Expert Systems with Applications最新文献

Hybrid intelligence–driven global path planning for ships in complex maritime environments 复杂海洋环境下船舶混合智能驱动的全局路径规划

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-05 DOI: 10.1016/j.eswa.2026.131473

Jiao Liu , Kaige Zhu , Yuanqiang Zhang , Miao Gao , Pengjun Zheng

Global ship path planning in complex maritime environments is challenged by dynamic disturbances, vessel-specific constraints, and long-range trajectory dependencies. This study develops an integrated hybrid planning framework that combines deep generative modeling with rule-based optimization. Automatic identification system trajectory time series are first transformed into Gramian Angular Field images to enhance spatio-temporal feature extraction. Vessel type and length are encoded as one-hot vectors and introduced as conditional variables, enabling personalized path generation. These inputs are processed by a Multi-Head Attention–based Conditional Wasserstein Generative Adversarial Network with Gradient Penalty (MHA-cWGAN-GP), in which multi-head attention is used to model long-range dependencies, and conditional Generative Adversarial Network (cGAN) training together with a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) objective is adopted to improve conditioning behavior and training robustness. The model generates initial navigation paths, which are further refined using an A* search procedure that incorporates wind and current disturbances, as well as constraints such as static obstacles, water depth, and Traffic Separation Scheme (TSS) regulations. The final path is smoothed to ensure feasibility and compliance. In case studies for the Ningbo–Zhoushan Port and Yangtze River Estuary, the hybrid planner reduces the number of search nodes from 45 to 57 to 29–35 while simultaneously enforcing TSS, water-depth, wind, and current constraints, with only about a 3–4% increase in path length relative to classical A* and Dijkstra algorithms. The results indicate that the proposed framework effectively integrates learning and optimization, offering a practical and intelligent solution for real-world maritime path planning.

在复杂的海洋环境中，全球船舶路径规划受到动态干扰、船舶特定约束和远程轨迹依赖的挑战。本研究开发了一个集成的混合规划框架，将深度生成建模与基于规则的优化相结合。首先将自动识别系统的轨迹时间序列转化为格莱曼角场图像，增强时空特征提取。船舶类型和长度被编码为单热向量，并作为条件变量引入，从而实现个性化路径生成。这些输入通过基于多头注意力的Wasserstein梯度惩罚条件生成对抗网络（mfa - cwgan - gp）进行处理，其中多头注意力用于建立远程依赖关系模型，并采用条件生成对抗网络（cGAN）训练与WGAN-GP目标相结合来提高条件反射行为和训练鲁棒性。该模型生成初始导航路径，并使用A*搜索程序进一步优化，该搜索程序包含风和电流干扰，以及静态障碍物、水深和交通分道制（TSS）法规等约束条件。最后的路径被平滑以确保可行性和合规性。以宁波-舟山港和长江口为例，混合规划器将搜索节点数从45 ~ 57个减少到29 ~ 35个，同时执行TSS、水深、风和电流约束，路径长度仅比经典a *和Dijkstra算法增加约3 ~ 4%。结果表明，该框架有效地将学习与优化相结合，为现实世界的海上路径规划提供了实用的智能解决方案。

{"title":"Hybrid intelligence–driven global path planning for ships in complex maritime environments","authors":"Jiao Liu , Kaige Zhu , Yuanqiang Zhang , Miao Gao , Pengjun Zheng","doi":"10.1016/j.eswa.2026.131473","DOIUrl":"10.1016/j.eswa.2026.131473","url":null,"abstract":"<div><div>Global ship path planning in complex maritime environments is challenged by dynamic disturbances, vessel-specific constraints, and long-range trajectory dependencies. This study develops an integrated hybrid planning framework that combines deep generative modeling with rule-based optimization. Automatic identification system trajectory time series are first transformed into Gramian Angular Field images to enhance spatio-temporal feature extraction. Vessel type and length are encoded as one-hot vectors and introduced as conditional variables, enabling personalized path generation. These inputs are processed by a Multi-Head Attention–based Conditional Wasserstein Generative Adversarial Network with Gradient Penalty (MHA-cWGAN-GP), in which multi-head attention is used to model long-range dependencies, and conditional Generative Adversarial Network (cGAN) training together with a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) objective is adopted to improve conditioning behavior and training robustness. The model generates initial navigation paths, which are further refined using an A* search procedure that incorporates wind and current disturbances, as well as constraints such as static obstacles, water depth, and Traffic Separation Scheme (TSS) regulations. The final path is smoothed to ensure feasibility and compliance. In case studies for the Ningbo–Zhoushan Port and Yangtze River Estuary, the hybrid planner reduces the number of search nodes from 45 to 57 to 29–35 while simultaneously enforcing TSS, water-depth, wind, and current constraints, with only about a 3–4% increase in path length relative to classical A* and Dijkstra algorithms. The results indicate that the proposed framework effectively integrates learning and optimization, offering a practical and intelligent solution for real-world maritime path planning.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131473"},"PeriodicalIF":7.5,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122663","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}

引用次数: 0

DORF-EASNet: physics-driven real-time seafloor classification via entropy‑regularized acoustic features and adaptive model activation DORF-EASNet：物理驱动的实时海底分类，通过熵正则化声学特征和自适应模型激活

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-04 DOI: 10.1016/j.eswa.2026.131461

Xi Zhao, Qiangqiang Yuan, Quanyin Zhang, Jiadan Xu

Real-time seabed sediment classification (SSC) is crucial for underwater navigation, operations, and habitat assessment. Conventional methods relying on post-mission multibeam-echosounder (MBES) data processing impede in situ decision-making. We propose a novel, real-time SSC method deployable on both shipborne and Autonomous Underwater Vehicle (AUV) platforms, integrating three core components. Primarily, an efficient preprocessing pipeline comprising georeferencing, radiometric normalization, noise suppression, and incidence-angle correction enables rapid conversion of raw MBES backscatter into geometry-consistent tiles, supporting real-time operation with sub-second responsiveness. Afterwards, the system extracts multi-modal descriptors by combining entropy-regularised angular-response fitting for acoustic backscatter, object-level texture analysis using adaptive graph segmentation, and curvature-aware terrain metrics derived from quadratic surface fitting under entropy constraints by considering the physical responses and spatial distribution of MBES images and point clouds. Finally, a Dynamic Optimal Random Forest with Entropy-Adaptive Subnetwork Selection (DORF-EASNet) dynamically selects between a global classifier and lightweight domain-specific sub-models to match local acoustic complexity, achieving a balance between inference efficiency and physical interpretability. Field experiments conducted in Jiaozhou Bay and the South China Sea demonstrate the proposed framework’s robustness across platforms and sensing configurations, achieving macro-F1 scores of 0.881 and 0.913, respectively, while maintaining real-time processing capability exceeding that of conventional offline methods.

实时海底沉积物分类（SSC）对水下导航、作业和栖息地评估至关重要。依靠任务后多波束测深（MBES）数据处理的传统方法阻碍了现场决策。我们提出了一种新型的实时SSC方法，可部署在舰载和自主水下航行器（AUV）平台上，集成了三个核心组件。首先，高效的预处理管道包括地理参考、辐射归一化、噪声抑制和入射角校正，可以将原始MBES背散射快速转换为几何一致的瓷砖，支持亚秒级响应的实时操作。然后，考虑MBES图像和点云的物理响应和空间分布，结合声学后向散射的熵正则化角响应拟合、自适应图分割的目标级纹理分析以及熵约束下二次曲面拟合的曲率感知地形度量，提取多模态描述符。最后，基于熵自适应子网络选择的动态最优随机森林（DORF-EASNet）在全局分类器和轻量化领域特定子模型之间动态选择以匹配局部声学复杂性，实现了推理效率和物理可解释性之间的平衡。在胶州湾和南海进行的现场实验表明，该框架具有跨平台、跨感知配置的鲁棒性，宏观f1得分分别达到0.881和0.913，同时保持了超过传统离线方法的实时处理能力。

{"title":"DORF-EASNet: physics-driven real-time seafloor classification via entropy‑regularized acoustic features and adaptive model activation","authors":"Xi Zhao, Qiangqiang Yuan, Quanyin Zhang, Jiadan Xu","doi":"10.1016/j.eswa.2026.131461","DOIUrl":"10.1016/j.eswa.2026.131461","url":null,"abstract":"<div><div>Real-time seabed sediment classification (SSC) is crucial for underwater navigation, operations, and habitat assessment. Conventional methods relying on post-mission multibeam-echosounder (MBES) data processing impede in situ decision-making. We propose a novel, real-time SSC method deployable on both shipborne and Autonomous Underwater Vehicle (AUV) platforms, integrating three core components. Primarily, an efficient preprocessing pipeline comprising georeferencing, radiometric normalization, noise suppression, and incidence-angle correction enables rapid conversion of raw MBES backscatter into geometry-consistent tiles, supporting real-time operation with sub-second responsiveness. Afterwards, the system extracts multi-modal descriptors by combining entropy-regularised angular-response fitting for acoustic backscatter, object-level texture analysis using adaptive graph segmentation, and curvature-aware terrain metrics derived from quadratic surface fitting under entropy constraints by considering the physical responses and spatial distribution of MBES images and point clouds. Finally, a Dynamic Optimal Random Forest with Entropy-Adaptive Subnetwork Selection (DORF-EASNet) dynamically selects between a global classifier and lightweight domain-specific sub-models to match local acoustic complexity, achieving a balance between inference efficiency and physical interpretability. Field experiments conducted in Jiaozhou Bay and the South China Sea demonstrate the proposed framework’s robustness across platforms and sensing configurations, achieving macro-F1 scores of 0.881 and 0.913, respectively, while maintaining real-time processing capability exceeding that of conventional offline methods.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131461"},"PeriodicalIF":7.5,"publicationDate":"2026-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122660","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}

引用次数: 0

Robot path planning based on multi-strategy enhanced aquila optimizer algorithm in complex environments 复杂环境下基于多策略增强aquila优化算法的机器人路径规划

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-03 DOI: 10.1016/j.eswa.2026.131489

Yu Zhou , Xing Liu , Jianqiao Long , Yitian Lu , Jiaoyang Cheng , Jichun Li

Path planning is a core challenge in autonomous navigation and continuously attracts significant attention in mobile robotics. While optimization algorithms are widely employed for solving robot path planning problems, the Aquila Optimizer (AO) suffers from slow convergence and a tendency to become trapped in local optima. To address these limitations, we propose a robot path planning method based on a Multi-strategy Enhanced Aquila Optimizer (MEAO). In MEAO, the initial population is enhanced using opposition-based learning, and an adaptive parameter mechanism balances exploration and exploitation. During the narrowed exploration phase, a phasor operator enables non-parametric optimization to improve global search capability, while a differential evolution mutation strategy is embedded to strengthen local exploitation. The algorithm’s performance is validated on the CEC2022 benchmark functions with ablation studies confirming the effectiveness and synergy of the various strategies. MEAO is further applied to robot path planning, with simulations performed on various complex two-dimensional grid maps, and comparisons made against several intelligent optimization-based algorithms. In addition, to address the limitations of the traditional Dynamic Window Approach (DWA) in terms of dynamic obstacle avoidance robustness and susceptibility to local minima, we introduce a dynamic threat response mechanism and an adaptive heading trap detection strategy. A collaborative framework combining MEAO-based global planning with the improved DWA for local obstacle avoidance is then established. Experimental results demonstrate that MEAO achieves shorter path lengths and faster convergence, while the improved DWA significantly enhances obstacle avoidance robustness in complex environments. The proposed collaborative framework thus ensures globally optimal paths and reliable real-time local obstacle avoidance, demonstrating the practicality and efficiency of the MEAO algorithm and improved DWA for mobile robot navigation.

路径规划是自主导航的核心问题，一直是移动机器人研究的热点。虽然优化算法被广泛用于解决机器人路径规划问题，但Aquila Optimizer （AO）存在收敛速度慢且容易陷入局部最优的问题。为了解决这些限制，我们提出了一种基于多策略增强Aquila优化器（MEAO）的机器人路径规划方法。在MEAO中，使用基于对立的学习增强初始种群，并采用自适应参数机制平衡探索和开发。在狭窄的勘探阶段，相量算子实现非参数优化以提高全局搜索能力，而嵌入差分进化突变策略以加强局部开发。该算法的性能在CEC2022基准函数上进行了验证，并进行了消融研究，证实了各种策略的有效性和协同性。MEAO进一步应用于机器人路径规划，在各种复杂的二维网格地图上进行了仿真，并与几种基于智能优化的算法进行了比较。此外，为了解决传统动态窗口方法在动态避障鲁棒性和局部最小敏感性方面的局限性，引入了动态威胁响应机制和自适应航向陷阱检测策略。建立了基于meao的全局规划与改进DWA相结合的局部避障协作框架。实验结果表明，MEAO的路径长度更短，收敛速度更快，改进的DWA显著提高了复杂环境下的避障鲁棒性。所提出的协作框架保证了全局最优路径和可靠的实时局部避障，证明了MEAO算法和改进的DWA在移动机器人导航中的实用性和高效性。

{"title":"Robot path planning based on multi-strategy enhanced aquila optimizer algorithm in complex environments","authors":"Yu Zhou , Xing Liu , Jianqiao Long , Yitian Lu , Jiaoyang Cheng , Jichun Li","doi":"10.1016/j.eswa.2026.131489","DOIUrl":"10.1016/j.eswa.2026.131489","url":null,"abstract":"<div><div>Path planning is a core challenge in autonomous navigation and continuously attracts significant attention in mobile robotics. While optimization algorithms are widely employed for solving robot path planning problems, the Aquila Optimizer (AO) suffers from slow convergence and a tendency to become trapped in local optima. To address these limitations, we propose a robot path planning method based on a Multi-strategy Enhanced Aquila Optimizer (MEAO). In MEAO, the initial population is enhanced using opposition-based learning, and an adaptive parameter mechanism balances exploration and exploitation. During the narrowed exploration phase, a phasor operator enables non-parametric optimization to improve global search capability, while a differential evolution mutation strategy is embedded to strengthen local exploitation. The algorithm’s performance is validated on the CEC2022 benchmark functions with ablation studies confirming the effectiveness and synergy of the various strategies. MEAO is further applied to robot path planning, with simulations performed on various complex two-dimensional grid maps, and comparisons made against several intelligent optimization-based algorithms. In addition, to address the limitations of the traditional Dynamic Window Approach (DWA) in terms of dynamic obstacle avoidance robustness and susceptibility to local minima, we introduce a dynamic threat response mechanism and an adaptive heading trap detection strategy. A collaborative framework combining MEAO-based global planning with the improved DWA for local obstacle avoidance is then established. Experimental results demonstrate that MEAO achieves shorter path lengths and faster convergence, while the improved DWA significantly enhances obstacle avoidance robustness in complex environments. The proposed collaborative framework thus ensures globally optimal paths and reliable real-time local obstacle avoidance, demonstrating the practicality and efficiency of the MEAO algorithm and improved DWA for mobile robot navigation.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131489"},"PeriodicalIF":7.5,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122751","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}

引用次数: 0

3D-MolGL: A multimodal framework for integrating 3D molecular graphs into language models 3D- molgl：用于将3D分子图集成到语言模型中的多模态框架

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-03 DOI: 10.1016/j.eswa.2026.131437

Huizhi Li , Dagang Li , Jinglin Zhang , Yuhui Zheng , Cong Bai

Large Language Models (LLMs) have exhibited remarkable capabilities in natural language generation and have been extensively applied to diverse tasks such as text generation and medical literature analysis, demonstrating robust proficiency in structured data processing and knowledge extraction. However, these models generally overlook the crucial three-dimensional (3D) molecular conformations, which are vital for understanding key chemical properties. This oversight significantly limits the potential of LLMs in the biomolecular field, particularly in complex tasks like drug structure discovery. To address this, we propose a Multimodal Framework for Integrating 3D Molecular Graphs into Language Models (3D-MolGL). It employs a Physics-Informed Equivariant Graph Neural Network (PI-EGNN) incorporating physically meaningful edge-level priors and physics-based regularization, aligning learned representations with empirical data and physical laws. Our approach incorporates an Iterative Cross-Modal Fusion module to reinforce structural and linguistic information, enabling the model to capture complex dependencies and improve the alignment between molecular data and natural language. Moreover, the Region-Phrase Semantic Grounding module enables fine-grained alignment between molecular substructures and linguistic tokens, thereby reinforcing the connection between molecular semantics and their textual representation. Additionally, the Best-of-N sampling strategy enhances output reliability. Notably, 3D-MolGL achieves competitive or state-of-the-art performance in molecule captioning and 3D-aware question answering tasks, while utilizing approximately 75% fewer parameters than existing large-scale multimodal architectures. This demonstrates that robust molecular reasoning capabilities can be achieved with more compact models, providing a promising new perspective for interpretable AI in chemistry.

大型语言模型（llm）在自然语言生成方面表现出卓越的能力，并被广泛应用于文本生成和医学文献分析等各种任务，显示出在结构化数据处理和知识提取方面的强大能力。然而，这些模型通常忽略了关键的三维（3D）分子构象，这对于理解关键的化学性质至关重要。这种疏忽极大地限制了法学硕士在生物分子领域的潜力，特别是在药物结构发现等复杂任务中。为了解决这个问题，我们提出了一个将3D分子图集成到语言模型中的多模态框架（3D- molgl）。它采用了一种物理知情的等变图神经网络（PI-EGNN），结合了物理上有意义的边缘级先验和基于物理的正则化，将学习到的表示与经验数据和物理定律对齐。我们的方法结合了一个迭代跨模态融合模块来强化结构和语言信息，使模型能够捕获复杂的依赖关系，并改善分子数据和自然语言之间的一致性。此外，区域-短语语义基础模块可以实现分子子结构和语言标记之间的细粒度对齐，从而加强分子语义与其文本表示之间的联系。此外，Best-of-N采样策略提高了输出可靠性。值得注意的是，3D-MolGL在分子标注和3d感知问答任务方面实现了具有竞争力或最先进的性能，同时使用的参数比现有的大规模多模态架构减少了约75%。这表明强大的分子推理能力可以通过更紧凑的模型来实现，为化学领域的可解释人工智能提供了一个有希望的新视角。

{"title":"3D-MolGL: A multimodal framework for integrating 3D molecular graphs into language models","authors":"Huizhi Li , Dagang Li , Jinglin Zhang , Yuhui Zheng , Cong Bai","doi":"10.1016/j.eswa.2026.131437","DOIUrl":"10.1016/j.eswa.2026.131437","url":null,"abstract":"<div><div>Large Language Models (LLMs) have exhibited remarkable capabilities in natural language generation and have been extensively applied to diverse tasks such as text generation and medical literature analysis, demonstrating robust proficiency in structured data processing and knowledge extraction. However, these models generally overlook the crucial three-dimensional (3D) molecular conformations, which are vital for understanding key chemical properties. This oversight significantly limits the potential of LLMs in the biomolecular field, particularly in complex tasks like drug structure discovery. To address this, we propose a Multimodal Framework for Integrating 3D Molecular Graphs into Language Models (<strong>3D-MolGL</strong>). It employs a Physics-Informed Equivariant Graph Neural Network (PI-EGNN) incorporating physically meaningful edge-level priors and physics-based regularization, aligning learned representations with empirical data and physical laws. Our approach incorporates an Iterative Cross-Modal Fusion module to reinforce structural and linguistic information, enabling the model to capture complex dependencies and improve the alignment between molecular data and natural language. Moreover, the Region-Phrase Semantic Grounding module enables fine-grained alignment between molecular substructures and linguistic tokens, thereby reinforcing the connection between molecular semantics and their textual representation. Additionally, the Best-of-N sampling strategy enhances output reliability. Notably, 3D-MolGL achieves competitive or state-of-the-art performance in molecule captioning and 3D-aware question answering tasks, while utilizing approximately 75% fewer parameters than existing large-scale multimodal architectures. This demonstrates that robust molecular reasoning capabilities can be achieved with more compact models, providing a promising new perspective for interpretable AI in chemistry.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131437"},"PeriodicalIF":7.5,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122659","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}

引用次数: 0

Discovering anomalous sequences in attributed graphs: A parameter-light approach 发现属性图中的异常序列：一种轻参数的方法

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-02 DOI: 10.1016/j.eswa.2026.131467

Cheng He , Xinyang Chen , Amaël Broustet , Guoting Chen

Graphs have been widely used across scientific disciplines, from sociology to biology, particularly when modeling temporal evolution. Although many algorithms have been developed to discover patterns in graphs, they face three main limitations. First, most algorithms assume that each node or edge is associated with a single attribute, whereas real-world applications often involve multiple attributes to capture events more comprehensively. Second, existing methods typically require tuning several hyperparameters, which can vary significantly across different datasets. Third, most approaches focus on identifying frequent patterns, often overlooking rare but meaningful ones. To address these limitations, this paper proposes a framework for discovering anomalous sequences in attributed graphs. Instead of relying on frequency-based measures, the framework adopts an entropy-based method for pattern mining, thereby requiring at most one hyperparameter. Experimental results on real-world datasets demonstrate the effectiveness of the proposed approach in detecting anomalous sequences. Moreover, we extend the framework to applications in optics, where it is used to evaluate phase differences.

从社会学到生物学，图形在科学学科中被广泛使用，特别是在建模时间进化时。尽管已经开发了许多算法来发现图中的模式，但它们面临三个主要限制。首先，大多数算法假设每个节点或边缘与单个属性相关联，而实际应用程序通常涉及多个属性以更全面地捕获事件。其次，现有的方法通常需要调优几个超参数，这在不同的数据集上可能会有很大的不同。第三，大多数方法专注于识别频繁的模式，往往忽略了罕见但有意义的模式。为了解决这些限制，本文提出了一个发现属性图中异常序列的框架。该框架不依赖基于频率的度量，而是采用基于熵的方法进行模式挖掘，因此最多只需要一个超参数。在实际数据集上的实验结果证明了该方法在检测异常序列方面的有效性。此外，我们将该框架扩展到光学应用中，用于评估相位差。

{"title":"Discovering anomalous sequences in attributed graphs: A parameter-light approach","authors":"Cheng He , Xinyang Chen , Amaël Broustet , Guoting Chen","doi":"10.1016/j.eswa.2026.131467","DOIUrl":"10.1016/j.eswa.2026.131467","url":null,"abstract":"<div><div>Graphs have been widely used across scientific disciplines, from sociology to biology, particularly when modeling temporal evolution. Although many algorithms have been developed to discover patterns in graphs, they face three main limitations. First, most algorithms assume that each node or edge is associated with a single attribute, whereas real-world applications often involve multiple attributes to capture events more comprehensively. Second, existing methods typically require tuning several hyperparameters, which can vary significantly across different datasets. Third, most approaches focus on identifying frequent patterns, often overlooking rare but meaningful ones. To address these limitations, this paper proposes a framework for discovering anomalous sequences in attributed graphs. Instead of relying on frequency-based measures, the framework adopts an entropy-based method for pattern mining, thereby requiring at most one hyperparameter. Experimental results on real-world datasets demonstrate the effectiveness of the proposed approach in detecting anomalous sequences. Moreover, we extend the framework to applications in optics, where it is used to evaluate phase differences.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131467"},"PeriodicalIF":7.5,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122664","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}

引用次数: 0

Local sharpness aware minimization in decentralized federated learning with privacy protection 具有隐私保护的分散联邦学习中的局部锐度感知最小化

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-02 DOI: 10.1016/j.eswa.2026.131510

Jifei Hu , Yanli Li , Huayong Xie , Lijun Xu , Hang Zhang , Xinqiang Zhou

Federated learning (FL) enables distributed model training while preserving data privacy, but it still faces challenges from data heterogeneity and privacy constraints. Existing optimization methods aiming to flatten loss landscapes often fail to resolve inconsistencies between local and global flatness due to hyperparameter dependencies and centralized aggregation. Moreover, privacy-preserving techniques such as differential privacy (DP) can distort gradients, leading to sharper loss landscapes and hindered convergence. To tackle these issues, we propose DFedLSAM (Decentralized Federated Local Sharpness-Aware Minimization), a novel framework that eliminates the central server and uses the Sharpness-Aware Minimization (SAM) optimizer on the client side to maintain locally flattened loss landscapes. DFedLSAM adopts a dual-model architecture, where each client trains a sharing model for cross-client knowledge exchange and a private model updated via knowledge distillation (KD) from the sharing model’s soft logits, thereby reducing data heterogeneity and mitigating DP-induced sharpness. Building on this design, we introduce a perturbation-based SAM variant, integrated into the framework as DFedLSAM-Pert, which allocates perturbations according to layer-wise sensitivity and improves generalization without sacrificing privacy. Extensive experiments on benchmark image datasets and real-world medical datasets show that DFedLSAM and its perturbation-based extension DFedLSAM-Pert outperform existing baselines, especially in non-IID settings and under strict privacy budgets. These results indicate that DFedLSAM and DFedLSAM-Pert provide practical solutions for privacy-preserving FL in healthcare and other sensitive application domains.

联邦学习（FL）在保护数据隐私的同时支持分布式模型训练，但它仍然面临数据异构和隐私约束的挑战。现有的旨在平坦化损失景观的优化方法往往由于超参数依赖和集中式聚集而无法解决局部平坦度与全局平坦度之间的不一致性。此外，差分隐私（DP）等隐私保护技术会扭曲梯度，导致更清晰的损失景观并阻碍收敛。为了解决这些问题，我们提出了DFedLSAM（去中心化联邦本地锐度感知最小化），这是一个新的框架，它消除了中央服务器，并在客户端使用锐度感知最小化（SAM）优化器来维护本地平坦的损失景观。DFedLSAM采用双模型架构，每个客户端训练一个用于跨客户端知识交换的共享模型和一个通过共享模型软逻辑的知识蒸馏（KD）更新的私有模型，从而降低了数据异构性并减轻了dp引起的锐度。在此设计的基础上，我们引入了一个基于扰动的SAM变体，作为DFedLSAM-Pert集成到框架中，它根据分层灵敏度分配扰动，并在不牺牲隐私的情况下提高泛化。在基准图像数据集和真实医疗数据集上的大量实验表明，DFedLSAM及其基于扰动的扩展DFedLSAM- pert优于现有基线，特别是在非iid设置和严格的隐私预算下。这些结果表明，DFedLSAM和DFedLSAM- pert为医疗保健和其他敏感应用领域的隐私保护FL提供了实用的解决方案。

{"title":"Local sharpness aware minimization in decentralized federated learning with privacy protection","authors":"Jifei Hu , Yanli Li , Huayong Xie , Lijun Xu , Hang Zhang , Xinqiang Zhou","doi":"10.1016/j.eswa.2026.131510","DOIUrl":"10.1016/j.eswa.2026.131510","url":null,"abstract":"<div><div>Federated learning (FL) enables distributed model training while preserving data privacy, but it still faces challenges from data heterogeneity and privacy constraints. Existing optimization methods aiming to flatten loss landscapes often fail to resolve inconsistencies between local and global flatness due to hyperparameter dependencies and centralized aggregation. Moreover, privacy-preserving techniques such as differential privacy (DP) can distort gradients, leading to sharper loss landscapes and hindered convergence. To tackle these issues, we propose DFedLSAM (Decentralized Federated Local Sharpness-Aware Minimization), a novel framework that eliminates the central server and uses the Sharpness-Aware Minimization (SAM) optimizer on the client side to maintain locally flattened loss landscapes. DFedLSAM adopts a dual-model architecture, where each client trains a sharing model for cross-client knowledge exchange and a private model updated via knowledge distillation (KD) from the sharing model’s soft logits, thereby reducing data heterogeneity and mitigating DP-induced sharpness. Building on this design, we introduce a perturbation-based SAM variant, integrated into the framework as DFedLSAM-Pert, which allocates perturbations according to layer-wise sensitivity and improves generalization without sacrificing privacy. Extensive experiments on benchmark image datasets and real-world medical datasets show that DFedLSAM and its perturbation-based extension DFedLSAM-Pert outperform existing baselines, especially in non-IID settings and under strict privacy budgets. These results indicate that DFedLSAM and DFedLSAM-Pert provide practical solutions for privacy-preserving FL in healthcare and other sensitive application domains.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131510"},"PeriodicalIF":7.5,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122658","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}

引用次数: 0

Chrysanthemum image quality assessment via multi-scale feature fusion and meta-learning 基于多尺度特征融合和元学习的菊花图像质量评估

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-02 DOI: 10.1016/j.eswa.2026.131378

Shun Zhu , Xichen Yang , Tianshu Wang , Zhongyuan Mao , Yifan Chen , Jian Jiang , Hui Yan

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 https://github.com/dart-into/a-chrysanthemum-Screening-Method.

菊花产地溯源对保证菊花品质具有重要意义。随着计算机视觉技术的发展，将视觉技术应用于物体原点跟踪是可行的。这使得智能溯源成为可能，从而提高效率和准确性。然而，在收集菊花图像时，图像失真是不可避免的。这些畸变，如菊花组织不完整和角度差，往往会降低产地追踪的准确性。因此，准确测量图像质量，进而进一步提高原点跟踪的精度是非常重要的。为此，我们提出了一种菊花图像质量评价方法。首先，设计两步筛选（two-step screening， TSS）模块，筛选现有的适合于菊花畸变图像的经典畸变图像。其次，利用深度特征提取模块提取不同感受野层次的特征；第三，使用语义分析模块对不同特征的语义信息进行分析和融合。最后，设计了元学习框架来提高模型的准确性和鲁棒性。利用元学习获得的先验知识对模型进行少量采样微调。实验结果表明，该方法能够准确地判断不完全和角度畸变，从而有效地提高了原点跟踪的精度。我们的代码和模型可在https://github.com/dart-into/a-chrysanthemum-Screening-Method上获得。

{"title":"Chrysanthemum image quality assessment via multi-scale feature fusion and meta-learning","authors":"Shun Zhu , Xichen Yang , Tianshu Wang , Zhongyuan Mao , Yifan Chen , Jian Jiang , 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":"2026-02-02","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}

引用次数: 0

Local attention alignment fusion network for domain adaptive water body segmentation 区域自适应水体分割的局部注意力对齐融合网络

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-02 DOI: 10.1016/j.eswa.2026.131382

Hao Liu , Xiaobin Zhu , Xu Qizhi , Chun Yang , Hongyang Zhou , Yongjie Xia , Xucheng Yin

Water body segmentation is crucial for various tasks, e.g., disaster early warning and ecological management. Existing deep learning-based methods mainly focus on specific scenarios, often encountering significant performance drops on multi-source satellite data with large domain differences. In this paper, we propose a novel Local Attention Alignment Fusion Network for domain adaptive water body segmentation (dubbed LAAFNet). Our LAAFNet explores spatial relationships between pseudo-RGB and pseudo-NIR images, and then extracts invariant features via local attention to improve the representative capability of cross-domain features. Moreover, we design a novel Difficult Sample Point Loss (DSPLoss) to address the presence of potential positive samples within negative regions across domains through a pixel-level contrastive learning strategy. DSPLoss leverages a Cauchy-Schwarz-based constraint to regulate the upper bound of feature similarity in the pixel-level inner product space. This constraint enhances the separation between water bodies and background in hard samples, allowing the model to learn a clearer decision boundary and thereby improving its generalization capability. Notably, we construct a large-scale Water Generation Testing Dataset (WGTDataset) to evaluate water body segmentation in real-world applications. Experimental results demonstrate that the LAAFNet outperforms the state-of-the-art (SOTA) methods. The codes and dataset are available on: https://github.com/LH325/LAAFNet.

水体分割对于灾害预警、生态管理等工作至关重要。现有的基于深度学习的方法主要针对特定场景，在多源卫星数据域差异较大的情况下，往往会出现显著的性能下降。在本文中，我们提出了一种新的区域自适应水体分割的局部注意力对齐融合网络（LAAFNet）。我们的LAAFNet探索伪rgb和伪nir图像之间的空间关系，然后通过局部关注提取不变性特征，以提高跨域特征的代表能力。此外，我们设计了一种新的困难样本点损失（DSPLoss），通过像素级对比学习策略来解决跨域负区域内潜在正样本的存在。DSPLoss利用基于cauchy - schwarz的约束来调节像素级内积空间中特征相似度的上界。该约束增强了硬样本中水体与背景的分离，使模型能够学习到更清晰的决策边界，从而提高模型的泛化能力。值得注意的是，我们构建了一个大规模的水生成测试数据集（WGTDataset）来评估实际应用中的水体分割。实验结果表明，LAAFNet优于最先进的SOTA方法。代码和数据集可在https://github.com/LH325/LAAFNet上获得。

{"title":"Local attention alignment fusion network for domain adaptive water body segmentation","authors":"Hao Liu , Xiaobin Zhu , Xu Qizhi , Chun Yang , Hongyang Zhou , Yongjie Xia , Xucheng Yin","doi":"10.1016/j.eswa.2026.131382","DOIUrl":"10.1016/j.eswa.2026.131382","url":null,"abstract":"<div><div>Water body segmentation is crucial for various tasks, e.g., disaster early warning and ecological management. Existing deep learning-based methods mainly focus on specific scenarios, often encountering significant performance drops on multi-source satellite data with large domain differences. In this paper, we propose a novel Local Attention Alignment Fusion Network for domain adaptive water body segmentation (dubbed LAAFNet). Our LAAFNet explores spatial relationships between pseudo-RGB and pseudo-NIR images, and then extracts invariant features via local attention to improve the representative capability of cross-domain features. Moreover, we design a novel Difficult Sample Point Loss (DSPLoss) to address the presence of potential positive samples within negative regions across domains through a pixel-level contrastive learning strategy. DSPLoss leverages a Cauchy-Schwarz-based constraint to regulate the upper bound of feature similarity in the pixel-level inner product space. This constraint enhances the separation between water bodies and background in hard samples, allowing the model to learn a clearer decision boundary and thereby improving its generalization capability. Notably, we construct a large-scale Water Generation Testing Dataset (WGTDataset) to evaluate water body segmentation in real-world applications. Experimental results demonstrate that the LAAFNet outperforms the state-of-the-art (SOTA) methods. The codes and dataset are available on: <span><span>https://github.com/LH325/LAAFNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131382"},"PeriodicalIF":7.5,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122716","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}

引用次数: 0

Visible-guided multigranularity prompt learning for visible-infrared person re-identification 可见制导的多粒度提示学习，用于可见红外人再识别

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-02 DOI: 10.1016/j.eswa.2026.131464

Yangyan Luo, Ying Chen

Visible-infrared person re-identification (VI-ReID) remains challenging due to substantial cross-modal discrepancies and the absence of explicit semantic correspondence. This paper presents a novel Visible-Guided Multigranularity Prompt Learning (VG-MPL) framework that integrates semantic reasoning into cross-modal alignment through language-guided prompt learning. A fine-grained adaptive prompt is constructed by decomposing textual templates into learnable semantic slots, whose activations are dynamically modulated by a Prompt Slot Router (PSR) guided by visible features. This design enables sample-specific semantic modeling and enhances interpretability. To establish coherent cross-modal representations, a multi-granularity consistency constraint is imposed across the hierarchical layers of the CLIP text encoder, ensuring that global identity and local attribute semantics remain aligned. Furthermore, an Alternative Cross-Modal Alignment (ACMA) strategy and its theoretical analysis promotes bidirectional learning between visible and infrared modalities, improving optimization stability and preventing one-sided collapse. Extensive experiments on SYSU-MM01 and RegDB datasets demonstrate that VG-MPL achieves state-of-the-art performance and superior cross-modal generalization, validating the effectiveness of adaptive semantic prompting and hierarchical alignment in bridging the modality gap.

由于大量的跨模态差异和缺乏明确的语义对应，可见-红外人再识别（VI-ReID）仍然具有挑战性。本文提出了一种新的可视引导多粒度提示学习（VG-MPL）框架，通过语言引导提示学习将语义推理集成到跨模态对齐中。通过将文本模板分解为可学习的语义槽，构建了细粒度的自适应提示，这些语义槽的激活由提示槽路由器（PSR）根据可见特征进行动态调节。这种设计支持特定于示例的语义建模并增强可解释性。为了建立一致的跨模态表示，在CLIP文本编码器的分层层上施加了多粒度一致性约束，确保全局标识和本地属性语义保持一致。此外，ACMA策略及其理论分析促进了可见光和红外模态之间的双向学习，提高了优化稳定性，防止了单侧坍塌。在SYSU-MM01和RegDB数据集上的大量实验表明，VG-MPL实现了最先进的性能和卓越的跨模态泛化，验证了自适应语义提示和分层校准在弥合模态差距方面的有效性。

{"title":"Visible-guided multigranularity prompt learning for visible-infrared person re-identification","authors":"Yangyan Luo, Ying Chen","doi":"10.1016/j.eswa.2026.131464","DOIUrl":"10.1016/j.eswa.2026.131464","url":null,"abstract":"<div><div>Visible-infrared person re-identification (VI-ReID) remains challenging due to substantial cross-modal discrepancies and the absence of explicit semantic correspondence. This paper presents a novel Visible-Guided Multigranularity Prompt Learning (VG-MPL) framework that integrates semantic reasoning into cross-modal alignment through language-guided prompt learning. A fine-grained adaptive prompt is constructed by decomposing textual templates into learnable semantic slots, whose activations are dynamically modulated by a Prompt Slot Router (PSR) guided by visible features. This design enables sample-specific semantic modeling and enhances interpretability. To establish coherent cross-modal representations, a multi-granularity consistency constraint is imposed across the hierarchical layers of the CLIP text encoder, ensuring that global identity and local attribute semantics remain aligned. Furthermore, an Alternative Cross-Modal Alignment (ACMA) strategy and its theoretical analysis promotes bidirectional learning between visible and infrared modalities, improving optimization stability and preventing one-sided collapse. Extensive experiments on SYSU-MM01 and RegDB datasets demonstrate that VG-MPL achieves state-of-the-art performance and superior cross-modal generalization, validating the effectiveness of adaptive semantic prompting and hierarchical alignment in bridging the modality gap.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131464"},"PeriodicalIF":7.5,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122750","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}

引用次数: 0

A heterogeneous Hopfield neural network with discrete memristor: modeling, dynamics, and application in medical image encryption 具有离散忆阻器的异构Hopfield神经网络：建模、动态和在医学图像加密中的应用

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2026-02-01 DOI: 10.1016/j.eswa.2026.131457

Huiqun Zou , Yang Lu , Wenjiao Li , Wenhui Li , Xiuli Chai

Memristors and activation functions critically shape the nonlinear dynamics of Hopfield neural networks. While previous studies explored memristor modeling and heterogeneous activation separately, their combination to form heterogeneous memristive networks remains insufficiently explored. This paper bridges this gap by proposing a novel Heterogeneous Hopfield Neural Network with Discrete Memristor (HHNN-DM), coupling a discrete memristor with heterogeneous activations to mimic neural diversity. By analyzing dissipation, equilibrium stability, bifurcation diagrams, and Lyapunov exponents, we demonstrate that heterogeneous activation mechanisms significantly enhance network complexity and unpredictability under memristive interactions. This synergy gives rise to rich chaotic behaviors, such as periodic orbits, bifurcations, transient chaos, and chaotic bursting. As these biologically inspired chaotic dynamics, the resulting high-quality chaotic sequences are well suited for cryptographic applications. Furthermore, a Heterogeneous Hopfield Neural Network–based Medical Image Encryption Algorithm (HHNN-MIEA) is developed to enhance security in remote medical image transmission, integrating an X-fractal curve sorting matrix for permutation with multi-logical diffusion driven by chaotic sequences. Experimental results verify that the HHNN-MIEA achieves high security in aspects such as key sensitivity, information entropy without compromising efficiency, highlighting its effectiveness, robustness and reliable solution for secure medical image transmission.

记忆电阻器和激活函数是Hopfield神经网络非线性动力学的关键。虽然以前的研究分别探讨了忆阻器建模和异质激活，但它们的组合形成异质忆阻网络的探索仍然不够充分。本文通过提出一种具有离散忆阻器的新型异质Hopfield神经网络（HHNN-DM）来弥补这一差距，将离散忆阻器与异质激活耦合起来以模拟神经多样性。通过对耗散、平衡稳定性、分岔图和Lyapunov指数的分析，我们证明了在记忆相互作用下，异质激活机制显著提高了网络的复杂性和不可预测性。这种协同作用产生了丰富的混沌行为，如周期轨道、分岔、瞬态混沌和混沌爆发。由于这些受生物学启发的混沌动力学，由此产生的高质量混沌序列非常适合密码学应用。在此基础上，提出了一种基于异构Hopfield神经网络的医学图像加密算法（HHNN-MIEA），将x分形曲线排序矩阵与混沌序列驱动的多逻辑扩散相结合，提高了医学图像远程传输的安全性。实验结果表明，在不影响效率的前提下，HHNN-MIEA在密钥灵敏度、信息熵等方面实现了较高的安全性，突出了其有效性、鲁棒性和可靠的医学图像安全传输解决方案。

{"title":"A heterogeneous Hopfield neural network with discrete memristor: modeling, dynamics, and application in medical image encryption","authors":"Huiqun Zou , Yang Lu , Wenjiao Li , Wenhui Li , Xiuli Chai","doi":"10.1016/j.eswa.2026.131457","DOIUrl":"10.1016/j.eswa.2026.131457","url":null,"abstract":"<div><div>Memristors and activation functions critically shape the nonlinear dynamics of Hopfield neural networks. While previous studies explored memristor modeling and heterogeneous activation separately, their combination to form heterogeneous memristive networks remains insufficiently explored. This paper bridges this gap by proposing a novel Heterogeneous Hopfield Neural Network with Discrete Memristor (HHNN-DM), coupling a discrete memristor with heterogeneous activations to mimic neural diversity. By analyzing dissipation, equilibrium stability, bifurcation diagrams, and Lyapunov exponents, we demonstrate that heterogeneous activation mechanisms significantly enhance network complexity and unpredictability under memristive interactions. This synergy gives rise to rich chaotic behaviors, such as periodic orbits, bifurcations, transient chaos, and chaotic bursting. As these biologically inspired chaotic dynamics, the resulting high-quality chaotic sequences are well suited for cryptographic applications. Furthermore, a Heterogeneous Hopfield Neural Network–based Medical Image Encryption Algorithm (HHNN-MIEA) is developed to enhance security in remote medical image transmission, integrating an X-fractal curve sorting matrix for permutation with multi-logical diffusion driven by chaotic sequences. Experimental results verify that the HHNN-MIEA achieves high security in aspects such as key sensitivity, information entropy without compromising efficiency, highlighting its effectiveness, robustness and reliable solution for secure medical image transmission.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131457"},"PeriodicalIF":7.5,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122715","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}

引用次数: 0