonstrate the superiority of the proposed SE-TRDF compared with baselines.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"644 ","pages":"Article 130392"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144099453","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Model-agnostic personalized adaptation for segment anything model","authors":"Juncheng Wang ,&nbsp;Lei Shang ,&nbsp;Wang Lu ,&nbsp;Xiangyang Ji ,&nbsp;Shujun Wang","doi":"10.1016/j.neucom.2025.130424","DOIUrl":"10.1016/j.neucom.2025.130424","url":null,"abstract":"<div><div>The Segment Anything Model (SAM) and its family of models have made significant strides in open-set, prompt-driven instance segmentation. However, some closed-source SAM family models often face ethical, copyright, or commercial restrictions, limiting their accessibility and further personalized adaptation. To overcome these limitations, we introduce MapSAM, a model-agnostic, personalized plugin for SAM family models. MapSAM features a lightweight threshold learner that enables nuanced post-hoc processing of confidence maps, leading to improved segmentation accuracy. By leveraging mask-focused learning, our approach determines pixel-wise and hardness-aware thresholds, allowing for more effective adaptation to diverse datasets. Furthermore, we critically examine the limitations of the commonly used Dice loss, which can overlook sample hardness when allocating penalties. We theoretically demonstrate that the Mean Squared Error (MSE) loss complements Dice loss by providing a stronger focus on sample hardness. Through extensive experiments on seven diverse datasets using multiple SAM family models, we validate the effectiveness of MapSAM in achieving superior segmentation results, particularly in challenging domains. Our findings open up new avenues for personalized, open-set instance segmentation across various application areas, leveraging any closed-source SAM family model. Code will be available at <span><span>https://github.com/wjc2830/MapSAM.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"645 ","pages":"Article 130424"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144107692","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Efficient incremental model reduction approach for time-varying spatially distributed processes","authors":"Ai Ling ,&nbsp;Hao Ru ,&nbsp;Xueqin Chen ,&nbsp;Kok Lay Teo","doi":"10.1016/j.neucom.2025.130390","DOIUrl":"10.1016/j.neucom.2025.130390","url":null,"abstract":"<div><div>The accurate and efficient modeling of complex spatially distributed processes (SDPs) represents a significant challenge for researchers in the field. A variety of modeling approaches have been developed with the objective of addressing issues related to spatiotemporal modeling. The conventional offline modeling approaches necessitate the availability of a spatiotemporal data set that encompasses the entirety of the system’s dynamic features. It is not possible to fully satisfy this prerequisite in the context of real-world applications, particularly in the case of time-varying SDPs. Furthermore, the process of updating the model by repeatedly applying the offline algorithm is inherently time-consuming. In this paper, we put forth an efficient incremental model reduction approach for time-varying SDPs. Following the initialization phase, the modified Gram–Schmidt orthogonalization method is employed to extract the feature subspace in a sequential manner. The time-space synthesis is then utilized to reconstruct the spatiotemporal dynamics. The efficacy of the model is evaluated on a representative diffusion-reaction process, and the comparative experimental outcomes demonstrate that the presented algorithm is an efficient and effective approach for the online learning of time-varying SDPs.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"643 ","pages":"Article 130390"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144071068","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Vision transformers on the edge: A comprehensive survey of model compression and acceleration strategies","authors":"Shaibal Saha,&nbsp;Lanyu Xu","doi":"10.1016/j.neucom.2025.130417","DOIUrl":"10.1016/j.neucom.2025.130417","url":null,"abstract":"<div><div>In recent years, vision transformers (ViTs) have emerged as powerful and promising techniques for computer vision tasks such as image classification, object detection, and segmentation. Unlike convolutional neural networks (CNNs), which rely on hierarchical feature extraction, ViTs treat images as sequences of patches and leverage self-attention mechanisms. However, their high computational complexity and memory demands pose significant challenges for deployment on resource-constrained edge devices. To address these limitations, extensive research has focused on model compression techniques and hardware-aware acceleration strategies. Nonetheless, a comprehensive review that systematically categorizes these techniques and their trade-offs in accuracy, efficiency, and hardware adaptability for edge deployment remains lacking. This survey bridges this gap by providing a structured analysis of model compression techniques, software tools for inference on edge, and hardware acceleration strategies for ViTs. We discuss their impact on accuracy, efficiency, and hardware adaptability, highlighting key challenges and emerging research directions to advance ViT deployment on edge platforms, including graphics processing units (GPUs), application-specific integrated circuit (ASICs), and field-programmable gate arrays (FPGAs). The goal is to inspire further research with a contemporary guide on optimizing ViTs for efficient deployment on edge devices.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"643 ","pages":"Article 130417"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144084545","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Framework for Neural Topic Modeling with Mutual Information and Group Regularization","authors":"Tung Nguyen ,&nbsp;Linh Ngo Van ,&nbsp;Anh Nguyen Duc ,&nbsp;Sang Dinh Viet","doi":"10.1016/j.neucom.2025.130420","DOIUrl":"10.1016/j.neucom.2025.130420","url":null,"abstract":"<div><div>Recent advances in topic modeling have leveraged pre-trained language models (PLMs) and refined topic-word associations to improve both topic discovery and document topic distributions. However, directly integrating PLMs often leads to higher inference costs, making them less suitable for low-latency applications. At the same time, effectively capturing inter-topic relationships remains a critical yet challenging task. In this paper, we propose NeuroMIG (<strong>Neu</strong>ral T<strong>o</strong>pic Modeling with <strong>M</strong>utual <strong>I</strong>nformation and <strong>G</strong>roup Topic Regularization), a novel framework that addresses both issues. NeuroMIG (1) maximizes mutual information between document topic distributions and PLM embeddings to efficiently incorporate PLM knowledge, and (2) employs Group Topic Regularization based on optimal transport to model interactions among topics. Compatible with a wide range of topic modeling architectures, NeuroMIG significantly improves performance over baselines while preserving efficient inference, as validated by experimental results.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"645 ","pages":"Article 130420"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144116652","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SqSFill : Joint spatial and spectral learning for high-fidelity image inpainting","authors":"Zihao Zhang ,&nbsp;Feifan Cai ,&nbsp;Qin Zhou ,&nbsp;Youdong Ding","doi":"10.1016/j.neucom.2025.130414","DOIUrl":"10.1016/j.neucom.2025.130414","url":null,"abstract":"<div><div>Image inpainting has made significant progress due to recent advances in deep learning. However, most generative inpainting networks face challenges such as producing blurry results that lack high-frequency details or introducing inconsistent structures. To address these issues, we propose a novel transformer-based approach, SqSFill, which exploits rich information in both spatial and spectral domains. Specifically, SqSFill incorporates Rectified Frequency Feature Extractor (RecFFE) in the early layers of the network to capture fine-grained details by leveraging frequency information, guided by frequency loss. Moreover, we design a Scout Attention Block with linear complexity to replace vanilla self-attention, thereby effectively capturing long-range dependencies with lower computational cost. By integrating the RecFFE and Scout Attention Block, SqSFill is able to generate both coherent structures and sharp textures. Extensive experiments demonstrate the proposed SqSFill achieves superior results, outperforming previous state-of-the-art approaches with fewer parameters.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"645 ","pages":"Article 130414"},"PeriodicalIF":5.5,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144099668","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}