Neural rendering algorithms have revolutionized computer graphics, yet their impact on real-time rendering under arbitrary lighting conditions remains limited due to strict latency constraints in practical applications. The key challenge lies in formulating a compact yet expressive material representation. To address this, we propose TransGI, a novel neural rendering method for real-time, high-fidelity global illumination. It comprises an object-centric neural transfer model for material representation and a radiance-sharing lighting system for efficient illumination. Traditional BSDF representations and spatial neural material representations lack expressiveness, requiring thousands of ray evaluations to converge to noise-free colors. Conversely, real-time methods trade quality for efficiency by supporting only diffuse materials. In contrast, our object-centric neural transfer model achieves compactness and expressiveness through an MLP-based decoder and vertex-attached latent features, supporting glossy effects with low memory overhead. For dynamic, varying lighting conditions, we introduce local light probes capturing scene radiance, coupled with an across-probe radiance-sharing strategy for efficient probe generation. We implemented our method in a real-time rendering engine, combining compute shaders and CUDA-based neural networks. Experimental results demonstrate that our method achieves real-time performance of less than 10 ms to render a frame and significantly improved rendering quality compared to baseline methods.

Recent advances in underwater image enhancement (UIE) have achieved notable progress using deep learning techniques; however, existing methods often struggle with limited receptive fields, inadequate frequency modeling, and poor structural perception, leading to sub-optimal visual quality and weak generalization in complex underwater environments. To tackle these issues, we propose FACMamba, a Mamba-based framework augmented with frequency-aware mechanisms, enabling efficient modeling of long-range spatial relations for underwater image restoration. Specifically, FACMamba incorporates three key components: a Multi-Directional Vision State-Space Module (MVSM) to model directional spatial context via the proposed 8-direction selective scan block (SS8D), a Frequency-Aware Guidance Module (FAGM) for learning informative frequency representations with low overhead, and a Structure-Aware Fusion Module (SAFM) to preserve fine-grained structural cues through adaptive multi-scale integration. Recognizing the importance of spatial-frequency interaction, our model fuses these representations via lightweight architecture to enhance both texture and color fidelity. Experiments on standard UIE benchmarks demonstrate that FACMamba achieves a favorable balance between enhancement quality and computational efficiency, outperforming many existing UIE methods.

Virtual Machines (VMs) serve as dynamic execution environments that trade-off workload isolation, performance, and elastic scalability in the cloud. However, the flexibility of VMs which allows for efficiency also makes them susceptible to stealthy and adaptive cyber threats such as resource exhaustion, privilege escalation, and lateral movement. In such environments, the traditional signature- and heuristic-based defenses often encounter difficulties, resulting in high false-positive rates and low-rank under changing attack conditions. To mitigate these limitations, we present a flexible defense system which combines feature extraction, anomaly detection, classification and mitigation in a single pipeline. The system consists of an Adaptive Feature Encoder for concise behavior representation, a Density-Aware Clustering for anomaly detection, a Transformer–Boosting Classifier for timely threat identification, and a Dynamic Mitigation Controller for prompt decision making at runtime, and with low overhead. Experiments on benchmark VM telemetry datasets (ToN-IoT and CSE-CIC-IDS2018) indicate that VMShield provides 99.8% accuracy, 99.7% precision, 99.6% F1-score, and reduces false positives by 35% compared to state-of-the-art baselines. Stress testing ensures scalability, keeping detection latency at ~ 240 ms and overhead under 7%. By integrating the accuracy with operational resilience, proposed adaptive and scalable protection framework offers a practical defense to protect the cloud-hosted VMs from the emerging adversarial threats.

Introduction Heating, Ventilation and Air Conditioning (HVAC) sensor fault diagnosis is essential for ensuring the reliability and energy efficiency of intelligent building systems. However, existing diagnostic methods suffer from insufficient adaptability to multi-scale features, weak temporal dependency modeling, and poor generalization under small samples, and are highly sensitive to Gaussian noise. Method To address these limitations, this study proposes a fault diagnosis method that integrates an improved one-dimensional convolutional neural network (1-D CNN) with wavelet packet clustering. First, a multi-scale convolution module is designed using parallel 3/5/7 convolution kernels and residual connections to extract temporal features across different receptive fields. Then, wavelet packet decomposition is used to divide the original signal into eight frequency bands and construct energy feature vectors. K-means clustering is performed in an unsupervised manner, and Softmax-based weight fusion is used to realize end-to-end diagnosis with low computational overhead. Results Experimental results show that the proposed method achieves a diagnostic accuracy of 97.84% and an F1-score of 0.97. Under 30% Gaussian white noise, the area under the curve decreases by only 4%, and the instantaneous robustness drop increases by 0.01 within the 10%-30% noise range, demonstrating strong noise resistance and generalized learning capability. Discussion and Conclusion The proposed method effectively balances feature-scale adaptability, temporal modeling, and robustness under noisy and small-sample conditions. With low inference complexity and high diagnostic stability, it provides a feasible paradigm for real-time fault detection and reliable operation and maintenance in intelligent building HVAC systems.

The Wireless Sensor Networks (WSNs) are utilized by many monitoring applications, and it is widely accessible. Restricted node energy and dynamic network backgrounds limits the effectiveness of WSN. Thus, premature node failures and shorter network lifetime (NL) may result from these limitations. For Cluster Head (CH) selection, conventional clustering methods are ineffective, because these conventional methods mostly utilized static metrics. So, these conventional methods fail to adapt to dynamic topologies and energy patterns. An AI-enhanced Firefly–Gradient Descent Hybrid Optimization (AI-FGDHO) model is suggested in this study for the purpose of resolving those issues. In the network structure, intelligent decision-making is integrated by AI-FGDHO model. The node-level local CH candidacy scoring with lightweight machine learning (ML) algorithms and CH level collaborative model updates without raw data sharing using federated learning (FL) are utilized by this suggested model. The CH rotation schedules and routing strategies are dynamically refined by the Reinforcement learning (RL). For enhancing CH placement, and exploiting network topology, graph neural networks (GNNs) are used. In the exploration ability of Firefly optimization and the exploitation strength of gradient descent, these AI components are integrated, and it facilitate in adaptive and energy-aware clustering. Then, simulation was conducted with the suggested AI-FGDHO and conventional methods. With higher residual energy (0.70 J), delivery ratio (80), NL (950 rounds), throughput (900 packets), lower overhead (120 packets), latency (200 ms), reduced CH rotations (22), and improved coverage (75), the suggested AI-FGDHO model executes better than conventional methods, and it was demonstrated by the simulation outcomes.

Approximate computing (AxC) is a method that allows tiny and controlled errors in exchange for lower power consumption and reduced hardware cost. It offers an effective way to balance computational accuracy and efficiency in error-tolerant tasks. This paper explains two representative designs that use approximation at both the circuit level and the system level. The first design is the Error-Reducing Progressive Prediction Approximate Adder (ERCPAA). It improves accuracy by using progressive carry prediction and constant truncation while maintaining low hardware overhead. The second design is an adaptive Least Mean Square (LMS) filter that uses a sign-aware approximate multiplier. This multiplier lowers computational cost and supports stable convergence. A detailed evaluation of accuracy, power, and area shows how these designs work together to produce efficient and error-resilient hardware.

The rapid adoption of GPS-enabled mobile devices has produced massive trajectory datasets that drive modern applications in traffic prediction, logistics, and spatio-temporal analytics. Yet traditional database management systems (DBMSs) still lack native operators to process such data efficiently. To overcome this limitation, we introduce a set of k-nearest neighbor (k-NN) user-defined aggregates (UDAs) that embed k-NN processing directly within the PostgreSQL engine. By integrating computation into the database core, our approach minimizes data transfer and latency while maintaining low storage overhead. Experiments on benchmarked BerlinMOD-derived datasets demonstrate that the proposed UDAs reduce query execution time by 6–23%, depending on dataset size and query complexity.

The quick development of Cloud-IoT systems has enabled large-scale automation, universal sensing, and intelligent decision-making across various application domains, including smart cities, industrial IoT, healthcare and transportation. However, the integration of resource-constrained IoT devices with cloud platforms familiarizes significant security encounters, predominantly in authentication and in data protection. Traditional cryptographic and authentication structures look to be computationally very intensive which makes them inappropriate for low-power IoT architectures. This research recommends a robust and energy effective security framework that integrates the Multi-Factor Authentication (MFA) with the lightweight encryption to ensure secure device access, user authentication and data confidentiality in Cloud-IoT environments. The proposed MFA model incorporates possession-based, knowledge-based and behavioral factors thereby minimalizing the threat of credential compromise and the unauthorized access. Lightweight encryption algorithms such as SPECK and PRESENT are labouring to produce cryptographically secure authentication tokens and protect announcement without overloading the constrained devices. Investigational evaluations validate better-quality resistance to replay, impersonation and brute-force attacks while achieving low computational overhead. The mixture of MFA and lightweight cryptography efficiently supports trust, enhances system resilience, and provides a scalable security architecture suitable for next-generation IoT deployments.

The operational status of photovoltaic modules directly impacts power generation efficiency, making rapid and precise fault detection crucial for intelligent operation and maintenance of Photovoltaic (PV) power plants. Addressing the perceptual limitations of single-modal images in complex environments, this study constructs an RGBIRPV multimodal dataset tailored for centralized PV power plants and proposes an RFE-YOLO model. This model enhances detection performance through three core mechanisms: The RC module employs a CBAM-based attention mechanism for multi-parameter feature extraction, utilizing heterogeneous RC_V and RC_I architectures to achieve differentiated feature enhancement for visible and infrared modalities. The lightweight adaptive fusion FA module introduces learnable modality balance and attention cascading mechanisms to optimize multimodal information fusion. Concurrently, the multi-scale enhanced EVG module based on GSConv achieves synergistic representation of shallow details and deep semantics with low computational overhead. The experiment employed an 8:1:1 data partitioning scheme. Compared to the YOLOv11n model employing feature-level mid-fusion, the model proposed in this study achieves improvements of 2.9%, 1.8%, and 1.5% in precision, mAP@50, and F1 score, respectively. It effectively meets the demand for rapid and accurate detection of PV module failures in real power plant environments, providing an effective technical solution for intelligent operation and maintenance of photovoltaic power plants.

Secure storage model for digital forensics represents essential progress in the domain, addressing the major problems associated with protecting and maintaining digital evidence. This method employs recent encryption systems and optimal key generation methods to ensure the confidentiality and integrity of data throughout the investigative process. DFA-AOKGE a Digital Forensics Architecture with Authentication and Optimal Key Generation-based Encryption—for secure evidence storage in cloud/edge settings. Evidence objects are split into four shards, each encrypted with an independently derived key (multikey model) using AES-GCM for confidentiality and integrity. A Secure Block Verification Mechanism (SBVM) authenticates every shard and its lineage using a Merkle-root and per-block HMACs, enabling tamper-evident audit. A lightweight Optimal Key Generation pipeline strengthens seeds with memory-hard KDF (scrypt) and context-bound HKDF to derive per-shard keys deterministically while preventing cross-shard compromise. The architecture supports homomorphic-ready storage (optional: replace per-shard AES with multikey homomorphic encryption for privacy-preserving computation). Experiments (design-time analysis) show improved resistance to key compromise, replay/tamper attempts, and insider risk, while maintaining low operational overhead and clean forensic chain-of-custody.

The widespread adoption of Android has made it a primary target for increasingly sophisticated malware, posing a significant challenge to mobile security. Traditional static or behavioural approaches often struggle with obfuscation and lack contextual integration across multiple feature domains. In this work, we propose GIT-GuardNet, a novel Graph-Informed Transformer Network that leverages multi-modal learning to detect Android malware with high precision and robustness. GIT-GuardNet fuses three complementary perspectives: (i) static code attributes captured through a Transformer encoder, (ii) call graph structures modelled via a Graph Attention Network (GAT), and (iii) temporal behaviour traces learned using a Temporal Transformer. These encoders are integrated using a cross-attention fusion mechanism that dynamically weighs inter-modal dependencies, enabling more informed decision-making under both benign and adversarial conditions. We conducted comprehensive experiments on a large-scale dataset comprising 15,036 Android applications, including 5,560 malware samples from the Drebin project. GIT-GuardNet achieves state-of-the-art performance, reaching 99.85% accuracy, 99.89% precision, and 99.94 AUC, outperforming traditional machine learning models, single-view deep networks, and recent hybrid approaches like DroidFusion. Ablation studies confirm the complementary impact of each modality and the effectiveness of the cross-attention design. Our results demonstrate the strong generalization of GIT-GuardNet in obfuscated and stealthy threats, low inference overhead, and practical applicability for real-world mobile threat detection. This study provides a powerful and extensible framework for future research in secure mobile computing and intelligent malware defence.

ABSTRACT Lightweight and secure authentication is a fundamental requirement for mobile roaming in edge‐assisted networks, particularly in the presence of resource constraints and the emerging threat of quantum‐capable adversaries. This paper proposes a blockchain‐assisted authentication protocol that employs post‐quantum cryptographic primitives to generate and validate device‐bound tokens. During registration, a Home Agent (HA) issues blockchain‐anchored tokens containing signed security metadata and a freshness counter to prevent replay attacks. In roaming scenarios, the Mobile User (MU) selectively discloses token metadata to the Foreign Agent (FA), which verifies its authenticity with the HA to enable efficient and trustworthy authentication. A hybrid key establishment using post‐quantum key encapsulation ensures forward secrecy and quantum‐resistant confidentiality. Formal verification through BAN logic reasoning and automated analysis using the Scyther tool confirm that the protocol withstands impersonation, replay, and man‐in‐the‐middle attacks. Experimental evaluation on mobile devices demonstrates low computational and communication overhead, showing that the protocol is practical and well‐suited for real‐world deployment in edge‐assisted mobility environments.

TOPLDM: Towards dynamic low overhead traffic obfuscation based on packet length distribution modification

This paper proposes a decorrelation scheme based on product quantization, termed Reference-Vector Removed Product Quantization (RvRPQ), for approximate nearest neighbor (ANN) search. The core idea is to capture the redundancy among database vectors by representing them with compactly encoded reference-vectors, which are then subtracted from the original vectors to yield residual vectors. We provide a theoretical derivation for obtaining the optimal reference-vectors. This preprocessing step significantly improves the quantization accuracy of the subsequent product quantization applied to the residuals. To maintain low online computational complexity and control memory overhead, we apply vector quantization to the reference-vectors and allocate only a small number of additional bits to store their indices. Experimental results show that RvRPQ substantially outperforms state-of-the-art ANN methods in terms of retrieval accuracy, while preserving high search efficiency.

ABSTRACT The Internet of Things (IoT) has witnessed rapid global adoption, driving the development of intelligent networks that provide smart services and computing at the network edge. This paper introduces a Multi‐Layered Aggregation and Lightweight Prediction Framework that integrates data aggregation and data prediction methods, specifically designed for IoT‐based Wireless Sensor Networks (WSNs). The framework first employs temporal and spatial data aggregation (TDA and SDA) to minimize transmissions between cluster member nodes (CMNs) and cluster heads (CHs). It then applies a lightweight data prediction (LDP) model, based on linear extrapolation with adaptive correction, to further reduce data transfer volume between CHs and the base station (BS). Unlike approaches relying solely on aggregation or prediction, the proposed framework leverages their synergy to achieve significant energy savings and prolong network lifetime. Experimental validation using real‐world LUYF data confirms its superiority over state‐of‐the‐art data reduction techniques, demonstrating simplicity, low computational overhead, and effective transmission reduction. Despite its lightweight design, LDP remains reliable and versatile, seamlessly integrating with various cluster‐based data aggregation schemes. Overall, the proposed framework preserves data integrity and quality while conserving energy and extending the operational lifespan of WSNs.

Network on Chip (NoC) architectures are essential subsystems for on-chip communication. They use routers and simplified protocols modeled after public data networks to transport packets using complex routing algorithms from their source to their destination. Reliable communication can be severely hampered by component failures, such as malfunctioning routers or cables, which can interrupt packet transfer. Performance may be harmed by the narrow criteria used by traditional fault-tolerant routing algorithms to find reliable routes. In order to improve routing reliability and Quality of Service (QoS) in scalable NoC architectures, this paper suggests a novel, adaptive fault-tolerant routing algorithm that incorporates the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), a multi-criteria decision-making technique. The suggested approach dynamically assesses and ranks alternate routes to choose the best ones, even when there are failures, by utilizing path length and density information from nearby nodes. On 8 × 8 meshes with 10% link failures, the approach reduces average delay by ~ 8–12% compared to EDAR and increases throughput by ~ 2–5% compared to EDAR; on application-driven traces, it reduces delay by ~ 5–15% at nearly equal throughput. It reduces energy per flit by around 15–20% compared to EDAR, improves throughput by about 3–4%, and lowers delay by about 8–10% under transient, thermal, and voltage disturbances. The two-stage decision core maintains the improvements on 16 × 16 meshes and reroutes locally in about 3–5 cycles without adding a critical-path cost. Additionally, the approach ensures scalability for large-scale NoC implementations by introducing low hardware overhead. The suggested algorithm is a viable answer for next-generation NoC designs, meeting the requirements of high-performance, dependable, and scalable on-chip communication systems thanks to its combination of fault tolerance, QoS awareness, and resource efficiency.

Under the context of “Energy Conservation and Emission Reduction,” low-sulfur fuel has become widely adopted in maritime operations, posing significant challenges to cylinder lubrication systems. Traditional oil injection strategies, heavily reliant on manual experience, suffer from instability and high costs. To address this, a lightweight image retrieval model for cylinder lubrication is proposed, leveraging deep learning and computer vision to support oiling decisions based on visual features. The model comprises three components: a backbone network, a feature enhancement module, and a similarity retrieval module. Specifically, EfficientNetB0 serves as the backbone for efficient feature extraction under low computational overhead. MobileViT Blocks are integrated to combine local feature perception of Convolutional Neural Networks (CNNs) with the global modeling capacity of Transformers. To further improve receptive field and multi-scale representation, Receptive Field Blocks (RFB) are introduced between the components. Additionally, the Convolutional Block Attention Module (CBAM) attention mechanism enhances focus on salient regions, improving feature discrimination. A high-quality image dataset was constructed using WINNING’s large bulk carriers under various sea conditions. The experimental results demonstrate that the EfficientNetB0 + RFB + MobileViT + CBAM model achieves excellent performance with minimal computational cost: 99.71% Precision, 99.69% Recall, and 99.70% F1-score—improvements of 11.81%, 15.36%, and 13.62%, respectively, over the baseline EfficientNetB0. With only a 0.3 GFLOP and 8.3 MB increase in model size, the approach balances accuracy and inference efficiency. The model also demonstrates good robustness and application stability in real-world ship testing, with potential for further adoption in the field of intelligent ship maintenance.

Wireless Sensor Networks (WSNs) have emerged as a foundational technology for monitoring and data collection in diverse domains such as environmental sensing, smart agriculture, and industrial automation. Precise node localization plays a vital role in WSNs, enabling effective data interpretation, reliable routing, and spatial context awareness. The challenge intensifies in range-free settings, where a lack of direct distance data demands efficient indirect estimation methods, particularly in large-scale, energy-constrained deployments. This work proposes a hybrid localization framework that integrates the distance vector-hop (DV-Hop) range-free localization algorithm with the Fire Hawk Optimizer (FHO), a nature-inspired metaheuristic method inspired by the predatory behavior of fire hawks. The proposed FHODV-Hop method enhances location estimation accuracy while maintaining low computational overhead by inserting the FHO into the third stage of the DV-Hop algorithm. Extensive simulations are conducted on multiple topologies, including random, circular, square-grid, and S-shaped, under various network parameters such as node densities, anchor rates, population sizes, and communication ranges. The results show that the proposed FHODV-Hop model achieves competitive performance in Average Localization Error (ALE), localization ratio, convergence behavior, computational, and runtime efficiency. Specifically, FHODV-Hop reduces the ALE by up to 35% in random deployments, 25% in circular networks, and nearly 45% in structured square-grid layouts compared to the classical DV-Hop. Even under highly irregular S-shaped conditions, the algorithm achieves around 20% improvement. Furthermore, convergence speed is accelerated by approximately 25%, and computational time is reduced by nearly 18%, demonstrating its scalability and practical applicability. Therefore, these results demonstrate that the proposed model offers a promising balance between accuracy and practicality for real-world WSN deployments.

This research presents an integrated approach to enhancing the performance of Copper Indium Gallium Selenide (CIGS) thin-film Photovoltaic (PV) panels through active thermal management for urban rooftop installations in high-temperature environments such as India. CIGS modules, while efficient under low-light and diffuse conditions, suffer significant power degradation at elevated temperatures. To address this, a water-based hybrid cooling system was developed and tested on a 10 W CIGS module, featuring aluminium-copper ductwork and an Arduino-based dual-threshold control system. Experimental trials across a wide irradiance spectrum (210-980 W/m²), with and without a 2× solar concentrator, demonstrated a 3.56% improvement in energy yield under standard conditions and a 28.5% gain with concentrated sunlight. Advanced predictive modelling using Random Forest regression further validated the thermal sensitivity and projected output behaviour of a 1 kW system, confirming a sharp decline in power beyond 50°C. The study also revealed improved durability through reduced thermal stress, as supported by the Arrhenius degradation model. Economically, the system reduces payback time and enhances ROI, particularly in high-insolation urban areas. These findings highlight the viability and scalability of intelligent cooling strategies for next-generation rooftop solar deployments in emerging economies. Major Findings: An Arduino-controlled dual-threshold cooling system significantly reduced CIGS module temperatures, improving output efficiency under rooftop conditions. A Random Forest model trained on experimental data achieved an R² of 0.932, accurately predicting nonlinear power losses due to thermal stress. The integrated system demonstrated environmental viability with low water consumption (1.2 L/day/10 W) and minimal energy overhead, supporting sustainable PV deployment in hot climates.

This paper presents a robust and computationally efficient fault classification framework for wind energy conversion systems (WECS), built upon a Robust Random Vector Functional Link Network (Robust-RVFLN) and validated through real-time simulations on a Real-Time Digital Simulator (RTDS). Unlike existing studies that depend on high-dimensional feature extraction or purely data-driven deep learning models, our approach leverages a compact set of five statistically significant and physically interpretable features derived from rotor torque, phase current, DC-link voltage, and dq-axis current components. This reduced feature set ensures both high discriminative power and low computational overhead, enabling effective deployment in resource-constrained edge devices and large-scale wind farms. A synthesized dataset representing seven representative fault scenarios—including converter, generator, gearbox, and grid faults—was employed to evaluate the model. Comparative analysis shows that the Robust-RVFLN consistently outperforms conventional classifiers (SVM, ELM) and deep models (CNN, LSTM), delivering accuracy rates of up to 99.85% for grid-side line-to-ground faults and 99.81% for generator faults. Beyond accuracy, evaluation metrics such as precision, recall, and F1-score further validate its robustness under transient operating conditions. By uniting interpretability, scalability, and real-time performance, the proposed framework addresses critical challenges in condition monitoring and predictive maintenance, offering a practical and transferable solution for next-generation renewable energy infrastructures.