Expert Systems with Applications最新文献

{"title":"Chrysanthemum image quality assessment via multi-scale feature fusion and meta-learning","authors":"Shun Zhu ,&nbsp;Xichen Yang ,&nbsp;Tianshu Wang ,&nbsp;Zhongyuan Mao ,&nbsp;Yifan Chen ,&nbsp;Jian Jiang ,&nbsp;Hui Yan","doi":"10.1016/j.eswa.2026.131378","DOIUrl":"10.1016/j.eswa.2026.131378","url":null,"abstract":"<div><div>The origin tracing of chrysanthemum is significant in ensuring the quality of chrysanthemum. With the development of computer vision, it is feasible to utilize vision technology to the origin tracing. This enables intelligent origin tracing, thereby improving efficiency and accuracy. However, image distortions are inevitable while collecting chrysanthemum images. These distortions, such as incomplete chrysanthemum tissue and poor angle, tend to reduce the accuracy of the origin tracing. Thus, it is important to measure the image quality accurately, and then further improve the accuracy of the origin tracing. Considering it, we proposed a chrysanthemum image quality assessment method. First, a two-step screening (TSS) module is designed to screen existing classically distorted images that are suitable for distorted chrysanthemum images. Second, a deep feature extraction module is utilized to extract features at different receptive field levels. Third, the semantic analysis module is used to analyze and fuse the semantic information of different features. Finally, the meta-learning framework is designed to improve the accuracy and robustness of the model. The prior knowledge acquired through meta-learning is utilized to fine-tune the model with few-shot samples. The experimental results demonstrate that the proposed method can accurately judge incomplete and angle distortions, and thus effectively promote the accuracy of origin tracing. Our codes and models are available at <span><span>https://github.com/dart-into/a-chrysanthemum-Screening-Method</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131378"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122752","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":"NeuroVision: EEG-to-image reconstruction via progressive neural encoding and cross-modal distillation","authors":"Tianwei Qu ,&nbsp;Zexue Yang ,&nbsp;Qixian Zhang","doi":"10.1016/j.eswa.2026.131526","DOIUrl":"10.1016/j.eswa.2026.131526","url":null,"abstract":"<div><div>Reconstructing visual imagery from electroencephalography (EEG) signals represents a fundamental challenge in brain-computer interfaces, as EEG recordings capture rich neural information about visual perception but lack the spatial resolution necessary for direct image reconstruction. Existing approaches typically employ diffusion models to synthesize images from EEG embeddings, yet these methods often fail to preserve the semantic content encoded in neural signals due to information loss during the encoding-decoding process. In this paper, we propose NeuroVision, a novel framework that adapts neural signal encoders for unified EEG understanding and visual reconstruction through progressive neural encoding and cross-modal distillation. Our approach addresses the fundamental trade-off between preserving EEG semantic information and achieving high-quality image reconstruction by introducing a three-stage progressive training scheme that gradually enhances reconstruction capability while maintaining original neural signal understanding. We develop a temporal-spatial attention fusion mechanism to capture multi-channel temporal dynamics in EEG signals, coupled with adaptive feature alignment that dynamically maps EEG representations to visual feature spaces. Furthermore, we introduce a semantic-preserving loss function that ensures reconstructed images faithfully reflect the semantic content of neural activity rather than generating visually plausible but semantically inconsistent outputs. Extensive experiments demonstrate that NeuroVision achieves superior reconstruction quality compared to existing diffusion-based approaches while significantly better preserving semantic correspondence between neural signals and visual content, establishing a new paradigm for EEG-to-image reconstruction that prioritizes semantic fidelity alongside visual quality.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131526"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192556","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":"PQS-BFL: A post-quantum secure blockchain-based federated learning framework","authors":"Daniel Commey ,&nbsp;Garth V. Crosby","doi":"10.1016/j.eswa.2026.131449","DOIUrl":"10.1016/j.eswa.2026.131449","url":null,"abstract":"<div><div>Federated Learning (FL) enables collaborative model training while preserving data privacy, but its classical cryptographic underpinnings are vulnerable to quantum attacks. This vulnerability is particularly critical in sensitive domains like healthcare, where data requires long-term forward secrecy. This paper introduces PQS-BFL (Post-Quantum Secure Blockchain-based Federated Learning), a framework integrating post-quantum cryptography (PQC) with blockchain verification to secure FL against quantum adversaries. We employ ML-DSA-65 (standardized in FIPS 204, formerly Dilithium) signatures to authenticate model updates and leverage optimized smart contracts for decentralized validation. Designed for permissioned consortium environments (e.g., healthcare research networks), our framework ensures update integrity independent of the underlying ledger’s quantum resistance. Extensive evaluations on diverse datasets (MNIST, SVHN, HAR) demonstrate that PQS-BFL achieves feasible cryptographic operations (average PQC sign time: 0.65 ms, verify time: 0.53 ms) with a fixed signature size of 3309 Bytes. While blockchain integration incurs higher gas usage averaging 1.72 × 10⁶ units per update due to PQC verification complexity, we demonstrate that this cost is negligible in private consortium settings where gas fees are nominal. Importantly, the cryptographic overhead relative to transaction time remains minimal (typically  &lt; 0.2%), confirming that PQS-BFL is a viable architecture for securing critical infrastructure against future quantum threats.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131449"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192558","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":"Vul2image: A quick image-inspired and CNN-based vulnerability detection system","authors":"Rong Ren ,&nbsp;Mushi Zhou ,&nbsp;Ni Liao ,&nbsp;Bing Zhang ,&nbsp;Guoyan Huang ,&nbsp;Haitao He ,&nbsp;Qian Wang","doi":"10.1016/j.eswa.2026.131468","DOIUrl":"10.1016/j.eswa.2026.131468","url":null,"abstract":"<div><div>Given the accuracy of deep learning (DL) in image classification, some studies have applied DL algorithms to vulnerability detection by characterizing software source code as RGB images. However, effectively utilizing RGB images to store multiple code semantics remains a challenge, impacting the effectiveness of vulnerability detection. To address this, we developed Vul2image, a quick Image-inspired and CNN-based Vulnerability Detection System. By focusing on Potential Vulnerable Code Fragments (PVCFs) and their context code, Vul2image minimized interference from irrelevant information and achieved comprehensive coverage of vulnerability features. It constructed an RGB fine-grained image model incorporating textual, semantic, and structural information from code text, Control Dependency Graphs (CDGs), and Data Dependency Graphs (DDGs), resulting in improved detection efficiency. Evaluated on three datasets with increasing vulnerability types (including our self-collected, VulCNN, and Devign), Vul2image achieved the best results on our dataset, outperforming 9 classic (incl. 4 LLM-based) and 2 SOTA image-based detectors (VulCNN, VulGAI) and demonstrating performance comparable to 7 transformer-encoder-based methods, showing strong precision for specific vulnerability types. In practice, Vul2image was 35 times faster than VulCNN and successfully identified 21 reported and 5 unreported vulnerabilities in various real-world systems and software within 67,352,085 lines of code, showcasing its large-scale vulnerability detection capability.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131468"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192559","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":"DAWN: Dimension-aware graph contrastive learning for few-shot dissolved gas analysis","authors":"Jiyuan Sun ,&nbsp;Huifang Ma ,&nbsp;Shuai Yang ,&nbsp;Rui Bing ,&nbsp;Zhixin Li","doi":"10.1016/j.eswa.2026.131504","DOIUrl":"10.1016/j.eswa.2026.131504","url":null,"abstract":"<div><div>Dissolved Gas Analysis (DGA) is a widely used technique for identifying characteristic gas signatures indicative of transformer faults. However, traditional DGA methods struggle to capture the complex interdependencies among gases. Although deep learning has shown promise in this domain, existing approaches face significant limitations under few-shot learning scenarios, primarily due to severe class imbalance and high inter-class similarity, which lead to diagnostic ambiguity. Moreover, these methods often overlook critical inter-gas relationships that are essential for understanding underlying fault mechanisms. To address these challenges, we propose <strong>DAWN</strong> (<strong>D</strong>imension-<strong>AW</strong>are Graph Co<strong>N</strong>trastive Learning), a novel framework that integrates two key components: (1) a contrastive few-shot learning module with clustering consistency loss to enhance the discriminability of similar fault categories through lightweight fine-tuning, and (2) a knowledge-enhanced dimension graph that explicitly models structural dependencies among gas features by combining statistical correlations with expert domain knowledge. Extensive evaluations on DGA datasets demonstrate that DAWN achieves state-of-the-art performance, improving rare fault detection accuracy by over 15% compared to conventional methods. To the best of our knowledge, this work represents the first contrastive few-shot learning framework tailored for DGA-based fault diagnosis.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131504"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192583","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":"MI-DGHCL: Motor imagery EEG domain generalization via hyperbolic contrastive learning","authors":"Junfu Chen ,&nbsp;Dechang Pi ,&nbsp;Feng Gao ,&nbsp;Cheng Ma ,&nbsp;Yang Chen","doi":"10.1016/j.eswa.2026.131477","DOIUrl":"10.1016/j.eswa.2026.131477","url":null,"abstract":"<div><div>Cross-subject motor imagery (MI) Electroencephalogram (EEG) decoding remains challenging due to significant domain shifts caused by inter-subject variability. Despite the success of Euclidean domain alignment techniques via deep learning, they fail to capture the potential hierarchical structures inherent in EEG signals. To address this challenge, we propose MI-DGHCL, a domain generalization model for MI-EEG signals that leverages hyperbolic contrastive learning. MI-DGHCL introduces a Mamba-based feature extractor, incorporating slice feature embedding and slice-aware scanning modules that simultaneously capture both global information and local contextual features of EEG signals. Furthermore, it enforces both feature consistency and semantic consistency. Feature consistency is attained by aligning domains through the minimization of feature covariance in Euclidean space. Subsequently, hierarchical representations are derived via hyperbolic embeddings, and supervised contrastive learning pushes intra-class samples across subjects close in hyperbolic space, thereby ensuring semantic consistency. Comprehensive experiments on 3 MI-EEG datasets demonstrate MI-DGHCL yields superior results, compared to advanced methods.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131477"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192780","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":"Towards privacy-preserving and communication-efficient federated distillation","authors":"Xinge Ma,&nbsp;Jin Wang,&nbsp;Xuejie Zhang","doi":"10.1016/j.eswa.2026.131503","DOIUrl":"10.1016/j.eswa.2026.131503","url":null,"abstract":"<div><div>Federated distillation (FD) has emerged as a promising alternative to federated learning (FL) for collaborative training across decentralized clients to benefit from their private data by exchanging model outputs associated with a large-scale unlabeled public dataset. However, recent studies have revealed that sharing model outputs still poses privacy risks of data exposure when encountering malicious attacks. Although incorporating differential privacy (DP) can provide strong privacy guarantees for FD by perturbing model parameters to produce secure model outputs or directly injecting calibrated noise to model outputs before sharing, it either suffer from inefficient knowledge transfer due to the limited domain-specific knowledge learned by local models or incurs a high privacy cost that significantly compromises the utility of model outputs because the required noise magnitude is proportional to the scale of the perturbed target. To balance the trade-off between knowledge utility and privacy protection, this paper presents FedLA, a privacy-preserving and communication-efficient FD framework empowered by local differential privacy and active data sampling, which proactively selects the most informative subset from the large-scale unlabeled public dataset as a high-quality carrier for local perturbation and knowledge transfer. The resulting reduction in the number of queries to local models minimizes privacy cost and communication overhead while maximizing model performance. Experiments on two popular benchmark datasets across diverse evaluation settings demonstrate the superiority of FedLA in terms of model accuracy, communication efficiency, privacy cost, and attack defense. The code is available at: <span><span>https://github.com/maxinge8698/FedLA</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131503"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192778","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":"A dynamic model of rumor propagation based on adversarial behavior and evolutionary games","authors":"Chaolong Jia ,&nbsp;Guicai Deng ,&nbsp;Xiaochuan Chen ,&nbsp;Kangle Chen ,&nbsp;Rong Wang ,&nbsp;Tun Li ,&nbsp;Yunpeng Xiao","doi":"10.1016/j.eswa.2026.131490","DOIUrl":"10.1016/j.eswa.2026.131490","url":null,"abstract":"<div><div>Rumors pose a serious threat to social stability across modern social networks. To address the ongoing confrontation between rumor and anti-rumor groups, this study proposes a framework for analyzing rumor dissemination based on confrontational behavior and evolutionary game theory. This study considers differences in users’ real-time preferences for rumors relative to anti-rumors, which significantly influence rumor spread and are complex to measure. Multivariate linear regression is applied to develop a real-time preference metric and identify multiple factors influencing user preferences. Evolutionary game theory is then employed to construct a game-theoretic mechanism for anti-rumor behavior. The coexistence and confrontation between the two groups were analyzed using the Rosenzweig–MacArthur equations to characterize overall rumor propagation dynamics. For individual users, three states are defined after exposure to both rumor types: a wavering state (I), a rumor-adopting state (P), and an anti-rumor state (O), which form the basis of a rumor dissemination dynamics model. Experimental results show that the proposed model achieves an average SMAPE of 16%, while complementary metrics such as RMSE and AUC further confirm its reliable performance across different evaluation dimensions.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131490"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192711","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":"Deep global-ranking hashing via average precision approximation for large-scale image retrieval","authors":"Lei Wang ,&nbsp;Yongyue Fu ,&nbsp;Qibing Qin ,&nbsp;Lei Huang ,&nbsp;Wenfang Zhang","doi":"10.1016/j.eswa.2026.131557","DOIUrl":"10.1016/j.eswa.2026.131557","url":null,"abstract":"<div><div>Deep hashing algorithms have become a mainstream solution for large-scale multimedia retrieval due to their advantages in search efficiency and storage space. The algorithm is able to jointly learn semantic features and hash functions to encode raw data into compact binary codes, with significant differentiation. For the retrieval task, the central goal is to learn a ranking relation that can efficiently rank candidate results. Most of the existing hash methods adopt pair-wise or multi-wise strategies to learn the ranking results by minimizing the relative distance between similar samples or maximizing the distance between dissimilar samples. This type of approach can indeed improve the consistency of the local neighborhood, but its optimization objective is essentially limited to the local ranking relationship and fails to globally rank the samples in the entire retrieval set. Ultimately, this leads to the model possibly achieving good performance in local ranking but being unable to guarantee the overall relevance and stability of the global retrieval list. Especially when the sample distribution is complex or there is significant semantic overlap within or between categories; the local ranking method cannot fully capture the relationships among different samples, thereby limiting the discriminative ability and ranking quality of hash codes in actual retrieval scenarios. To address this issue, by introducing the average precision (AP) metric as the optimization objective, a novel Deep Global-ranking Hashing framework via average precision approximation (DGrH) is proposed to learn hash spaces with ranking relationships. Specifically, based on the discrete Heaviside function, a novel Ranking-AP optimization strategy is introduced into deep hashing to learn the global semantic relationships. The Sigmoid function is employed to smoothly approximate the non-differentiable discrete Heaviside function, making it differentiable. On this basis, the overall objective function and the novel Ranking-AP loss could enhance the learning of global ranking information. This helps capture and preserve high-quality high-quality global ranking relationships among samples more effectively in hash code learning. Extensive experiments on several benchmark datasets validate the efficacy of our designed DGrH framework, which consistently outperforms the mainstream deep hashing by large gaps. The code for the implementation of our DGrH framework is available at <span><span>https://github.com/QinLab-WFU/DGrH</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131557"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192701","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":"Privacy preservation in face soft biometrics via attribute disentanglement","authors":"Yue Wang ,&nbsp;Biao Jin ,&nbsp;Zheyu Chen ,&nbsp;Jinsen Lin ,&nbsp;Zhiqiang Yao","doi":"10.1016/j.eswa.2026.131520","DOIUrl":"10.1016/j.eswa.2026.131520","url":null,"abstract":"<div><div>Soft biometric privacy enhancement techniques have been widely adopted in face recognition systems to prevent attackers from inferring sensitive attributes such as gender, age, and ethnicity from facial images. Although existing facial attribute privacy protection methods can conceal multiple attributes simultaneously, they still face three key challenges: (1) precisely modifying target attributes while preserving non-target attributes; (2) achieving a balanced trade-off between privacy preservation and identity recognition utility; and (3) providing flexible and user-controllable options for attribute protection. To address these challenges, this paper proposes a novel face soft biometric privacy protection framework based on attribute disentanglement, which effectively conceals sensitive facial attributes while maximizing identity recognition accuracy. First, a mapping module guided by an attribute supervision loss is introduced to learn a disentangled latent space, where the semantic representations of different attributes are separated for controllable manipulation. Second, a face matcher combined with a dedicated face matching loss enforces identity consistency, enabling the model to preserve recognition utility while suppressing sensitive attribute leakage. Finally, an attribute selection module (ASM) is incorporated during the inference stage, allowing users to flexibly specify which attributes (e.g., gender, age, and smile) to protect, thereby enhancing adaptability and user-level controllability in privacy-sensitive applications. Experimental results demonstrate that the proposed method effectively safeguards the privacy of facial attributes while maintaining high identity recognition utility. Code is available at <span><span>https://github.com/Forestmumu/PrivAD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131520"},"PeriodicalIF":7.5,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192851","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}