Internet of Things最新文献

Bibliometric analysis of secure IoT for quantum computing 量子计算安全物联网的文献计量分析

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2026-01-16 DOI: 10.1016/j.iot.2026.101872

Hamza Ibrahim , Love Allen Chijioke Ahakonye , Jae-Min Lee , Dong-Seong Kim

The convergence of Quantum Machine Learning (QML) and Blockchain is emerging as a transformative paradigm to address escalating security and scalability challenges in 6G-enabled Industrial Internet of Things (IIoT) networks. This study presents the first comprehensive bibliometric and meta-analysis of this nascent interdisciplinary field. We analyzed 159 peer-reviewed publications (indexed from January 2022 through December 22, 2024) from Scopus, employing a systematic Kitchenham-based methodology for literature selection and VOSviewer for science mapping. Our analysis reveals a 75% annual growth rate since 2022, with India (37.7%), the USA (12.6%), and South Korea (12.6%) as the leading contributors. Keyword co-occurrence analysis identified four dominant thematic clusters: “6G Network Security,” “Quantum Computing and AI,” “Blockchain and Decentralization,” and “IIoT Applications.” The study’s novelty lies in synthesizing bibliometric insights with a proposed five-layer QML-Blockchain integration framework and a comparative analysis against existing reviews. Quantitative performance metrics indicate that QML can improve anomaly detection accuracy by 5–9% over classical models, while advanced consensus mechanisms like PoA² can reduce transaction latency by 35%. However, significant challenges persist, including quantum hardware limitations (e.g., qubit coherence  < 100 μs), scalability challenges in achieving consensus across massive IIoT device densities, and a critical lack of empirical testbeds. This research provides a foundational roadmap, emphasizing the urgent need for standardized benchmarks, hybrid orchestration models, and quantum-resistant cryptography to realize secure, intelligent, and autonomous IIoT ecosystems in the 6G era.

量子机器学习（QML）和区块链的融合正在成为一种变革范例，以解决支持6g的工业物联网（IIoT）网络中不断升级的安全性和可扩展性挑战。本研究首次对这一新兴的跨学科领域进行了全面的文献计量和荟萃分析。我们分析了来自Scopus的159篇同行评审的出版物（从2022年1月到2024年12月22日），采用了基于kitchenham的系统文献选择方法和VOSviewer的科学制图方法。我们的分析显示，自2022年以来，年增长率为75%，其中印度（37.7%）、美国（12.6%）和韩国（12.6%）是主要贡献者。关键词共现分析确定了四个主要主题集群：“6G网络安全”、“量子计算与人工智能”、“区块链与去中心化”和“工业物联网应用”。该研究的新颖之处在于将文献计量学的见解与提出的五层qml -区块链集成框架相结合，并与现有综述进行比较分析。定量性能指标表明，与经典模型相比，QML可以将异常检测准确率提高5-9%，而像PoA2这样的高级共识机制可以将事务延迟减少35%。然而，重大挑战仍然存在，包括量子硬件限制（例如，量子比特相干性 <； 100 μs），在大规模工业物联网设备密度上达成共识的可扩展性挑战，以及经验测试平台的严重缺乏。本研究提供了一个基本路线图，强调了在6G时代实现安全、智能和自主的工业物联网生态系统对标准化基准、混合编排模型和抗量子加密的迫切需求。

{"title":"Bibliometric analysis of secure IoT for quantum computing","authors":"Hamza Ibrahim , Love Allen Chijioke Ahakonye , Jae-Min Lee , Dong-Seong Kim","doi":"10.1016/j.iot.2026.101872","DOIUrl":"10.1016/j.iot.2026.101872","url":null,"abstract":"<div><div>The convergence of Quantum Machine Learning (QML) and Blockchain is emerging as a transformative paradigm to address escalating security and scalability challenges in 6G-enabled Industrial Internet of Things (IIoT) networks. This study presents the first comprehensive bibliometric and meta-analysis of this nascent interdisciplinary field. We analyzed 159 peer-reviewed publications (indexed from January 2022 through December 22, 2024) from Scopus, employing a systematic Kitchenham-based methodology for literature selection and VOSviewer for science mapping. Our analysis reveals a 75% annual growth rate since 2022, with India (37.7%), the USA (12.6%), and South Korea (12.6%) as the leading contributors. Keyword co-occurrence analysis identified four dominant thematic clusters: “6G Network Security,” “Quantum Computing and AI,” “Blockchain and Decentralization,” and “IIoT Applications.” The study’s novelty lies in synthesizing bibliometric insights with a proposed five-layer QML-Blockchain integration framework and a comparative analysis against existing reviews. Quantitative performance metrics indicate that QML can improve anomaly detection accuracy by 5–9% over classical models, while advanced consensus mechanisms like PoA<sup>2</sup> can reduce transaction latency by 35%. However, significant challenges persist, including quantum hardware limitations (e.g., qubit coherence  < 100 <em>μ</em>s), scalability challenges in achieving consensus across massive IIoT device densities, and a critical lack of empirical testbeds. This research provides a foundational roadmap, emphasizing the urgent need for standardized benchmarks, hybrid orchestration models, and quantum-resistant cryptography to realize secure, intelligent, and autonomous IIoT ecosystems in the 6G era.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"36 ","pages":"Article 101872"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077991","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

FedWKD: Federated learning weighted aggregation with knowledge distillation for IoT forecasting FedWKD：基于知识蒸馏的物联网预测的联邦学习加权聚合

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2025-12-11 DOI: 10.1016/j.iot.2025.101849

Bouchra Fakher , Mohamed El Amine Brahmia , Ismail Bennis , Abdelhafid Abouaissa

Federated Learning (FL) has emerged as a promising solution for decentralized Machine Learning (ML) that does not have direct access to datasets in a centralized manner. However, the traditional FL methods are prone to overfitting and model drift at the client level and server divergence during classic aggregation in case of heterogeneous, non-independent and identically distributed (non-IID) time-series sensor data. In this paper, we propose a novel approach that integrates bidirectional Knowledge Distillation (KD) by using distilled soft predictions of each client model, called logits, as well as server model distilled logits. Specifically, clients use KD regularization techniques using the received server logits during model training, while the server uses received client logits to build a score for weighted global aggregation each round. Thus, we avoid local training overhead for clients, while also improving global aggregation using weighting on the server-side for each training round for non-IID data. Experimental results highlight its ability to improve forecasting metrics compared to other methods such as CADIS and FEDGKD, using loss, error, and execution time metrics, hence bettering generalization and minimizing client drift and bias.

联邦学习（FL）已经成为去中心化机器学习（ML）的一个有前途的解决方案，它不能以集中的方式直接访问数据集。然而，对于异构、非独立、同分布（non-IID）的时间序列传感器数据，传统的FL方法在经典聚合过程中容易出现客户端的过拟合和模型漂移，而服务器端的发散。在本文中，我们提出了一种新的方法，通过使用每个客户端模型（称为logits）的蒸馏软预测以及服务器模型蒸馏logits来集成双向知识蒸馏（KD）。具体来说，客户端在模型训练期间使用接收到的服务器logit使用KD正则化技术，而服务器则使用接收到的客户端logit为每轮加权全局聚合构建分数。因此，我们避免了客户端的本地训练开销，同时还在服务器端使用非iid数据的每个训练轮的加权来改进全局聚合。实验结果表明，与其他方法（如CADIS和FEDGKD）相比，它能够使用损失、误差和执行时间指标来改进预测指标，从而更好地泛化并最大限度地减少客户端漂移和偏差。

{"title":"FedWKD: Federated learning weighted aggregation with knowledge distillation for IoT forecasting","authors":"Bouchra Fakher , Mohamed El Amine Brahmia , Ismail Bennis , Abdelhafid Abouaissa","doi":"10.1016/j.iot.2025.101849","DOIUrl":"10.1016/j.iot.2025.101849","url":null,"abstract":"<div><div>Federated Learning (FL) has emerged as a promising solution for decentralized Machine Learning (ML) that does not have direct access to datasets in a centralized manner. However, the traditional FL methods are prone to overfitting and model drift at the client level and server divergence during classic aggregation in case of heterogeneous, non-independent and identically distributed (non-IID) time-series sensor data. In this paper, we propose a novel approach that integrates bidirectional Knowledge Distillation (KD) by using distilled soft predictions of each client model, called logits, as well as server model distilled logits. Specifically, clients use KD regularization techniques using the received server logits during model training, while the server uses received client logits to build a score for weighted global aggregation each round. Thus, we avoid local training overhead for clients, while also improving global aggregation using weighting on the server-side for each training round for non-IID data. Experimental results highlight its ability to improve forecasting metrics compared to other methods such as CADIS and FEDGKD, using loss, error, and execution time metrics, hence bettering generalization and minimizing client drift and bias.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"36 ","pages":"Article 101849"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791909","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

An edge-intelligent three-tier framework for real-time forest fire detection, integrating WSNs, WMSNs, and UAVs 一种边缘智能三层框架，用于实时森林火灾探测，集成了wsn、wmsn和无人机

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2025-12-29 DOI: 10.1016/j.iot.2025.101861

Hamzeh Abu Ali , Enver Ever , Burak Kizilkaya , Muhammad Toaha Raza Khan , Masood Ur Rehman , Shuja Ansari , Muhammad Ali Imran , Adnan Yazici

Forest fires are becoming prevalent, threatening ecosystems, economies, and public safety while creating an urgent demand for rapid and reliable detection systems. Conventional approaches such as watchtowers, manual patrols, and satellite imaging suffer from limited coverage, delays, and inadequate precision. To address these challenges, we propose a three-tier, edge-centric framework that integrates wireless sensor networks (WSNs), wireless multimedia sensor networks (WMSNs), unmanned aerial vehicles (UAVs), and lightweight machine learning (ML) and deep learning (DL) models for efficient detection. In the first tier, scalar sensors provide early hazard identification; in the second, smart sensors execute a lightweight ML model for intermediate verification, achieving a 94% F1-score with a minimal feature set; and in the third, UAVs equipped with sensors, cameras, and a compact convolutional neural network (CNN) deliver final confirmation. The CNN achieves state-of-the-art results with a 100% F1 score on the FireMan-UAV-RGBT dataset and 99.5% on UAV-FFDB while remaining compact (1.6 MB) and efficient (157 ms inference on Raspberry Pi 5), enabling real-time edge deployment. Simulations show reduced end-to-end delay (813.59 ms) compared to WSN-only (865.84 ms) and WMSN (1066.18 ms) baselines, improved throughput (7.05 kbps vs 3.80 kbps and 3.06 kbps), and a 100% delivery ratio. Real-world WSN testbed experiments further validate the framework, achieving a 97% delivery ratio, 144.39 ms latency (vs. 258.37 ms in simulations), and energy consumption of 0.0559 J/s (closely matching 0.0442 J/s in simulations). These results collectively demonstrate the practicality and effectiveness of the framework for real-time forest fire monitoring and rapid emergency response.

森林火灾越来越普遍，威胁着生态系统、经济和公共安全，同时迫切需要快速可靠的检测系统。传统的方法，如瞭望塔、人工巡逻和卫星成像，受到覆盖范围有限、延迟和精度不足的困扰。为了应对这些挑战，我们提出了一个三层、以边缘为中心的框架，该框架集成了无线传感器网络（wsn）、无线多媒体传感器网络（wmsn）、无人机（uav）以及轻量级机器学习（ML）和深度学习（DL）模型，以实现高效检测。在第一层，标量传感器提供早期危险识别；其次，智能传感器执行轻量级ML模型进行中间验证，以最小的特征集实现94%的f1得分；第三，配备传感器、摄像头和紧凑卷积神经网络（CNN）的无人机提供最终确认。CNN在FireMan-UAV-RGBT数据集上获得了100%的F1分数，在UAV-FFDB上获得了99.5%的分数，同时保持了紧凑（1.6 MB）和高效（在Raspberry Pi 5上推断157 ms），实现了实时边缘部署。仿真显示，与纯wsn （865.84 ms）和WMSN （1066.18 ms）基线相比，端到端延迟（813.59 ms）减少，吞吐量提高（7.05 kbps vs 3.80 kbps和3.06 kbps），传输率达到100%。真实WSN测试平台实验进一步验证了该框架，实现了97%的传输率、144.39 ms的延迟（模拟为258.37 ms）和0.0559 J/s的能耗（与模拟中的0.0442 J/s非常接近）。这些结果共同证明了该框架在森林火灾实时监测和快速应急响应方面的实用性和有效性。

{"title":"An edge-intelligent three-tier framework for real-time forest fire detection, integrating WSNs, WMSNs, and UAVs","authors":"Hamzeh Abu Ali , Enver Ever , Burak Kizilkaya , Muhammad Toaha Raza Khan , Masood Ur Rehman , Shuja Ansari , Muhammad Ali Imran , Adnan Yazici","doi":"10.1016/j.iot.2025.101861","DOIUrl":"10.1016/j.iot.2025.101861","url":null,"abstract":"<div><div>Forest fires are becoming prevalent, threatening ecosystems, economies, and public safety while creating an urgent demand for rapid and reliable detection systems. Conventional approaches such as watchtowers, manual patrols, and satellite imaging suffer from limited coverage, delays, and inadequate precision. To address these challenges, we propose a three-tier, edge-centric framework that integrates wireless sensor networks (WSNs), wireless multimedia sensor networks (WMSNs), unmanned aerial vehicles (UAVs), and lightweight machine learning (ML) and deep learning (DL) models for efficient detection. In the first tier, scalar sensors provide early hazard identification; in the second, smart sensors execute a lightweight ML model for intermediate verification, achieving a 94% F1-score with a minimal feature set; and in the third, UAVs equipped with sensors, cameras, and a compact convolutional neural network (CNN) deliver final confirmation. The CNN achieves state-of-the-art results with a 100% F1 score on the FireMan-UAV-RGBT dataset and 99.5% on UAV-FFDB while remaining compact (1.6 MB) and efficient (157 ms inference on Raspberry Pi 5), enabling real-time edge deployment. Simulations show reduced end-to-end delay (813.59 ms) compared to WSN-only (865.84 ms) and WMSN (1066.18 ms) baselines, improved throughput (7.05 kbps vs 3.80 kbps and 3.06 kbps), and a 100% delivery ratio. Real-world WSN testbed experiments further validate the framework, achieving a 97% delivery ratio, 144.39 ms latency (vs. 258.37 ms in simulations), and energy consumption of 0.0559 J/s (closely matching 0.0442 J/s in simulations). These results collectively demonstrate the practicality and effectiveness of the framework for real-time forest fire monitoring and rapid emergency response.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"36 ","pages":"Article 101861"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885164","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Medvault: Privacy-enhancing medical record retrieval for ioMT-Enabled healthcare with query-pattern protection Medvault：为支持iomt的医疗保健提供具有查询模式保护的增强隐私的医疗记录检索

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2026-02-11 DOI: 10.1016/j.iot.2026.101890

Yuxi Li , Dong Ji , Fa Zhu , Xingchi Chen , Athanasios V. Vasilakos , Francesco Piccialli , David Camacho

The Internet of Medical Things (IoMT) increasingly relies on remote EHR repositories to answer contextual clinical queries that combine structured signals, free-text notes, and medical images. Outsourcing retrieval to a cloud provider, however, risks exposing not only query content but also query patterns—which index regions and records are accessed and how often—enabling sensitive inference even if payloads are encrypted. We present MedVault, a privacy-enhancing multimodal EHR retrieval system that protects both query content and query pattern under a split-trust two-server model. MedVault computes multimodal embeddings at the authorized client boundary, secret-shares embeddings and record payloads across two non-colluding servers, privately selects candidate clusters via distributed point functions (DPF), and retrieves padded top-k results so that transferred volumes are independent of the target. A clustering index confines secure scoring to a sublinear candidate set and enables a constant-round (1-RTT) query protocol. On de-identified MIMIC-IV datasets, MedVault not only achieves stable retrieval quality with R@10  ≈  0.83–0.84 and nDCG@10  ≈  0.85–0.86, but also improves over a keyword baseline by about 18%–19%. These results suggest that MedVault offers a deployable building block for privacy-preserving clinical retrieval in bandwidth- and latency-sensitive IoMT settings.

医疗物联网（IoMT）越来越依赖远程电子病历存储库来回答结合结构化信号、自由文本注释和医学图像的上下文临床查询。但是，将检索外包给云提供商，不仅会暴露查询内容，还会暴露查询模式（访问哪些索引区域和记录以及访问频率），即使对有效负载进行了加密，也会启用敏感推断。我们提出MedVault，一个增强隐私的多模式EHR检索系统，在分离信任的双服务器模型下保护查询内容和查询模式。MedVault在授权客户端边界计算多模态嵌入，在两个非串通服务器上秘密共享嵌入和记录有效负载，通过分布式点函数（DPF）私下选择候选集群，并检索填充的top-k结果，以便传输的卷与目标无关。聚类索引将安全评分限制在次线性候选集，并启用恒定轮询（1-RTT）查询协议。在去识别的MIMIC-IV数据集上，MedVault不仅获得了R@10 ≈ 0.83-0.84和nDCG@10 ≈ 0.85-0.86的稳定检索质量，而且在关键词基线上提高了约18%-19%。这些结果表明，MedVault为带宽和延迟敏感的IoMT设置中保护隐私的临床检索提供了可部署的构建块。

{"title":"Medvault: Privacy-enhancing medical record retrieval for ioMT-Enabled healthcare with query-pattern protection","authors":"Yuxi Li , Dong Ji , Fa Zhu , Xingchi Chen , Athanasios V. Vasilakos , Francesco Piccialli , David Camacho","doi":"10.1016/j.iot.2026.101890","DOIUrl":"10.1016/j.iot.2026.101890","url":null,"abstract":"<div><div>The Internet of Medical Things (IoMT) increasingly relies on remote EHR repositories to answer contextual clinical queries that combine structured signals, free-text notes, and medical images. Outsourcing retrieval to a cloud provider, however, risks exposing not only query content but also query patterns—which index regions and records are accessed and how often—enabling sensitive inference even if payloads are encrypted. We present <span>MedVault</span>, a privacy-enhancing multimodal EHR retrieval system that protects both query content and query pattern under a split-trust two-server model. <span>MedVault</span> computes multimodal embeddings at the authorized client boundary, secret-shares embeddings and record payloads across two non-colluding servers, privately selects candidate clusters via distributed point functions (DPF), and retrieves padded top-<em>k</em> results so that transferred volumes are independent of the target. A clustering index confines secure scoring to a sublinear candidate set and enables a constant-round (1-RTT) query protocol. On de-identified MIMIC-IV datasets, <span>MedVault</span> not only achieves stable retrieval quality with R@10  ≈  0.83–0.84 and nDCG@10  ≈  0.85–0.86, but also improves over a keyword baseline by about 18%–19%. These results suggest that <span>MedVault</span> offers a deployable building block for privacy-preserving clinical retrieval in bandwidth- and latency-sensitive IoMT settings.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"36 ","pages":"Article 101890"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146173165","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Transformer-based classification of IoT network traffic with flow-to-window aggregation 基于变压器的物联网网络流量分类与流到窗口聚合

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2026-01-21 DOI: 10.1016/j.iot.2026.101879

Sergio Martin-Reizabal , Adrian Caballero-Quiroga , Beatriz Gil-Arroyo , Nuño Basurto , Ruben Ruiz-Gonzalez

The explosive growth of the IoT has led to an increasingly complex and heterogeneous network traffic, posing major challenges for intrusion detection. Most existing machine learning and deep learning approaches model network traffic at the level of individual flows, which limits their ability to capture contextual relationships among concurrent communications. This paper introduces a Transformer-based framework for IoT intrusion detection that aggregates network flows into fixed-duration windows and treats each flow as a token within the input sequence. The self-attention mechanism captures contextual relationships among concurrent flows, enabling effective modeling of temporal dependencies without recurrence. Experiments conducted on the CICIoT2023 dataset show that the proposed model achieves a weighted F1-score of 97.9% and a macro ROC–AUC of 99.6% under temporally blocked cross-validation, while maintaining high computational efficiency. These results demonstrate that flow-to-window aggregation combined with self-attention provides a robust and scalable foundation for IoT network security, suitable for deployment in edge and smart-home environments.

物联网的爆炸式增长导致网络流量日益复杂和异构，对入侵检测提出了重大挑战。大多数现有的机器学习和深度学习方法都是在单个流的层面上对网络流量进行建模，这限制了它们捕捉并发通信之间上下文关系的能力。本文介绍了一种基于transformer的物联网入侵检测框架，该框架将网络流聚合到固定持续时间的窗口中，并将每个流视为输入序列中的令牌。自关注机制捕获并发流之间的上下文关系，从而能够有效地对时间依赖性进行建模，而不会重复出现。在CICIoT2023数据集上进行的实验表明，在时间阻塞交叉验证下，该模型的加权f1得分为97.9%，宏观ROC-AUC为99.6%，同时保持了较高的计算效率。这些结果表明，流到窗口聚合与自关注相结合，为物联网网络安全提供了强大且可扩展的基础，适合部署在边缘和智能家居环境中。

{"title":"Transformer-based classification of IoT network traffic with flow-to-window aggregation","authors":"Sergio Martin-Reizabal , Adrian Caballero-Quiroga , Beatriz Gil-Arroyo , Nuño Basurto , Ruben Ruiz-Gonzalez","doi":"10.1016/j.iot.2026.101879","DOIUrl":"10.1016/j.iot.2026.101879","url":null,"abstract":"<div><div>The explosive growth of the IoT has led to an increasingly complex and heterogeneous network traffic, posing major challenges for intrusion detection. Most existing machine learning and deep learning approaches model network traffic at the level of individual flows, which limits their ability to capture contextual relationships among concurrent communications. This paper introduces a Transformer-based framework for IoT intrusion detection that aggregates network flows into fixed-duration windows and treats each flow as a token within the input sequence. The self-attention mechanism captures contextual relationships among concurrent flows, enabling effective modeling of temporal dependencies without recurrence. Experiments conducted on the CICIoT2023 dataset show that the proposed model achieves a weighted F1-score of 97.9% and a macro ROC–AUC of 99.6% under temporally blocked cross-validation, while maintaining high computational efficiency. These results demonstrate that flow-to-window aggregation combined with self-attention provides a robust and scalable foundation for IoT network security, suitable for deployment in edge and smart-home environments.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"36 ","pages":"Article 101879"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077990","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Neuromorphic solar edge AI for sustainable wildfire detection 用于可持续野火探测的神经形态太阳边缘AI

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2025-12-29 DOI: 10.1016/j.iot.2025.101862

Raúl Parada

This paper presents a feasibility study of a solar-autonomous wildfire detection system using neuromorphic edge AI on fixed-wing drones. Through a comprehensive year-long simulation over Parc del Garraf (Catalonia), we evaluate three edge computing platforms, Raspberry Pi 4, Google Coral TPU, and BrainChip Akida, integrated into solar-optimized eBee X drones. Results show that the BrainChip Akida achieves 4200 patrol hrs per yr, nearly three times that of traditional CPU systems, while maintaining 87 % solar energy autonomy. The Google Coral TPU and Raspberry Pi 4 reach 66 % and 52 % autonomy, respectively. Fleet scaling analysis demonstrates that increasing drone count from one to eight reduces median wildfire detection time from 18 to 2.2 hrs, surpassing critical response thresholds. Seasonal analysis reveals Akida-based systems can operate fully on solar energy during summer and most of spring and fall, minimizing grid dependency. These findings establish neuromorphic computing as a foundational technology for sustainable, perpetual environmental monitoring within the Internet of Robotic Things (IoRT).

本文研究了一种基于神经形态边缘人工智能的太阳能自主野火探测系统在固定翼无人机上的可行性。通过对Parc del Garraf （Catalonia）进行为期一年的全面模拟，我们评估了三个边缘计算平台，树莓派4，谷歌Coral TPU和BrainChip Akida，集成到太阳能优化的eBee X无人机中。结果表明，BrainChip Akida实现了每年4200小时的巡逻时间，几乎是传统CPU系统的三倍，同时保持了87%的太阳能自主性。谷歌Coral TPU和Raspberry Pi 4分别达到66%和52%的自主性。机队规模分析表明，将无人机数量从1架增加到8架，将野火探测时间的中位数从18小时减少到2.2小时，超过了关键响应阈值。季节性分析显示，秋田系统可以在夏季和春季和秋季的大部分时间完全依靠太阳能运行，从而最大限度地减少对电网的依赖。这些发现确立了神经形态计算作为机器人物联网（IoRT）中可持续、永久环境监测的基础技术。

引用次数: 0

Stability analysis for fog computing via Lyapunov function 基于Lyapunov函数的雾计算稳定性分析

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2026-01-12 DOI: 10.1016/j.iot.2026.101871

Shih-Yu Chang , Mayank Kapadia , Peishun Yan , Wei Duan

Fog computing is an emerging paradigm for the Internet of Things (IoT), where system stability directly impacts reliability, performance, and user experience. Existing stability models often ignore application-level service completion or fail to capture dynamic interactions among sensors and fog nodes. This study addresses these gaps by establishing necessary conditions for fog nodes to process application services within finite time. First, we introduce a fluid model based on a partial differential equation (PDE) to quantify the dynamics of service counts for each sensor when fog nodes are shared. Second, we design a Lyapunov function derived from the PDE solution to analyze system stability and convergence. Third, we apply this Lyapunov function to derive conditions that guarantee timely service completion. Finally, numerical experiments validate the fluid model, investigate PDE solution behavior, and assess the convergence speed of the Lyapunov function under various system parameters. These results provide actionable insights for ensuring stability and efficiency in fog computing systems for IoT applications.

雾计算是物联网（IoT）的一种新兴范例，系统稳定性直接影响可靠性、性能和用户体验。现有的稳定性模型经常忽略应用程序级的服务完成，或者无法捕获传感器和雾节点之间的动态交互。本研究通过建立雾节点在有限时间内处理应用服务的必要条件来解决这些差距。首先，我们引入了一个基于偏微分方程（PDE）的流体模型来量化共享雾节点时每个传感器的服务计数动态。其次，我们设计了一个由PDE解导出的Lyapunov函数来分析系统的稳定性和收敛性。第三，我们应用这个李雅普诺夫函数来推导保证服务及时完成的条件。最后，通过数值实验验证了流体模型，研究了PDE解行为，并评估了不同系统参数下Lyapunov函数的收敛速度。这些结果为确保物联网应用的雾计算系统的稳定性和效率提供了可操作的见解。

{"title":"Stability analysis for fog computing via Lyapunov function","authors":"Shih-Yu Chang , Mayank Kapadia , Peishun Yan , Wei Duan","doi":"10.1016/j.iot.2026.101871","DOIUrl":"10.1016/j.iot.2026.101871","url":null,"abstract":"<div><div>Fog computing is an emerging paradigm for the Internet of Things (IoT), where system stability directly impacts reliability, performance, and user experience. Existing stability models often ignore application-level service completion or fail to capture dynamic interactions among sensors and fog nodes. This study addresses these gaps by establishing necessary conditions for fog nodes to process application services within finite time. First, we introduce a fluid model based on a partial differential equation (PDE) to quantify the dynamics of service counts for each sensor when fog nodes are shared. Second, we design a Lyapunov function derived from the PDE solution to analyze system stability and convergence. Third, we apply this Lyapunov function to derive conditions that guarantee timely service completion. Finally, numerical experiments validate the fluid model, investigate PDE solution behavior, and assess the convergence speed of the Lyapunov function under various system parameters. These results provide actionable insights for ensuring stability and efficiency in fog computing systems for IoT applications.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"36 ","pages":"Article 101871"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146023231","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Classifying user-created passwords using machine learning and natural language processing techniques 使用机器学习和自然语言处理技术对用户创建的密码进行分类

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2025-12-13 DOI: 10.1016/j.iot.2025.101854

Binh Le Thanh Thai, Tsubasa Takii, Hidema Tanaka

Passwords are the dominant authentication method. However, evaluating the strength of user-created passwords remains a significant challenge due to the influence of various external factors, such as language, culture, and keyboard layout. In this paper, we address the problem of classifying user-created passwords into predefined groups, rather than directly evaluating their strength. First, we assess the performance of classifiers utilizing eight machine learning (ML) algorithms and four Natural Language Processing techniques to identify the optimal combination of ML algorithms and feature extraction methods. Through this experiment, we determine that the classifier combining Bag-of-Words and Logistic Regression is the most effective approach for classifying user-created passwords. Subsequently, we propose a hierarchical classification model to enhance the performance of this classifier. Experimental results demonstrate that the proposed model achieves accuracy of 97.81 % and recall of 99.66 % for weak passwords.

密码是主要的认证方法。然而，由于语言、文化和键盘布局等各种外部因素的影响，评估用户创建的密码的强度仍然是一个重大挑战。在本文中，我们解决了将用户创建的密码分类到预定义组的问题，而不是直接评估它们的强度。首先，我们评估了使用八种机器学习（ML）算法和四种自然语言处理技术的分类器的性能，以确定ML算法和特征提取方法的最佳组合。通过本实验，我们确定结合词袋和逻辑回归的分类器是对用户创建的密码进行分类的最有效方法。随后，我们提出了一种层次分类模型来提高该分类器的性能。实验结果表明，该模型对弱密码的识别率为97.81%，召回率为99.66%。

引用次数: 0

FedMamba: Robust multimodal federated intrusion detection for heterogeneous IoT systems FedMamba：针对异构物联网系统的鲁棒多模态联邦入侵检测

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2026-01-20 DOI: 10.1016/j.iot.2026.101877

Hafiz Bilal Ahmad , Haichang Gao , Naila Latif , Tanjeena Manzoor

The convergence of Information Technology (IT) and Operational Technology (OT) in Industry 4.0 produces diverse data streams, such as system logs, sensor readings, and network traffic, which are vital for industrial security. However, existing security analytics are siloed by modality and rely on centralized processing, raising concerns regarding privacy, latency, and scalability. Although Federated Learning (FL) mitigates privacy risks, most frameworks remain unimodal, lack support for non-IID data distributions, and face adversarial evasion challenges. We propose FedMamba, a novel multimodal Federated Learning (MMFL) framework that creates a unified Mamba-based model to address these issues via (i) efficient cross-modal learning, (ii) a FedProx-based protocol for stable non-IID training that remains compatible with secure aggregation, and (iii) modality-specific adversarial training for robustness. Experiments on HDFS, SWaT, and CICIoMT-2024 datasets show that the standard FedMamba achieved competitive macro F1-scores of 0.9584, 0.9795, and 0.9665 relative to centralized baselines, but degraded on HDFS and SWaT under PGD attack (0.3791 and 0.5147), whereas CICIoMT-2024 remained robust under the same attack (0.9665). The adversarially trained FedMamba-AT sustains robust F1-scores (0.9480, 0.8357, 0.9645). FedMamba offers a robust and scalable solution for secure IIoT monitoring.

信息技术（IT）和操作技术（OT）在工业4.0中的融合产生了多种数据流，如系统日志、传感器读数和网络流量，这对工业安全至关重要。然而，现有的安全分析因模式而孤立，依赖于集中处理，引起了对隐私、延迟和可伸缩性的担忧。尽管联邦学习（FL）减轻了隐私风险，但大多数框架仍然是单模态的，缺乏对非iid数据分布的支持，并且面临对抗性规避的挑战。我们提出了FedMamba，一个新的多模态联邦学习（MMFL）框架，它创建了一个统一的基于mamba的模型，通过(i)高效的跨模态学习来解决这些问题，（ii）基于fedprox的稳定非iid训练协议，与安全聚合保持兼容，以及（iii）针对鲁棒性的特定模态对抗性训练。在HDFS、SWaT和CICIoMT-2024数据集上的实验表明，与集中式基线相比，标准FedMamba的宏观f1得分为0.9584、0.9795和0.9665，但在PGD攻击下，标准FedMamba在HDFS和SWaT上的性能下降（0.3791和0.5147），而CICIoMT-2024在相同的攻击下仍然保持鲁棒性（0.9665）。对抗性训练的FedMamba-AT保持稳健的f1得分（0.9480,0.8357,0.9645）。FedMamba为安全IIoT监控提供了强大且可扩展的解决方案。

{"title":"FedMamba: Robust multimodal federated intrusion detection for heterogeneous IoT systems","authors":"Hafiz Bilal Ahmad , Haichang Gao , Naila Latif , Tanjeena Manzoor","doi":"10.1016/j.iot.2026.101877","DOIUrl":"10.1016/j.iot.2026.101877","url":null,"abstract":"<div><div>The convergence of Information Technology (IT) and Operational Technology (OT) in Industry 4.0 produces diverse data streams, such as system logs, sensor readings, and network traffic, which are vital for industrial security. However, existing security analytics are siloed by modality and rely on centralized processing, raising concerns regarding privacy, latency, and scalability. Although Federated Learning (FL) mitigates privacy risks, most frameworks remain unimodal, lack support for non-IID data distributions, and face adversarial evasion challenges. We propose FedMamba, a novel multimodal Federated Learning (MMFL) framework that creates a unified Mamba-based model to address these issues via (i) efficient cross-modal learning, (ii) a FedProx-based protocol for stable non-IID training that remains compatible with secure aggregation, and (iii) modality-specific adversarial training for robustness. Experiments on HDFS, SWaT, and CICIoMT-2024 datasets show that the standard FedMamba achieved competitive macro F1-scores of 0.9584, 0.9795, and 0.9665 relative to centralized baselines, but degraded on HDFS and SWaT under PGD attack (0.3791 and 0.5147), whereas CICIoMT-2024 remained robust under the same attack (0.9665). The adversarially trained FedMamba-AT sustains robust F1-scores (0.9480, 0.8357, 0.9645). FedMamba offers a robust and scalable solution for secure IIoT monitoring.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"36 ","pages":"Article 101877"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077994","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 robust weighted late fusion approach for IoT 一种面向物联网的鲁棒加权后期融合方法

IF 7.6 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Internet of Things

Pub Date : 2026-03-01 Epub Date: 2025-12-17 DOI: 10.1016/j.iot.2025.101857

Vipin Nirala, Ratneshwer

IoT systems consist of numerous sensors that generate multiple modalities. Consequently, IoT-based systems often rely on these multiple modalities to perform their intended functionalities. However, the dynamic, heterogeneous, and openly-deployed nature of IoT-based systems makes them susceptible to several issues, including data corruption, inconsistencies, and unavailability. Catastrophic impacts such as node failures or security vulnerabilities further hinder the consistent availability of modalities. Therefore, achieving reliable access to all modalities at all times is almost impossible. In this work, we aim to improve decision-making in IoT systems by adaptively weighing different modalities. To this end, we propose an adaptive weighted late fusion, which combines heuristic-based strategies with optimization-based weight adaptation. This hybrid approach maintains a balance between heuristics and optimization, thereby improving overall system performance. We compare the proposed work against popular multimodal fusion approaches, including accuracy-based, F1-based, and entropy-based weighted fusion methods. Experimental results show that our proposed fusion method outperforms these approaches in terms of raw performance. Additionally, we simulate scenarios involving data corruption and modality unavailability, in which our proposed fusion method demonstrates superior performance compared to benchmark methods. In conclusion, the proposed fusion approach performs better in both ideal scenarios and challenging conditions with modality unavailability and inconsistencies.

物联网系统由产生多种模式的众多传感器组成。因此，基于物联网的系统通常依赖于这些多种模式来执行其预期功能。然而，基于物联网的系统的动态、异构和开放部署的特性使它们容易受到几个问题的影响，包括数据损坏、不一致和不可用。节点故障或安全漏洞等灾难性影响进一步阻碍了模式的一致可用性。因此，在任何时候实现所有模式的可靠访问几乎是不可能的。在这项工作中，我们的目标是通过自适应权衡不同的模式来改善物联网系统的决策。为此，我们提出了一种自适应加权后期融合算法，该算法将启发式策略与基于优化的权重自适应相结合。这种混合方法保持了启发式和优化之间的平衡，从而提高了系统的整体性能。我们将所提出的工作与流行的多模态融合方法进行了比较，包括基于精度的、基于f1的和基于熵的加权融合方法。实验结果表明，我们提出的融合方法在原始性能方面优于这些方法。此外，我们还模拟了涉及数据损坏和模态不可用的场景，与基准测试方法相比，我们提出的融合方法表现出更好的性能。总之，所提出的融合方法在理想场景和具有模态不可用和不一致的挑战性条件下都表现更好。

{"title":"A robust weighted late fusion approach for IoT","authors":"Vipin Nirala, Ratneshwer","doi":"10.1016/j.iot.2025.101857","DOIUrl":"10.1016/j.iot.2025.101857","url":null,"abstract":"<div><div>IoT systems consist of numerous sensors that generate multiple modalities. Consequently, IoT-based systems often rely on these multiple modalities to perform their intended functionalities. However, the dynamic, heterogeneous, and openly-deployed nature of IoT-based systems makes them susceptible to several issues, including data corruption, inconsistencies, and unavailability. Catastrophic impacts such as node failures or security vulnerabilities further hinder the consistent availability of modalities. Therefore, achieving reliable access to all modalities at all times is almost impossible. In this work, we aim to improve decision-making in IoT systems by adaptively weighing different modalities. To this end, we propose an adaptive weighted late fusion, which combines heuristic-based strategies with optimization-based weight adaptation. This hybrid approach maintains a balance between heuristics and optimization, thereby improving overall system performance. We compare the proposed work against popular multimodal fusion approaches, including accuracy-based, F1-based, and entropy-based weighted fusion methods. Experimental results show that our proposed fusion method outperforms these approaches in terms of raw performance. Additionally, we simulate scenarios involving data corruption and modality unavailability, in which our proposed fusion method demonstrates superior performance compared to benchmark methods. In conclusion, the proposed fusion approach performs better in both ideal scenarios and challenging conditions with modality unavailability and inconsistencies.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"36 ","pages":"Article 101857"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841108","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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