ABSTRACT Cloud base height (CBH) is a fundamental atmospheric parameter for weather forecasting, aviation safety, and climate research, as it provides key information on boundary layer structure, atmospheric stability, and cloud–radiation interactions. In this study, a two-step hybrid framework combining statistical cloud detection with machine learning (ML) regression for CBH estimation is developed and evaluated. In the first step, cloud presence is identified using a Variability Index (VI) defined as the ratio of the standard deviation of the backscatter profile to its peak value. This physically interpretable index shows strong class separability, with a large effect size (Cohen’s d ≈ 1.96), a maximum F1 score of about 0.83 at a VI of approximately 0.24, and an area under the ROC curve of 0.88, indicating effective cloud detection performance. In the second step, Multiple Linear Regression, Fine Tree Regression, Random Forest, and Gaussian Process Regression (GPR) models are applied to cloud-present profiles to estimate CBH. Among these, Random Forest, a tree based nonlinear ensemble model perform best, achieving a high correlation coefficients of about 0.94 ± 0.04. GPR, a kernel-based model, demonstrated slightly lower performance compared to Random Forest, achieving correlation coefficients of R = 0.91 ± 0.05, while the Fine Tree model showed the weakest performance among the nonlinear models tested in this study, achieving R = 0.89 ± 0.07. In contrast, Multiple Linear Regression model showed lowest accuracy with R = 0.58 ± 0.13. The results demonstrate that combining a simple, explainable statistical classification approach with advanced machine learning regression significantly improves the reliability and accuracy of CBH retrieval from Lidar backscatter data. The proposed framework is computationally efficient and shows strong potential for operational implementation in real-time atmospheric monitoring networks.

The rapid rise of the Internet of Things (IoT) has revolutionized connectivity across various domains, including smart homes, healthcare, and industrial systems. However, the large-scale integration of heterogeneous devices has significantly increased security vulnerabilities and cyberattack risks. Traditional intrusion detection systems (IDS) are often insufficient for IoT environments due to limited device resources and dynamic network behavior. This study proposes a machine learning–based IDS for detecting and classifying malicious activities in IoT networks in real time. Supervised learning algorithms, including Decision Tree, Random Forest, and Support Vector Machine (SVM), were employed to analyze network traffic and identify anomalies. Experimental evaluation using benchmark IoT datasets showed that the Random Forest model achieved the best performance with an accuracy of 98.1%, detection rate of 98.2%, precision of 98.0%, recall of 98.1%, and a low false positive rate of 1.9%. Comparative analysis demonstrated that the proposed approach outperformed conventional IDS techniques in both detection capability and reliability. These results highlight the effectiveness of intelligent learning models in enhancing IoT network security and supporting trustworthy network operations

Precision aquaculture demands robust communication networks capable of coordinating thousands of distributed sensors across marine and freshwater facilities. Current aquaculture IoT networks face critical challenges including underwater signal attenuation reaching 98% loss at 100 m depth, dynamic topology changes from fish movement and water currents, and severe energy constraints on battery-powered sensor nodes. This paper introduces SQUID-COMM, a novel bio-inspired communication framework emulating the signaling mechanisms of the Colossal Squid (Mesonychoteuthis hamiltoni). The framework introduces seven innovative mechanisms: Bioluminescent Pulse-Coded Modulation (BPCM) achieving 34% higher spectral efficiency through adaptive signal encoding; Chromatophore-Inspired Channel Adaptation (CICA) enabling 15ms frequency hopping response time; Distributed Axon-Ganglia Routing Protocol (DAGRP) maintaining 99.7% packet delivery under 40% node mobility; Tentacle-Topology Self-Organization (TTSO) for dynamic mesh network formation; Giant Fiber Emergency Broadcast (GFEB) achieving sub-50ms critical alert propagation; Photophore Synchronization Protocol (PSP) for microsecond-accurate time coordination; and Ink-Cloud Congestion Control (ICCC) reducing packet loss by 82%. The Enhanced SQUID-COMM variant incorporates Neuromorphic Edge Processing reducing cloud communication by 78%, Federated Learning Coordination for distributed model updates, and Quantum-Resistant Encryption for future-proof security. Experimental evaluation across five aquaculture deployment scenarios demonstrates end-to-end latency of 12.3ms representing 78% reduction compared to LoRaWAN, throughput of 2.4 Mbps in turbid conditions spanning 5-150 NTU, energy efficiency of 0.23 mJ/bit constituting 67% improvement over Zigbee, and network lifetime extension of 340%. Real-world deployment at four commercial facilities across Norway, Egypt, Thailand, and Greece over 120 days processed 2.3 billion sensor readings with 99.94% reliability, enabling fish behavior detection at 94.7% accuracy and early disease detection with 4.2-day lead time. Statistical analysis confirms significant improvements with p-values below 0.001 and Cohen’s d exceeding 1.2, while economic evaluation demonstrates annual savings of €89,000-€340,000 per facility.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-54545-6.

We present the development and evaluation of a real-time Graph Neural Network-based trigger module for the electromagnetic calorimeter of the Belle II experiment at the SuperKEKB collider. The algorithm processes calorimeter trigger cells as graph nodes to perform clustering, feature extraction, and per-cluster signal classification with deterministic latency. The model predicts cluster positions and energies and provides a signal classification score, enabling a more flexible clustering strategy than the baseline trigger algorithm. Implemented on an FPGA and integrated into the Belle II trigger readout infrastructure for synchronous operation, the system sustains the 8 MHz trigger throughput with an end-to-end latency of 3.168 μs. The performance is evaluated on simulated events and collision data. The energy resolution is comparable to the baseline trigger, while the position resolution for high-energy clusters improves by up to 18 % in the central detector region. Cluster purity increases by up to 20 % at low energies for isolated clusters, and cluster efficiency improves by up to 20 % for overlapping clusters. The signal classifier enables additional background suppression at fixed signal retention. These results demonstrate a first step towards GNN-based real-time reconstruction on FPGAs in a collider trigger. While the end-to-end latency exceeds the trigger decision budget, the system already sustains full operational conditions with 100 % uptime.

Efficient FPGA realization of neural network activation functions using adaptive piecewise polynomial approximations with Chebyshev nodes

• Interpretable ML framework detects shallow-lake stratification and mixing regimes. • Data-driven ΔT threshold for regime separation is physically validated using Ri scaling. • Off-the-shelf ML models classify regimes from standard high-frequency sensor data. • SHAP analysis reveals key hydrometeorological drivers of regime transitions. • Pipeline is reusable and implementable in real-time shallow-lake monitoring networks. Thermal stratification and mixing control vertical heat, oxygen, and nutrient exchange in lakes, shaping ecological functioning, biogeochemical processes, and ecosystem resilience. Operational detection of these regime shifts in shallow lakes remains challenging, as many existing methods rely on fixed, site-independent thresholds or computationally demanding hydrodynamic models that are difficult to deploy for routine monitoring. Here, we introduce and test a multi-stage, data-driven framework that combines unsupervised clustering, supervised classification, and simple physical diagnostics to identify mixing-stratification regimes in a small, shallow lake. Using one year of high-frequency measurements from Lake Eymir (Türkiye), k-means clustering of vertical temperature differences yields an empirical stratification threshold of approximately 2.4°C between 0.25 m and 4 m, which is consistent with reported criteria for shallow lakes and with gradient Richardson number scaling for this site. The resulting mixed and stratified labels are then used to train supervised models (XGBoost, K-nearest neighbours, Gradient Boosting, Decision Tree), with XGBoost achieving a classification accuracy of about 95% when dissolved oxygen is included as a predictor and around 90% when it is excluded. SHAP analysis indicates that dissolved-oxygen differences act as a strong proxy for the stratified state, while irradiation and surface temperature emerge as the main physical predictors once outcome variables are removed. The framework therefore offers a physically interpretable, computationally efficient way to derive site-specific temperature thresholds and to track stratification–mixing regimes from standard monitoring data, supporting applications such as near-real-time lake phase tracking and early warning for stratification-driven water-quality events in shallow, rapidly responding systems.

Wireless networks are faced with growing demands for high-speed connectivity and reliable coverage in smart buildings, industrial, and space environments. Traditional radio frequency (RF) networks struggle to meet these requirements due to limited bandwidth and spectrum congestion, especially in dense deployments. Visible Light communication (VLC) has emerged as a complementary solution that delivers multi-Gbps backbone links through advanced optical transmission technologies. However, integrating VLC with RF mesh networks presents significant challenges due to VLC's line-of-sight dependence and coverage limitations. Since the physical-layer characteristics of hybrid VLC/RF networks differ fundamentally from RF-only networks, existing routing protocols, such as Q-learning based Traffic-Aware Routing (QTAR) and Multi-Objective Optimized Link State Routing (MO-OLSR), often suffer severe performance degradation. To address this issue, we propose Q-learning-based Hybrid Link Selection (QHLS), a cross-layer routing protocol that employs reinforcement learning to intelligently coordinate VLC and RF transmissions. QHLS introduces a unified link quality metric that jointly considers physical-layer capacity, MAC-layer delay, and network congestion. Using this metric, QHLS enables distributed, adaptive routing decisions that dynamically adjust to real-time network conditions and heterogeneous link characteristics. Simulation results show that QHLS significantly outperforms existing protocols under diverse traffic and mobility scenarios. In high-congestion conditions, QHLS improves throughput by 24.5%–41.8%, reduces end-to-end delay by 60.9%–71.5%, and enhances packet delivery ratio by 6.1%–10.3% compared to QTAR and MO-OLSR. These results highlight the effectiveness of QHLS for hybrid VLC/RF wireless mesh networks.

BCAG-Net: A multi-branch deep learning framework with Butterworth–Chebyshev filtering and attention mechanisms for accurate cellular network traffic prediction

Abstract—The increasing complexity and volume of cyber attacks have rendered traditional signature-based intrusion detection systems inadequate for modern network environments. This paper presents the Smart Cyber Defence System (SCDS), a full-stack Django web application that integrates a supervised machine-learning pipeline with a role-based management portal to provide real-time network threat detection, multi-class classification, and automated IP-level mitigation. A Random Forest classifier is trained on a 63-feature network-traffic dataset covering five threat categories: DDoS, Mal-ware, Phishing, Intrusion, and Benign traffic. A rule-based fallback mechanism ensures continuous protection when no trained model artefact is available. High-severity detections trigger automatic IP blocking persisted in a relational database with full administrator-controlled lifecycle management. The system achieves an overall classification accuracy of 93.4 percent with a macro-averaged F1-score of 0.92 on the hold-out test set, and a prediction latency of 42 ms for trained-model inference. These results demonstrate the practical viability of embedding ML-based cyber defence within an accessible, open-source web platform suitable for organisations without dedicated security teams. Index Terms—cyber defence, intrusion detection, Random Forest, machine learning, network security, DDoS detection, IP blocking, Django, threat classification, anomaly detection, automated mitigation

ABSTRACT In this article, a real-time Network Intrusion Detection System (NIDS) with machine learning is introduced to tackle the growing cybersecurity challenges and enhance the continuous monitoring aspect of IT audit and governance. Conventional signature-based detection and manual audit methods are no longer sufficient for the more complex and voluminous cyberattacks. To address these challenges, the proposed system combines Apache Kafka with Apache Spark to enable real-time data streaming and processing of live network traffic. It uses an automated feature extraction method and an ensemble classification model based on the combination of Random Forest (RF) and Deep Neural Network (DNN) approaches to enhance the detection accuracy and scalability. Three benchmark datasets in cybersecurity—NSL-KDD, CIC-IDS2017, and UNSW-NB15—are used for training and testing the model. Evaluated using metrics like precision, recall, F1-score, ROC-AUC, latency, and throughput. Experimental results show better performance than other classifiers, such as SVM and kNN, with the F1 score ranging between 0.98 and 0.99, latency less than 10 ms per flow, and processing speed higher than 1000 flows per second. Furthermore, the study maps the system to COBIT DS5, ISO 27001 Annex A.12 and NIST CSF DE.CM-1, a continuous monitoring, audit readiness, security tracking, and SIEM integrated governance support.

The integration of multimodal imaging into operating rooms paves the way for comprehensive surgical scene understanding. In ophthalmic surgery, by now, two complementary imaging modalities are available: operating microscope (OPMI) imaging and real-time intraoperative optical coherence tomography (iOCT). This first work toward temporal OPMI and iOCT feature fusion demonstrates the potential of multimodal image processing for multi-head prediction through the example of precise instrument tracking in vitreoretinal surgery. We propose a multimodal, temporal, real-time capable network architecture to perform joint instrument detection, keypoint localization, and tool-tissue distance estimation. Our network design integrates a cross-attention fusion module to merge OPMI and iOCT image features, which are efficiently extracted via a Yolo-NAS and a CNN encoder, respectively. Furthermore, a region-based recurrent module leverages temporal coherence. Our experiments demonstrate reliable instrument localization and keypoint detection (95.79% mAP50) and show that the incorporation of iOCT significantly improves tool-tissue distance estimation, while achieving real-time processing rates of 22.5 ms per frame. Especially for close distances to the retina (below 1mm), the distance estimation accuracy improved from (OPMI only) to (multimodal). Feature fusion of multimodal imaging can enhance multitask prediction accuracy compared to single-modality processing, and real-time processing performance can be achieved through tailored network design. While our results demonstrate the potential of multimodal processing for image-guided vitreoretinal surgery, they also underline key challenges that motivate future research toward more reliable, consistent, and comprehensive surgical scene understanding.

HighlightsWhat are the main findings?An integrated IoT–Digital Twin–CPS framework with edge-based LSTM enables closed-loop predictive control in ornamental aquaculture.The proposed system achieves a 26.86% reduction in energy consumption by exploiting predictive timing and natural environmental dynamics during a 45-day real-world deployment.What are the implications of the main findings?Demonstrates that Digital Twin-assisted validation can bridge the gap between AI prediction and safe control execution, improving system explainability and reliability.Provides a scalable and energy-aware aquaculture framework applicable to resource-constrained and small-scale smart farming environments.Reducing energy consumption while maintaining stable water quality remains a major challenge in ornamental aquaculture. This study proposes an integrated predictive and energy-aware aquaculture management framework combining Internet of Things (IoT) sensing, Long Short-Term Memory (LSTM)-based prediction, Digital Twin (DT) simulation, and Cyber-Physical System (CPS) control. Real-time sensor networks monitored dissolved oxygen (DO), ammonia (NH3), temperature, pH, turbidity, and energy consumption in a koi pond over a 45-day deployment period. Forecasted environmental states generated by the LSTM model were validated through a physics-informed Digital Twin prior to actuator execution to improve operational reliability and control safety. Experimental results demonstrated strong agreement between the Digital Twin and observed pond dynamics, achieving R2 values of 0.97 for dissolved oxygen and 0.94 for ammonia. Compared with conventional manual operation, the proposed smart predictive control mode reduced total energy consumption by 26.86%. Statistical analysis confirmed that the reduction was highly significant (p < 0.001), with average daily energy consumption decreasing from 212 ± 6.06 Wh/day under manual operation to 154.71 ± 4.52 Wh/day under smart predictive control.

This study provides a comparative performance evaluation of commercial Precise Point Positioning Real-Time Kinematic (PPP-RTK) and public Network RTK (NRTK) services for vehicle-based positioning in urban and suburban environments. Using low-cost u-blox ZED-F9 receivers, the research assesses the accuracy, availability, and robustness of the u-blox PointPerfect service against a regional NRTK network across diverse real-world scenarios, including high-speed highway conditions and signal-challenging urban corridors. The experimental framework utilizes a rigid-bar setup for high-precision ground-truth validation and incorporates an independent vertical accuracy assessment against a LiDAR-derived digital elevation model (DEM). The results demonstrate that all tested configurations achieve decimeter-level accuracy. Notably, the integration of PPP-RTK with an inertial measurement unit (IMU) delivers performance nearly equivalent to NRTK, effectively mitigating vertical biases and ensuring positioning continuity in GNSS-denied areas such as tunnels. These results confirm that low-cost GNSS solutions, when paired with modern augmentation services and IMU integration, can meet the stringent demands of mass-market applications like Cooperative Intelligent Transport Systems (C-ITS) and autonomous mobility.

The rapid proliferation of sophisticated cyber threats has exposed critical limitations in conventional security architectures that rely on isolated, reactive tools. This paper presents ZeroGuardian-XDR, an intelligent and lightweight Extended Detection and Response (XDR) framework engineered to deliver real-time network threat detection, automated vulnerability assessment, and proactive incident alerting through a unified platform. The proposed system employs a trained autoencoder neural network for behavioral anomaly detection, enabling the identification of zero-day and previously unknown threats without reliance on static signature databases. ZeroGuardian-XDR integrates nine live global threat intelligence feeds including AlienVault OTX, Abuse.ch, Feodo Tracker, URLhaus, Blocklist.de, ThreatFox, NVD CVEs, MITRE ATT&amp;CK, and EmergingThreats, collectively maintaining over 22,000 dynamic threat indicators automatically refreshed every six hours. The system maps all detections to the MITRE ATT&amp;CK framework with 87% technique coverage across 8 tactical phases and 691 monitored techniques. A professional SOC-style web dashboard, multi-channel alert delivery via Telegram and email, automated PDF report generation, and an Nmap-powered CVE vulnerability scanner complete the integrated architecture. Experimental evaluation using five simulated zero-day attack scenarios demonstrated 100% detection accuracy with minimal false positive rates. The framework is deployed on Ubuntu Server 24.04 and made publicly available through open-source distribution with Windows and Linux installer packages. ZeroGuardian-XDR represents a scalable, cost-effective, and academically reproducible cybersecurity solution for modern network protection

Responding to a natural disaster is a short-lived and complex task requiring fast coordination and analytical skills. Responses to natural disasters must be fast, and resources need to be used in an efficient and effective manner. An automated system can significantly reduce human decision errors, increase speed, and lower operational costs. This study presents an automated system that leverages real-time mobile network data to optimize team deployment and coordination in order to repair base station alarms and re-deploy existing cellular communication networks whose core nodes have been damaged or congested. Algorithms presenting solutions from optimization to fast heuristics are adopted for automated assignment of technical teams to alarms as a response to a natural disaster. Algorithms are evaluated and tested for their multi-objective assignment performance, subject to constraints. A real alarm dataset logged from base stations is used. The average number of alarms assigned, assignment rate, average latency of the automated assignment system, and travel distance of the technical teams to the base station are used as the performance metrics. Cellular communication has to be re-deployed to sustain coordination and resilience as an immediate response to a natural disaster. The presented approach and comparative results show that technical teams located in neighboring cities can be assigned an immediate response, and automatic assignment of new arrival alarms with high priority can be assured by minimizing the travel distance of technical teams to the level of a few kilometers. This study fills specific gaps in comparison to prior studies by using the real alarm data logged after a destructive earthquake event, adopting a multi-objective optimization along with the performance metrics used by telecom operators.

Dead fuel moisture content (DFMC) critically influences wildfire behaviour, and its modelling underpins many fire management decision support systems. Recent modelling advances have enabled accurate forecast of point-scale fuel moisture, but their reliance on continuous real-time sensor functionality creates operational vulnerabilities when sensors may fail. Maintaining sensor networks across large, remote domains is costly and unreliable. Therefore, we developed a spatially continuous DFMC forecast system that eliminates real-time sensor dependency by replacing sensor initialisation with remotely sensed and modelled proxies for landscape fuel moisture states. Using 23,354 site-day observations from 27 forested sites in Victoria, Australia, our machine learning model produces 7-day ahead sub-canopy DFMC forecasts with median RMSE of 11.5% and 12.8% for day 1 and 7. The approach delivers reliable spatial forecasts across forested landscapes without sensor-dependent vulnerabilities, representing a significant advancement in operational fire risk management by providing comprehensive coverage for wildfire suppression planning and prescribed burning. • We present the first ML model for sensor-independent spatial forest DFMC forecasts. • Training used 23,354 site-days from 27 below-canopy forest monitoring sites. • Our approach eliminates sensor failure vulnerabilities during wildfire events. • Day 1-7 forecasts achieve 11.5-12.8% RMSE across 150,000km 2 of forested landscape.

Thermal Infrared (TIR) tracking technology offers significant potential for applications in Internet of Things (IoT) systems, such as urban security and intelligent transportation systems, due to its ability to track targets in all weather conditions. However, the lack of color, texture, and other detailed features in infrared images hinders the ability of infrared trackers to effectively manage background noise and interference from similar objects. Furthermore, current deep learning-based TIR trackers face challenges when deployed in resource-limited IoT systems. To address these issues, this paper introduces a cloud-edge architecture spatiotemporal enhanced infrared target tracking framework for the Internet of Things, which enhances infrared tracking tasks from both spatial and temporal perspectives. This framework includes a Spatial Enhancement Network (SEN) designed for deployment on cloud devices to effectively improve details such as textures and edges in images, as well as a Real-time Temporal Enhancement Tracking Network (RTETN) optimized for edge device deployment, which further enhances infrared tracking performance by continuously extracting, storing, and updating temporal context. The two-stage architecture also significantly improves the flexibility, real-time capabilities, and stability of smart IoT systems while reducing maintenance costs. Experiments on two extensively used TIR target tracking datasets, LSOTB-TIR and PTB-TIR, demonstrate that our approach achieves success rates of 73.1% and 71.6%, respectively, at a speed of 46 fps, outperforming state-of-the-art algorithms in both accuracy and speed. Moreover, when using RTETN alone for tracking, it can reach a speed of 109 fps while maintaining tracking accuracy comparable to current state-of-the-art algorithms.

LRTISS: a lightweight real-time network for industrial surface defect segmentation

The rapid advancement of Internet of Medical Things (IoMT) technologies and the growing demand for ubiquitous healthcare services have created an urgent need for intelligent networking solutions that can seamlessly integrate space, air, and ground communication infrastructures. This paper presents a novel intent-based networking framework designed for IoMT communications within space-air-ground integrated systems. The proposed framework addresses the complex requirements of healthcare applications by introducing an intelligent intent understanding and translation mechanism that can automatically configure network resources based on high-level medical service requirements. Our approach incorporates four core components: medical intent construction for IoMT scenarios, intelligent intent classification using enhanced BERT-CNN models, sophisticated intent parsing through GlobalPointer-based entity extraction, and dynamic intent translation for real-time network policy generation. Experimental results demonstrate that our framework achieves 94.37% accuracy in intent classification and 83.76% F1-score in entity extraction. Experimental validation across space-air-ground network simulations demonstrates substantial improvements in resource utilization efficiency (23.4% increase), bandwidth allocation optimization, and latency reduction (18.6% improvement), directly enhancing patient care capabilities and clinical decision-making reliability in distributed healthcare environments.

Wireless Sensor Networks (WSNs) enable energy-efficient data collection in dynamic environments but continue to face the dual challenges of severely constrained node energy and the spatiotemporal heterogeneity of data traffic. Inspired by honeybee foraging behavior, this paper proposes a hybrid optimization framework that integrates mixed-integer linear programming (MILP) and Markov decision processes (MDP), utilizing Q-learning for adaptive decision-making. The proposed framework systematically maps the dual-layer decision-making mechanism of honeybee foraging onto a synergistic architecture combining MILP-based global planning and MDP-based local adaptation, offering a novel bio-inspired solution for mobile sink trajectory planning and adaptive routing. Specifically, the upper-level MILP module simulates a colony-level global assessment of distant nectar sources, generating an initial global trajectory by determining the optimal access sequence of cluster heads to minimize the movement cost of the mobile sink. The lower-level Q-learning module simulates the individual-level local adaptation, where bees adjust harvesting behavior in real-time based on nectar quality and distance. This module continuously optimizes routing parameters based on real-time network states, including residual energy, the ratio of surviving nodes, data queue lengths, and cluster head density. The algorithm employs an -greedy strategy to balance exploration and exploitation, while a periodic decision-update mechanism is introduced to harmonize computational efficiency with learning stability. Furthermore, a multi-objective reward function is designed to jointly optimize energy efficiency, network lifetime, end-to-end latency, and path length. Extensive simulation results demonstrate that the proposed MILP-MDP hybrid framework significantly outperforms several representative baseline algorithms in terms of network lifetime extension and energy balance. These findings validate that the integration of bio-inspired foraging strategies and reinforcement learning provides an efficient and robust solution for trajectory planning and adaptive routing in dynamic WSNs.