Information Fusion最新文献

{"title":"Do multimodal large language models understand welding?","authors":"Grigorii Khvatskii ,&nbsp;Yong Suk Lee ,&nbsp;Corey Angst ,&nbsp;Maria Gibbs ,&nbsp;Robert Landers ,&nbsp;Nitesh V. Chawla","doi":"10.1016/j.inffus.2025.103121","DOIUrl":"10.1016/j.inffus.2025.103121","url":null,"abstract":"<div><div>This paper examines the performance of Multimodal LLMs (MLLMs) in skilled production work, with a focus on welding. Using a novel data set of real-world and online weld images, annotated by a domain expert, we evaluate the performance of two state-of-the-art MLLMs in assessing weld acceptability across three contexts: RV &amp; Marine, Aeronautical, and Farming. While both models perform better on online images, likely due to prior exposure or memorization, they also perform relatively well on unseen, real-world weld images. Additionally, we introduce WeldPrompt, a prompting strategy that combines Chain-of-Thought generation with in-context learning to mitigate hallucinations and improve reasoning. WeldPrompt improves model recall in certain contexts but exhibits inconsistent performance across others. These results underscore the limitations and potentials of MLLMs in high-stakes technical domains and highlight the importance of fine-tuning, domain-specific data, and more sophisticated prompting strategies to improve model reliability. The study opens avenues for further research into multimodal learning in industry applications.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"120 ","pages":"Article 103121"},"PeriodicalIF":14.7,"publicationDate":"2025-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143714554","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"GenRAN: GenFusion-guided Reversible Anonymization Network for face privacy preserving","authors":"Ruilin Wang ,&nbsp;Lingchen Gu ,&nbsp;Jing Li ,&nbsp;Jun Wang ,&nbsp;Jiande Sun ,&nbsp;Wenbo Wan","doi":"10.1016/j.inffus.2025.103120","DOIUrl":"10.1016/j.inffus.2025.103120","url":null,"abstract":"<div><div>The rapid advancement of social networks has made it possible to obtain personal face images without permission. While recent advances in face privacy preservation focus on anonymizing facial features, their effectiveness is limited by challenges in achieving high fidelity for both anonymized and recovered faces in practical scenarios. To address these challenges, we introduce GenFusion, which incorporates Virtual Face Generation (VFG) into the Bi-branch Fusion process with coupling reversibility. Accordingly, we propose a GenFusion-based Reversible Anonymization Network (GenRAN) for enhanced face privacy preservation. Our approach integrates a Multi-Fusion (MF) module, enabling an anonymization encoder to create natural and realistic anonymized faces by fusing original images with virtual faces generated through the VFG module. Furthermore, high-fidelity recovery of the original face from the anonymized version is achieved via an anonymization decoder, which employs a Multi-Recovery module that shares unified parameters with the MF module. Additionally, we introduce a Feature Mixing guided Dense Block to adaptively blend original facial details into the anonymized images, maintain high realism across varying image types. Extensive experiments show that proposed GenRAN can achieve better performance on actual privacy preserving scenarios, while obtaining high perceptual fidelity of anonymized and recovered faces than SOTA methods.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103120"},"PeriodicalIF":14.7,"publicationDate":"2025-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143715394","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"RGB-D Domain adaptive semantic segmentation with cross-modality feature recalibration","authors":"Qizhe Fan ,&nbsp;Xiaoqin Shen ,&nbsp;Shihui Ying ,&nbsp;Juan Wang ,&nbsp;Shaoyi Du","doi":"10.1016/j.inffus.2025.103117","DOIUrl":"10.1016/j.inffus.2025.103117","url":null,"abstract":"<div><div>Unsupervised domain adaptive (UDA) semantic segmentation aims to train models that effectively transfer knowledge from synthetic to real-world images, thereby reducing the reliance on manual annotation. Currently, most existing UDA methods primarily focus on RGB image processing, largely overlooking depth information as a valuable geometric cue that complements RGB representations. Additionally, while some approaches attempt to incorporate depth information by inferring it from RGB images as an auxiliary task, inaccuracies in depth estimation can still result in localized blurring or distortion in segmentation outcomes. To comprehensively address these limitations, we propose an innovative RGB-D UDA framework CMFRDA, which seamlessly integrates both RGB and depth images as inputs, fully leveraging their distinct yet complementary properties to improve segmentation performance. Specifically, to mitigate the prevalent object boundary noise in depth information, we propose a Depth Feature Rectification Module (DFRM), which effectively suppresses noise while enhancing the representation of fine structural details. Nevertheless, despite the effectiveness of DFRM, challenges remain due to the presence of noisy signals arising from incomplete surface data beyond the operational range of depth sensors, as well as potential mismatches between modalities. In order to overcome these challenges, we further introduce a Cross-Modality Feature Recalibration (CMFR) block. CMFR comprises two key components: Channel-wise Consistency Recalibration (CCR) and Spatial-wise Consistency Recalibration (SCR). CCR suppresses noise from incomplete surfaces in depth by leveraging the complementary information provided by RGB features, while SCR exploits the distinctive advantages of both modalities to mutually recalibrate each other, thereby ensuring consistency between RGB and depth modalities. By seamlessly integrating DFRM and CMFR, our CMFRDA framework effectively improves the performance of UDA semantic segmentation. Multitudinous experiments demonstrate that our CMFRDA achieves competitive performance on two widely-used UDA benchmarks GTA <span><math><mo>→</mo></math></span> Cityscapes and Synthia <span><math><mo>→</mo></math></span> Cityscapes.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"120 ","pages":"Article 103117"},"PeriodicalIF":14.7,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143714553","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Social network group decision making: Characterization, taxonomy, challenges and future directions from an AI and LLMs perspective","authors":"Mingshuo Cao ,&nbsp;Tiantian Gai ,&nbsp;Jian Wu ,&nbsp;Francisco Chiclana ,&nbsp;Zhen Zhang ,&nbsp;Yucheng Dong ,&nbsp;Enrique Herrera-Viedma ,&nbsp;Francisco Herrera","doi":"10.1016/j.inffus.2025.103107","DOIUrl":"10.1016/j.inffus.2025.103107","url":null,"abstract":"<div><div>In the past decade, social network group decision making (SNGDM) has experienced significant advancements. This breakthrough is largely attributed to the rise of social networks, which provides crucial data support for SNGDM. As a result, it has emerged as a rapidly developing research field within decision sciences, attracting extensive attention and research over the past ten years. SNGDM events involve complex decision making processes with multiple interconnected stakeholders, where the evaluation of alternatives is influenced by network relationships. Since this research has evolved from group decision making (GDM) scenarios, there is currently no clear definition for SNGDM problems. This article aims to address this gap by first providing a clear definition of the SNGDM framework. It describes basic procedures, advantages, and challenges, serving as a foundational portrait of the SNGDM framework. Furthermore, this article offers a macro description of the literature on SNGDM over the past decade based on bibliometric analysis. Solving SNGDM problems effectively is challenging and requires careful consideration of the impact of social networks among decision-makers and the facilitation of consensus between different participants. Therefore, we propose a classification and overview of key elements for SNGDM models based on the existing literature: trust models, internal structure, and consensus mechanism for SNGDM. This article identifies the research challenges in SNGDM and points out the future research directions from two dimensions: first, the key SNGDM methodologies and second, the opportunities from artificial intelligence technology, in particular, combining large language models and multimodal fusion technologies. This look will be analyzed from a double perspective, both from the decision problem and from the technology views.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"120 ","pages":"Article 103107"},"PeriodicalIF":14.7,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143678376","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Wearable gait analysis of Cervical Spondylotic Myelopathy patients by fusing bipedal inertial information","authors":"Xin Shi ,&nbsp;Zhelong Wang ,&nbsp;Sen Qiu ,&nbsp;Fang Lin ,&nbsp;Ruichen Liu ,&nbsp;Kai Tang ,&nbsp;Pengrong Hou ,&nbsp;Qinghao Chu ,&nbsp;Yongtao Chen","doi":"10.1016/j.inffus.2025.103115","DOIUrl":"10.1016/j.inffus.2025.103115","url":null,"abstract":"<div><div>Cervical Spondylotic Myelopathy (CSM) is a degenerative disorder caused by cervical spinal cord compression, resulting in neurological impairment that disrupts motor function and leads to gait disturbances. Refined gait analysis through parameter quantification and multi-level feature fusion is essential for advancing precision medicine and rehabilitation research for CSM. This paper explores portable gait analysis for CSM patients using inertial sensors to enable multi-level analysis via bipedal information fusion. An adaptive threshold-based gait phase segmentation method is proposed, allowing zero-velocity update-aided spatial parameter calculation, with results consistent across optical laboratory and practical test. Using high-precision spatiotemporal parameters and movement intensity features, we further analyzed gait rhythm, stability, and symmetry, introducing an improved EWT-based indicator for rhythm and symmetry. Finally, statistical analysis and machine learning-based feature ranking of CSM gait characteristics were performed, accompanied by a detailed discussion on feature types. The results underscore the critical role of fused features in capturing CSM gait patterns, offering a valuable reference for comprehensive gait analysis for CSM patients.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"120 ","pages":"Article 103115"},"PeriodicalIF":14.7,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143714373","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Cloud storage-based secure big data analytics mechanism for drone-assisted healthcare 5.0 data fusion system","authors":"Mohammad Wazid ,&nbsp;Jaskaran Singh ,&nbsp;Ashok Kumar Das ,&nbsp;Sahil Garg ,&nbsp;Willy Susilo ,&nbsp;Mohammad Mehedi Hassan","doi":"10.1016/j.inffus.2025.103085","DOIUrl":"10.1016/j.inffus.2025.103085","url":null,"abstract":"<div><div>Drone-assisted healthcare 5.0 data fusion system is the amalgamation of some cutting-edge technologies, like the Internet of Things (IoT), blockchain, robotics, drones or Internet of Drones (IoD), big data, artificial intelligence (AI), and cloud computing. It offers a range of various medical services, like healthcare decision making, patient remote monitoring and tracking, operating virtual clinics, patient shifting through smart ambulances, ambient supported living, self-illness management, reminders of treatment, compliance, and adherence. Since the devices communicate through the Internet, the potential attackers have some chances to launch different attacks on the drone-assisted healthcare 5.0 data fusion system. Under these circumstances, the healthcare data of the system may be revealed to unauthorized parties, or it may be updated in an unauthorized way. Therefore, to mitigate these issues, a cloud storage-based secure big data analytics mechanism for drone-assisted healthcare 5.0 data fusion system is proposed in the paper (in short, CSDM-DHF). Due to the deployed security mechanism, the proposed CSDM-DHF is capable enough to provide security to multi-sensor data fusion process. The provided security analysis proves the robustness of CSDM-DHF against various potential attacks. Moreover, the formal security verification of CSDM-DHF is also provided through the well-known scyther software validation tool. A comparative analysis of the existing schemes shows that the proposed CSDM-DHF outperforms other schemes. Finally, a practical demonstration of CSDM-DHF is provided, and essential performance parameters are computed and analyzed.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103085"},"PeriodicalIF":14.7,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143746895","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CPIFuse: Toward realistic color and enhanced textures in color polarization image fusion","authors":"Yidong Luo,&nbsp;Junchao Zhang,&nbsp;Chenggong Li","doi":"10.1016/j.inffus.2025.103111","DOIUrl":"10.1016/j.inffus.2025.103111","url":null,"abstract":"<div><div>Conventional image fusion aims to integrate multiple sets of source images into one with more details, representing the merging of intensity information. In contrast, polarization image fusion seeks to enhance texture of the intensity image <span><math><msub><mrow><mi>S</mi></mrow><mrow><mn>0</mn></mrow></msub></math></span> in the corresponding spectral bands by integrating strong texture features reflected by DoLP (Degree of Linear Polarization) images, representing the combination of intensity and polarization, which are both physical properties of light. However, the 3-Dimensional information contained in DoLP is presented in a highlighted form within the 2-Dimensional image, and fusing it directly can result in spectrum discontinuities and obscuring necessary details of the fused image. Existing polarization image fusion methods do not analyze this phenomenon and fail to examine the physical information represented by DoLP images. Instead, they simply integrate this interference information in the same manner as fusing infrared images, leading to fused images that suffer from information loss and significant color discrepancies. In this paper, we propose a new color polarization image fusion strategy that takes into account the physical properties reflected in the <span><math><msub><mrow><mi>S</mi></mrow><mrow><mn>0</mn></mrow></msub></math></span> and DoLP images, namely CPIFuse. CPIFuse designs a customized loss function to optimize parameters through a lightweight Transformer-based image fusion framework, and color polarization image fusion has been achieved with color fidelity, enhanced texture and high efficiency. These advantages can be demonstrated in the visual effects, quantitative metrics, and car detection tasks of our comparative experiments. Furthermore, a new polarization dataset is constructed by the mechanism of division of focal plane polarimeter camera, which addresses the scarcity of datasets in the field of polarization image fusion. The source code and CPIF-dataset will be available at <span><span>https://github.com/roydon-luo/CPIFuse</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"120 ","pages":"Article 103111"},"PeriodicalIF":14.7,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143678379","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Object-Level and Scene-Level Feature Aggregation with CLIP for scene recognition","authors":"Qun Wang ,&nbsp;Feng Zhu ,&nbsp;Ge Wu ,&nbsp;Pengfei Zhao ,&nbsp;Jianyu Wang ,&nbsp;Xiang Li","doi":"10.1016/j.inffus.2025.103118","DOIUrl":"10.1016/j.inffus.2025.103118","url":null,"abstract":"<div><div>Scene recognition is a fundamental task in computer vision, pivotal for applications like visual navigation and robotics. However, traditional methods struggle to effectively capture and aggregate scene-related features due to the inherent complexity and diversity of scenes, often leading to sub-optimal performance. To address this limitation, we propose a novel method, named OSFA (<strong>O</strong>bject-level and <strong>S</strong>cene-level <strong>F</strong>eature <strong>A</strong>ggregation), that leverages CLIP’s multimodal strengths to enhance scene feature representation through a two-stage aggregation strategy: Object-Level Feature Aggregation (OLFA) and Scene-Level Feature Aggregation (SLFA). In OLFA, we first generate an initial scene feature by integrating the average-pooled feature map of the base visual encoder and the CLIP visual feature. The initial scene feature is then used as a query in object-level cross-attention to extract object-level details most relevant to the scene from the feature map, thereby enhancing the representation. In SLFA, we first use CLIP’s textual encoder to provide category-level textual features for the scene, guiding the aggregation of corresponding visual features from the feature map. OLFA’s enhanced scene feature then queries these category-aware features using scene-level cross-attention to further capture scene-level information and obtain the final scene representation. To strengthen training, we employ a multi-loss strategy inspired by contrastive learning, improving feature robustness and discriminative ability. We evaluate OSFA on three challenging datasets (i.e. Places365, MIT67, and SUN397), achieving substantial improvements in classification accuracy. These results highlight the effectiveness of our method in enhancing scene feature representation through CLIP-guided aggregation. This advancement significantly improves scene recognition performance. Our code is public at <span><span>https://github.com/WangqunQAQ/OSFA</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"120 ","pages":"Article 103118"},"PeriodicalIF":14.7,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143678377","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Insufficient task description can impair in-context learning: A study from information perspective","authors":"Meidai Xuanyuan ,&nbsp;Tao Yang ,&nbsp;Jingwen Fu ,&nbsp;Sicheng Zhao ,&nbsp;Yuwang Wang","doi":"10.1016/j.inffus.2025.103116","DOIUrl":"10.1016/j.inffus.2025.103116","url":null,"abstract":"<div><div>In-context learning, an essential technique in transformer-based models, relies on two main sources of information: in-context examples and task descriptions. While extensive research has focused on the influence of in-context examples, the role of task descriptions remains underexplored, despite its practical significance. This paper investigates how task descriptions impact the in-context learning performance of transformers and how these two sources of information can be effectively fused. We design a synthetic experimental framework to control the information provided in task descriptions and conduct a series of experiments where task description details are systematically varied. Our findings reveal the dual roles of task descriptions: an insufficient task description will cause the model to overlook in-context examples, leading to poor in-context performance; once the amount of information in the task description exceeds a certain threshold, the impact of the task description shifts from negative to positive, and a performance emergence can be observed. We replicate these findings on GPT-4, observing a similar double-sided effect. This study highlights the critical role of task descriptions in in-context learning, offering valuable insights for future applications of transformer models.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"120 ","pages":"Article 103116"},"PeriodicalIF":14.7,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143714374","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"TSCMamba: Mamba meets multi-view learning for time series classification","authors":"Md Atik Ahamed ,&nbsp;Qiang Cheng","doi":"10.1016/j.inffus.2025.103079","DOIUrl":"10.1016/j.inffus.2025.103079","url":null,"abstract":"<div><div>Multivariate time series classification (TSC) is critical for various applications in fields such as healthcare and finance. While various approaches for TSC have been explored, important properties of time series, such as shift equivariance and inversion invariance, are largely underexplored by existing works. To fill this gap, we propose a novel multi-view approach to capture patterns with properties like shift equivariance. Our method integrates diverse features, including spectral, temporal, local, and global features, to obtain rich, complementary contexts for TSC. We use continuous wavelet transform to capture time–frequency features that remain consistent even when the input is shifted in time. These features are fused with temporal convolutional or multilayer perceptron features to provide complex local and global contextual information. We utilize the Mamba state space model for efficient and scalable sequence modeling and capturing long-range dependencies in time series. Moreover, we introduce a new scanning scheme for Mamba, called tango scanning, to effectively model sequence relationships and leverage inversion invariance, thereby enhancing our model’s generalization and robustness. Experiments on two sets of benchmark datasets (10+20 datasets) demonstrate our approach’s effectiveness, achieving average accuracy improvements of 4.01–6.45% and 7.93% respectively, over leading TSC models such as TimesNet and TSLANet.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"120 ","pages":"Article 103079"},"PeriodicalIF":14.7,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143678378","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}