Information FusionPub Date : 2025-09-16DOI: 10.1016/j.inffus.2025.103714
Zhenyu Liu , Heye Zhang , Yiwen Wang , Zhifan Gao
{"title":"Embracing knowledge integration from the vision-language model for federated domain generalization on multi-source fused data","authors":"Zhenyu Liu , Heye Zhang , Yiwen Wang , Zhifan Gao","doi":"10.1016/j.inffus.2025.103714","DOIUrl":"10.1016/j.inffus.2025.103714","url":null,"abstract":"<div><div>Federated Domain Generalization (FedDG) has attracted attention for its potential to enable privacy-preserving fusion of multi-source data. It aims to develop a global model in a distributed manner that generalizes to unseen clients. However, it faces the challenge of the tradeoff between inter-client and intra-client domain shifts. Knowledge distillation from the vision-language model may address this challenge by transferring its zero-shot generalization ability to client models. However, it may suffer from distribution discrepancies between the pretraining data of the vision-language model and the downstream data. Although pre-distillation fine-tuning may alleviate this issue in centralized settings, it may not be compatible with FedDG. In this paper, we introduce an in-distillation selective adaptation framework for FedDG. It selectively fine-tunes unreliable outputs while directly distilling reliable ones from the vision-language model, effectively using knowledge distillation to address the challenge in FedDG. Furthermore, we propose a federated energy-driven reliability appraisal (FedReap) method to support this framework by appraising the reliability of outputs from the vision-language model. It includes hypersphere-constraint energy construction and label-guided energy partition. These two processes enable FedReap to acquire reliable and unreliable outputs for direct distillation and adaptation. In addition, FedReap employs a dual-level distillation strategy and a dual-stage adaptation strategy for distillation and adaptation. Extensive experiments on five datasets demonstrate the effectiveness of FedReap compared to twelve state-of-the-art methods.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103714"},"PeriodicalIF":15.5,"publicationDate":"2025-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145119850","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":"DeepFake detection in the AIGC era: A survey, benchmarks, and future perspectives","authors":"Shichuang Xie , Tong Qiao , Sheng Li , Xinpeng Zhang , Jiantao Zhou , Guorui Feng","doi":"10.1016/j.inffus.2025.103740","DOIUrl":"10.1016/j.inffus.2025.103740","url":null,"abstract":"<div><div>In recent years, DeepFake has further developed, driven by continuous advances in data, computing power, and deep generative models. This emerging digital media forgery technique can manipulate or generate fake face content, increasingly blurring the boundaries between real and fake media. With the growing misuse of DeepFake, the associated risks are also intensifying. Although some research on DeepFake detection has been conducted, the research on detection is obviously falling behind DeepFake generation, and there is a lack of comprehensive and up-to-date surveys on DeepFake detection. Therefore, to effectively counter the proliferation of DeepFake face and promote the evolution of DeepFake detection, we conduct comprehensive survey and analysis. Specifically, (1) we analyze the key factors driving the proliferation of DeepFake, and we review the four representative types of DeepFake face and introduce a novel cross-modal face manipulation based on foundation models; (2) we reorganize DeepFake detection methods and establish a detection evaluation benchmark, emphasizing the potential of emerging detectors; (3) we focus on the current challenges of DeepFake forensic research and the corresponding development trends, and provide future perspectives, aiming to provide new insights for DeepFake forensic research in the AIGC era.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103740"},"PeriodicalIF":15.5,"publicationDate":"2025-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145221680","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}
Information FusionPub Date : 2025-09-16DOI: 10.1016/j.inffus.2025.103752
Shan Jiang , Wenchang Chai , Mingjin Zhang , Jiannong Cao , Shichang Xuan , Jiaxing Shen
{"title":"Verifying energy generation via edge LLM for web3-based decentralized clean energy networks","authors":"Shan Jiang , Wenchang Chai , Mingjin Zhang , Jiannong Cao , Shichang Xuan , Jiaxing Shen","doi":"10.1016/j.inffus.2025.103752","DOIUrl":"10.1016/j.inffus.2025.103752","url":null,"abstract":"<div><div>The global transition to clean energy is critical to achieving climate goals, yet traditional centralized systems face challenges in flexibility, grid resilience, and equitable access. While decentralized web3-based energy networks offer promising alternatives, existing solutions lack robust architectures to integrate distributed generation with real-time demand and fail to provide trustworthy energy verification mechanisms. This work introduces DeCEN, a decentralized clean energy network that synergizes collaborative edge computing and web3 technologies to address these gaps. DeCEN leverages autonomous edge devices to collect and process sensory data from renewable generators, enabling localized decision-making and verification of energy production. A layer-2 blockchain solution establishes a transparent web3 ecosystem, connecting clean energy generators and consumers through tokenized incentives for green energy activities. To combat fraud, DeCEN incorporates a novel large language model (LLM)-based energy verification protocol that analyzes sensory data to validate renewable claims, ensuring accountability and stabilizing token value. Additionally, a distributed LLM inference algorithm partitions LLMs into shards deployable on resource-constrained edge devices, enabling decentralized, low-latency processing while preserving data privacy and minimizing communication overhead. By integrating edge computing, blockchain, and AI-driven verification, DeCEN improves the reliability, trust, and efficiency of decentralized clean energy networks, offering a scalable pathway toward global renewable energy targets.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103752"},"PeriodicalIF":15.5,"publicationDate":"2025-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145119847","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":"Brain tumor segmentation via cross-modality semi-supervised transfer learning with 3D MRI diffusion model synthetic ultrasound","authors":"Yuhua Li , Shan Jiang, Zhiyong Yang, Liwen Wang, Shuangying Wang, Zeyang Zhou","doi":"10.1016/j.inffus.2025.103757","DOIUrl":"10.1016/j.inffus.2025.103757","url":null,"abstract":"<div><div>Accurate ultrasound segmentation is crucial for intraoperative brain navigation and can improve non-rigid registration between preoperative MRI and intraoperative ultrasound, compensating for brain shift. However, limited annotated ultrasound data hinder the application of deep learning methods. Given recent advances in brain MRI-based medical image processing, transferring MRI datasets and deep learning models to US image research via cross-modal translation may potentially enhance intelligent brain US image processing. In this paper, we propose a novel cross-modality semi-supervised transfer learning from MRI to US by leveraging annotated data in the MRI modality. A diffusion model, leveraging conditional texture features and guided mutual information, transforms well-annotated MRI images into synthetic US images with a distribution closer to real US images. Subsequently, we employ a segmentation framework that involves pretraining with synthetic US images derived from MRI through image translation, followed by semi-supervised fine-tuning using a hybrid dataset that integrates both labeled and unlabeled ultrasound data. Extensive assessments are reported on the utility of SL-DDPM against competing GAN and diffusion models in MRI-US translation. The experimental results demonstrate that our proposed transfer learning strategy achieves a segmentation accuracy of DSC of 93.43 ± 3.72 %. The effectiveness of our strategy is validated through ablation studies on fine-tuning strategies and semi-supervised learning, as well as comparisons with other state-of-the-art methods. Our transfer learning strategy enhances the accuracy and generalization of brain ultrasound segmentation models, even with limited hybrid training data, thereby assisting surgeons in identifying lesion.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103757"},"PeriodicalIF":15.5,"publicationDate":"2025-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107549","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}
Information FusionPub Date : 2025-09-15DOI: 10.1016/j.inffus.2025.103722
Huaping Zhou , Tao Wu , Kelei Sun , Jin Wu , Bin Deng
{"title":"Task-oriented multi-scale dynamic feature fusion for robust conveyor belt monitoring","authors":"Huaping Zhou , Tao Wu , Kelei Sun , Jin Wu , Bin Deng","doi":"10.1016/j.inffus.2025.103722","DOIUrl":"10.1016/j.inffus.2025.103722","url":null,"abstract":"<div><div>Existing conveyor belt monitoring methods suffer from unreasonable multi-task feature allocation and limited boundary feature extraction capability. To address these issues, this study develops a novel information fusion framework integrating Mask R-CNN-based detection and segmentation for conveyor belt status monitoring. Firstly, we propose the Multi-Scale Dynamic Feature Fusion (MS-DFF) module. It uses a multi-stage parallel multi-scale convolution network and dynamic weight adjustment mechanism to flexibly fuse and optimize multi-scale features. Secondly, we propose the Task-Oriented Module (TOM). It optimizes task adaptability between the detection and segmentation branches, combining frequency domain and spatial-domain features to meet multi-task requirements. Thirdly, we also design a Laplacian convolution fixed-weight structure to enhance target boundary information, leading to the new Boundary Enhanced (BE) segmentation head. Finally, we design the Dynamic Weighted Hybrid Loss (DWH Loss), combining Dice loss, Focal loss, and BCE loss. It dynamically adjusts weights to balance multi-task optimization, further improving segmentation boundary clarity and overall performance. We conduct extensive experiments on the conveyor belt monitoring dataset and the COCO dataset. On the conveyor belt dataset, the AP<span><math><msub><mrow></mrow><mn>50</mn></msub></math></span> for the detection task reaches 98.4 %, and the AP<span><math><msub><mrow></mrow><mn>50</mn></msub></math></span> for the segmentation task reaches 73.5 %. These results outperform most state-of-the-art methods.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103722"},"PeriodicalIF":15.5,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107554","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}
Information FusionPub Date : 2025-09-15DOI: 10.1016/j.inffus.2025.103724
Wenlong Liu , Jiaohua Qin
{"title":"Focus and learn: boosting deep multi-view clustering via hard instance awareness","authors":"Wenlong Liu , Jiaohua Qin","doi":"10.1016/j.inffus.2025.103724","DOIUrl":"10.1016/j.inffus.2025.103724","url":null,"abstract":"<div><div>Deep contrastive multi-view clustering aims to use contrastive mechanisms to exploit the complementary information from multiple features, which has attracted increasing attention in recent years. However, we observe that most contrastive multi-view clustering methods neglect the false sample pairs caused by hard samples during the process of constructing contrastive sample pairs, including negative samples exhibit high similarity and positive samples exhibit low similarity. To address this problems, we propose a novel deep contrastive multi-view clustering network for hard sample mining, termed <strong>MVC-HSM</strong>. Specifically, we propose a strategy that incorporates both coarse-grained and fine-grained perspectives. At the coarse-grained level, we perform contrastive learning by utilizing prototypes from each view, thereby mitigating hard samples at the sample level. At the fine-grained level, we first construct a comprehensive evaluation function to measure the similarity for the samples based on representation relationships and structures. In combination with the filtering effect of high-confidence pseudo-labels, we further design a contrastive learning loss for hard samples. Thus, the model could automatically increase the weight of hard samples while reducing the weight of easy samples. The superior of MVC-HSM is verified by extensive experiments on public multi-view datasets, demonstrating the proposed MVC-HSM outperforms other state-of-the-art multi-view clustering.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103724"},"PeriodicalIF":15.5,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107897","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":"CM2-STNet: Cross-modal image matching with modal-adaptive feature modulation and sparse transformer fusion","authors":"Zhizheng Zhang , Pengcheng Wei , Peilian Wu , Jindou Zhang , Boshen Chang , Zhenfeng Shao , Mingqiang Guo , Liang Wu , Jiayi Ma","doi":"10.1016/j.inffus.2025.103750","DOIUrl":"10.1016/j.inffus.2025.103750","url":null,"abstract":"<div><div>Multimodal image matching is a fundamental task in geospatial analysis, aiming to establish accurate correspondences between images captured by heterogeneous imaging devices. However, significant geometric inconsistencies and nonlinear radiometric distortions lead to large distribution gaps, posing a major challenge for cross-modal matching. Moreover, existing methods often struggle to adaptively capture intra- and inter-modal features at multiple scales and to focus on semantically relevant regions in large-scale scenes. To address these issues, we propose a novel cross-modal image matching network called CM<sup>2</sup>-STNet. Specifically, we introduce a modal-adaptive feature modulation (MAFM) module that dynamically adjusts cross-modal feature representations at multiple scales, thereby enhancing semantic consistency between modalities. In addition, a cross-modal sparse transformer fusion (CM-STF) module is developed to guide the network to concentrate on the most relevant regions, where a Top-k selection mechanism is employed to retain discriminative features while filtering out irrelevant content. Extensive experiments on multimodal remote sensing datasets demonstrate that CM<sup>2</sup>-STNet achieves accurate and robust matching performance, validating its effectiveness and generalization ability in complex real-world scenarios. Code and pre-trained model are available at https://github.com/whuzzzz/CM<sup>2</sup>-STNet.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103750"},"PeriodicalIF":15.5,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145119852","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}
Information FusionPub Date : 2025-09-15DOI: 10.1016/j.inffus.2025.103739
Han Zhou, Shuli Sun
{"title":"Distributed estimation for multi-sensor networked stochastic uncertain systems with correlated noises under a general stochastic communication protocol","authors":"Han Zhou, Shuli Sun","doi":"10.1016/j.inffus.2025.103739","DOIUrl":"10.1016/j.inffus.2025.103739","url":null,"abstract":"<div><div>The distributed state estimation problem is studied for multi-sensor networked stochastic uncertain systems with correlated noises under a stochastic communication protocol (SCP). Random parameter matrices are utilized to describe the stochastic uncertainties within the system model. Given the limited channel bandwidth among sensor nodes, a general SCP is set to randomly select multiple components from the complete state prediction estimate for transmission. A set of random variables is introduced to indicate which combination of state prediction components is selected for transmission at each time step. In the case that the sensor node does not know which combination of state prediction components from each neighboring node is transmitted to it at each time step, a distributed Kalman-like recursive estimator structure that depends on the probability distributions of random variables is developed. Under this estimator structure, an optimal distributed estimation algorithm is presented based on the linear unbiased minimum variance criterion, which necessitates the computation of estimation error cross-covariance matrices between different nodes. To avert the computation of cross-covariance matrices, a suboptimal distributed estimation algorithm is also proposed, where optimal gains are achieved by minimizing the upper bound of estimation error covariance matrix at each node. In addition, the scalar parameters in the upper bound of the covariance matrix are optimized to obtain a minimum upper bound. Stability and steady-state properties of two distributed estimation algorithms are analyzed. Finally, the effectiveness of the presented algorithms is validated through a simulation example.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103739"},"PeriodicalIF":15.5,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145269567","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}
Information FusionPub Date : 2025-09-15DOI: 10.1016/j.inffus.2025.103749
Chuang Yu , Yunpeng Liu , Jinmiao Zhao , Dou Quan , Zelin Shi , Xiangyu Yue
{"title":"Relational representation learning network for cross-spectral image patch matching","authors":"Chuang Yu , Yunpeng Liu , Jinmiao Zhao , Dou Quan , Zelin Shi , Xiangyu Yue","doi":"10.1016/j.inffus.2025.103749","DOIUrl":"10.1016/j.inffus.2025.103749","url":null,"abstract":"<div><div>Recently, feature relation learning has drawn widespread attention in cross-spectral image patch matching. However, existing related research focuses on capturing diverse feature relations between image patches and ignores sufficient intrinsic feature representations of individual image patches. To address this limitation, we propose an innovative relational representation learning that simultaneously focuses on sufficiently mining the intrinsic features of individual image patches and the feature relations between image patches. Based on this, we construct a <strong><u>R</u></strong>elational <strong><u>R</u></strong>epresentation <strong><u>L</u></strong>earning <strong><u>Net</u></strong>work (<strong>RRL-Net</strong>). Specifically, we innovatively construct an autoencoder to effectively characterize the individual intrinsic features, and introduce a feature interaction learning (FIL) module to extract deep-level feature relations. Meanwhile, to further fully mine individual intrinsic features, a lightweight multi-dimensional global-to-local attention (MGLA) module is constructed to enhance the global feature extraction of individual image patches and capture local dependencies within global features. By combining the MGLA module, we further explore the feature extraction network and construct an attention-based lightweight feature extraction (ALFE) network. Furthermore, a multi-loss post-pruning (MLPP) optimization strategy is proposed, which can greatly facilitate network optimization while avoiding increases in parameters and inference time. Extensive experiments demonstrate that our RRL-Net achieves state-of-the-art (SOTA) performance on multiple public datasets. Our code is available at <span><span>https://github.com/YuChuang1205/RRL-Net</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103749"},"PeriodicalIF":15.5,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145159141","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}
Information FusionPub Date : 2025-09-15DOI: 10.1016/j.inffus.2025.103741
Bin Pu , Mingjian Yang , Yan Kang , Guanyuan Chen , Pengchen Liang
{"title":"Hierarchical knowledge fusion for enhanced health event prediction: Discriminating between frequent and new diseases","authors":"Bin Pu , Mingjian Yang , Yan Kang , Guanyuan Chen , Pengchen Liang","doi":"10.1016/j.inffus.2025.103741","DOIUrl":"10.1016/j.inffus.2025.103741","url":null,"abstract":"<div><div>The prediction of future diseases from patients’ historical Electronic Health Records (EHRs) is of great importance for promoting patient empowerment and preventive healthcare. However, existing studies often overlook the distinction between frequent diseases and new diseases, as well as the complex and hidden relationships among diseases and patients. To address these issues, we propose HKLHEP, a novel hierarchical knowledge-learning algorithm that models health event prediction from both disease and patient perspectives. The method extracts and represents frequent and new diseases within a dynamic graph framework and enriches disease embeddings through a category tree aggregation approach; it further captures both high-level and low-level patient features in a patient-centric manner, evaluates the temporal significance of patient visits by designing a time attention mechanism, and incorporates discharge summaries via transfer learning to enhance textual representations. Experimental results on two large-scale real-world EHR datasets demonstrate that HKLHEP outperforms 11 state-of-the-art methods in health event prediction. The source code is available at <span><span>https://github.com/yangCode-res/HKLHEP</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103741"},"PeriodicalIF":15.5,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145119997","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}