Search Results (8,476)

With the rapid deployment of deep neural networks (DNNs) on edge devices, traditional hardware accelerators face significant challenges in terms of data security, computational redundancy caused by sparsity, and uneven utilization of on-chip resources. This paper proposes SaE-FPGA, a secure and efficient DNN accelerator designed specifically for edge FPGA platforms. The architecture introduces three core innovations: (1) Hash-Bypass Processing Unit (HBPU): Integrating a high-speed SHA-256 hardware engine with a hash-sparse bitmap mechanism, it enables real-time data integrity verification within a single clock cycle while skipping computations for redundant zero-value data. (2) Flexible Mixed-Precision Processing Element (FMP): By reconfiguring idle BRAM and LUT resources into an active lookup table multiplication engine, it overcomes the physical bit-width limitations of DSP blocks and supports INT8/INT6/INT4 mixed-precision multiplication. (3) Multi-mode Reconfigurable Streaming Frame (MRSF): A sparse-aware, elastic load balancing and data routing mechanism designed to mask long memory access latencies and ensure high hardware resource utilization. Experimental results on the Zynq 7045 platform demonstrate that SaE-FPGA reduces redundant computations by 23.2% while maintaining high precision and minimizing precision loss. The system effectively mitigates the risk of physical tampering. When tested on ResNet-50, it achieved a 27.2% improvement in energy efficiency and a 2.97× speedup compared to DSP-based FPGA solutions. Furthermore, by fully exploiting the hybrid BRAM-LUT and DSP configuration, the proposed accelerator achieves a remarkable peak throughput of 782.4 GOPS. Full article

Urban navigation in dynamic environments remains a challenging problem for blind and visually impaired users due to the presence of unpredictable obstacles and the limitations of conventional navigation systems, which rely primarily on static map-based information and lack real-time environmental awareness. This paper presents a real-time sensor-driven navigation system based on a multi-agent architecture incorporating a reinforcement-learning navigation policy for assistive mobility in urban environments. The proposed system integrates GPS-based global localization with vision-based perception to enable continuous fusion of global route planning and local obstacle detection. This integration allows the system to dynamically adjust navigation strategies in response to changing environmental conditions. The architecture is designed as a modular multi-agent system comprising agents for perception, navigation, sensor fusion, personalization, safety arbitration, interface management, and system monitoring. The reinforcement learning component formulates local navigation as a sequential decision-making problem, where the navigation policy is trained to balance path efficiency, obstacle avoidance, and safety constraints through interaction with simulated environments. Prototype implementation is developed and evaluated in both simulation and controlled real-world scenarios. Experimental results demonstrate that the proposed system shows improved obstacle avoidance performance and navigation stability under the evaluated conditions while maintaining low-latency responsiveness compared to baseline navigation approaches. The system also exhibits robust behaviour under varying environmental conditions, supporting its potential applicability to assistive navigation tasks in controlled urban environments. The proposed approach contributes to a scalable architecture that integrates a reinforcement-learning navigation policy within a multi-agent coordination framework and real-time sensor perception, providing a foundation for the development of intelligent and deployable assistive navigation systems. Full article

Background: Sleep disorders, particularly insomnia, represent a major public health concern, while currently available hypnotic drugs are often limited by adverse effects and poor long-term tolerability. Methods: In this study, we investigated the sleep-promoting effects of a mixture of Hypericum perforatum L. and Melissa officinalis L. extract (HME) and its underlying mechanisms in male ICR and C57BL/6 mice. In a pentobarbital-induced sleep model in mice, sleep onset latency and total sleep time were measured. Pharmacological studies using various antagonists and agonists were conducted to elucidate receptor-mediated mechanisms. Immunohistochemical and immunofluorescence analyses were performed to assess neuronal activity, and cortical mRNA expression was evaluated by quantitative analysis. HPLC analysis was used to identify the major constituents of HME, and their pharmacological profiles were functionally evaluated. Results: HME significantly reduced sleep onset latency and prolonged total sleep time. These hypnotic effects were shown to be mediated through adenosine and melatonin receptor signaling pathways. Immunohistochemical and immunofluorescence analyses showed that HME suppressed neuronal activity in wake-promoting cholinergic and orexinergic neurons of the basal forebrain and lateral hypothalamus, while enhancing activation of sleep-promoting GABAergic neurons in the ventrolateral preoptic nucleus. At the molecular level, HME increased cortical mRNA expression levels of adenosine A₁ receptor, adenosine A_2A receptor, melatonin receptor ₁, and melatonin receptor ₂. From the HPLC analysis, rosmarinic acid and hyperoside were identified as the major constituents of HME. Functional evaluation of these compounds revealed complementary pharmacological profiles, with hyperoside primarily acting through adenosine receptors and rosmarinic acid engaging both adenosine and melatonin receptor pathways. Conclusion: These findings suggest that HME enhances both sleep initiation and maintenance through adenosine and melatonin receptor signaling pathways, thereby supporting its potential as a multitarget therapeutic agent for improving sleep quality. Full article

Reconfigurable intelligent surface (RIS)-assisted unmanned aerial vehicle (UAV)-enabled mobile edge computing (MEC) has become an effective architecture for supporting computation-intensive and latency-sensitive applications by enabling flexible deployment and enhanced wireless coverage. However, when non-orthogonal multiple access (NOMA) is incorporated, the joint optimization of computation offloading, wireless resource allocation, RIS phase configuration, and UAV trajectory design becomes highly challenging owing to the strong coupling among decision variables, problem non-convexity, and time-varying system dynamics. To address these challenges, this paper investigates the energy consumption minimization problem in a NOMA-based RIS-assisted UAV-MEC system by jointly optimizing user offloading ratios, transmit power, UAV computing resource allocation, and flight trajectory. A long short-term memory (LSTM)-embedded proximal policy optimization (PPO) algorithm is developed to capture the temporal dependencies of system states and enable adaptive decision-making in dynamic environments. In addition, a closed-form phase conjugation-based optimal RIS configuration is derived and incorporated into the environment model to reduce the action space and improve training efficiency. The simulation results show that the proposed LSTM-PPO method converges faster and achieves lower energy consumption than conventional PPO, deep deterministic policy gradient (DDPG), and fixed offloading schemes, while exhibiting improved stability and scalability in the tested multi-user scenarios. These results demonstrate the effectiveness of combining temporal learning and model-assisted RIS optimization for energy efficient resource management in RIS-assisted UAV-MEC systems. Full article

Implementing cryptographic primitives on resource-constrained IoT devices involves tight latency, code-size, and energy budgets. This work proposes a general LLVM backend instruction-selection strategy that recognizes single-bit update idioms—typically expressed as LOAD–-(AND/OR)–-STORE sequences in SHA-256 and similar bit-oriented code—and lowers them to the most efficient target-specific bit-manipulation primitive when legality and cost conditions are met. As a concrete instantiation, we implement the strategy for the Renesas RL78/G23 ISA by rewriting eligible patterns into SET1/CLR1 instructions when the constant mask targets exactly one bit. We evaluate the resulting backend on an RL78/G23 platform using cycle counts and code size (bytes) across SHA-256-driven workloads motivated by firmware integrity checking, Merkle-tree hashing, HMAC-based authentication, password-based key derivation (PBKDF2), and chunk-level update validation. The observed cycle reductions are also converted to absolute time across the device’s supported on-chip oscillator frequencies to quantify latency impact under different clocking modes. The experimental validation in this work is limited to the RL78/G23 backend implementation. The underlying instruction-selection idea may be adaptable to other RL78-family devices or to other embedded architectures that provide equivalent single-bit set/clear or bitfield operations; however, such adaptations require target-specific legality checks, cost modeling, and separate experimental validation. Full article

Conventional adaptive active noise control (ANC) techniques, such as filtered-x normalized least mean square (FxNLMS), frequently run into issues when the noise environment changes, leading to longer reaction times. Moreover, fixed-filter approaches may lose the essential phase information necessary for efficient noise cancellation. This paper introduces 2D-BiSpecNet, a novel, effectively delayless feedforward active noise control system that uses a deep learning co-processor to address these difficulties. The technique converts one-dimensional audio signals into 64 × 64 bispectrum matrices, which transform sounds into visual representations. Therefore, it focuses on nonlinear quadratic phase couplings (QPCs), which provide robust and amplitude-independent views of the noise structure. The system then applies a quick multilabel classifier to examine these representations and immediately generates a control filter via 15 parallel subcontrol filters. The paper specifies a 5 × 5 convolutional receptive field that had the maximum efficacy. Simulations with real acoustic data indicate that this configuration yields an average noise reduction of −14.48 dB for aircraft noise, outperforming the usual FxNLMS technique by nearly 6 dB. The technology conducts classification and filtering nearly seven times faster than adaptive approaches, thus reducing convergence delays and delivering a more reliable and low-latency solution for noise-canceling environments. Full article

This study proposes a two-stage Intrusion Detection System (IDS) for Controller Area Networks (CAN) that leverages protocol-specific timing characteristics. Modern vehicular networks are vulnerable to injection attacks due to the CAN protocol’s lack of built-in authentication. Our methodology transforms raw CAN traffic into a structured feature space consisting of CAN IDs, message offsets, and inter-message intervals derived from the CAN Remote Frame request–response mechanism. The first stage applies unsupervised z-score statistical thresholding, requiring no labeled attack data. The second stage employs three independent binary Random Forest (RF) classifiers for precise characterization. Individual classifiers achieve F1-scores of 0.96 (Fuzzy), 0.77 (DoS), and 0.79 (Impersonation). In the integrated end-to-end pipeline, while the system effectively filters 97% of legitimate traffic, a performance stratification is observed: high detection is maintained for timing-disruptive attacks (Fuzzy), whereas timing-preserving attacks (DoS, Impersonation) exhibit lower recall due to the restrictive nature of the timing-only first-stage gating mechanism. Hardware profiling confirmed an inference latency of ∼0.018 ms and footprint of 8.8–19.2 MB, offering a deployable, computationally efficient defense for legacy automotive environments. Full article

Manual counting of surgical tools, known as surgical counting, is a time-consuming and error-prone task that increases the risk of retained surgical instruments and extends operating room (OR) time. Presently, in hospitals around the world, surgical counting is often performed manually with paper or tablet checklists, often leading to delays, increased infection risk, and financial cost. RFID, barcode-based, and computer vision solutions exist but are expensive and have challenges with sterilization and signal interference. This paper presents ToolSafe, a low-cost, portable system that classifies surgical tools by their acoustic signatures when dropped into a detection box. A pilot dataset of 4004 audio samples from four tool types (n = 996, tissue forceps; n = 1005, iris scissors; n = 1006, scalpel handle; n = 997, testing needle) was collected using ToolSafe. A convolutional neural network (CNN) was evaluated using stratified five-fold cross-validation on the laboratory dataset, with a k-nearest neighbors (KNN) classifier implemented as a control model. In each fold, both models were trained on 80% of the data and tested on the remaining 20%, ensuring that all samples were used for both training and evaluation. The CNN achieved a mean (±standard deviation) classification accuracy of 99.55% (±0.19%) across the validation folds, outperforming the KNN model, which achieved a mean accuracy of 97.28% (±0.50%). The difference was statistically significant according to a paired t-test across folds (p = 0.0003), indicating CNN’s superior performance on the dataset. For a run of 100 additional samples using the Raspberry Pi-based system, spectrogram generation averaged 0.121 s (±0.025 s), CNN inference averaged 0.180 s (±0.033 s), and total end-to-end latency averaged 1.851 s (±0.253 s) per tool. This pilot study proposes a possible technological solution for surgical counting that reduces human error and enhances patient safety. ToolSafe may be subsequently improved by increasing the number of surgical tools used in the training dataset, testing under more robust OR-like environments, and comparing to other classification algorithms. Further refinement and incorporation of ToolSafe in operating room workflows have the potential to reduce patient risks from extended surgical times and retained surgical instruments. Full article

This systematic review maps how Artificial Intelligence (AI) operationalises Material and Digital Product Passports (MP/DPP) for circular construction (January 2020–March 2026). However, most existing AI-to-passport implementations lack standardised reporting metrics, interoperability frameworks, and benchmarks for end-of-life decision support, leaving a critical operational gap that this review systematically addresses. The review follows the PRISMA 2020 guidelines to ensure methodological transparency and reproducibility. Of 2810 records identified across Web of Science, ScienceDirect, and Scopus, 49 peer-reviewed studies met the inclusion criteria for explicitly linking AI to MP/DPP under a circular economy lens. Evidence is synthesised across AI task families (computer vision, NLP including large language models, machine learning, knowledge graphs, digital twin/IoT), lifecycle phases (design, construction, operation, end-of-life), and circularity functions (traceability, lifecycle data enrichment, reuse/recycling readiness, recovery/EoL planning). The literature concentrates on traceability and lifecycle enrichment, while decision support for reuse and end-of-life remains sparse. Methodological weaknesses include narrow field validation, limited reporting of passport-level service metrics, and weak interoperability between AI pipelines and passport schemas. The regulatory landscape has intensified: Regulation (EU) 2024/3110 (Construction Products Regulation) entered into force in January 2025, and Regulation (EU) 2024/1781 (ESPR) launched its first product groups in April 2025, transforming AI-enabled MP/DPP from a prospective research topic into an immediate operational requirement. This review also notes the emergence of Large Language Models and blockchain as pivotal technologies for NLP-based field extraction and governed trust in twin-to-passport pipelines, respectively. Three framework elements are contributed and formalised as the Digital Circularity AI Framework (DCAF): (i) a minimum reporting bundle for AI-to-passport pipelines; (ii) a governance pack for twin-to-passport updates covering provenance, versioning, latency, and blockchain-trust; and (iii) open benchmark definitions for reuse grading, deconstruction sequencing, and residual value estimation. Together, these elements aim to shift MP/DPP from identification-oriented tools toward actionable decision support for circular recovery. Full article

Background: Prolonged exposure to moderate and loud noise is known to impair hearing; however, the safety of long-duration exposure to low-level sound, such as that encountered during everyday conversation, remains unclear. This study aimed to determine the effect of continuous exposure to sound at a 65 dB sound pressure level (SPL) on auditory processing. Methods: Auditory brainstem responses (ABRs) were recorded in C57BL/6 mice before and after a 1 h exposure to a continuous pure tone at 65 dB SPL. Changes in ABR thresholds, wave amplitudes, and latencies were analyzed across frequencies and time points. Correlations between amplitude and latency changes across ABR waves were also assessed. Results: Tone exposure induced a significant, frequency-specific increase in ABR thresholds, with a mean elevation of approximately 6 dB and a maximum shift of 15 dB. Significant reductions in amplitudes and prolongations of latencies were observed in Waves I–III, while Wave V amplitude remained relatively stable. A strong negative correlation between amplitude reduction and latency increase was found in Wave I, which progressively weakened from Wave II to Wave V. These functional changes persisted for up to three hours following exposure before gradually returning to baseline. Conclusions: Prolonged exposure to low-level sound at intensities typical of conversational speech can transiently impair auditory function and alter early neural processing in the auditory pathway. These findings suggest that sound levels commonly considered safe may still pose a risk when exposure is sustained, with implications for understanding hidden hearing loss and improving early diagnostic approaches. Full article

Aiming at the problems of weak background adaptive ability, high dependence on edge features, high computational complexity of some traditional license plate location algorithms, high deployment cost and strong training dependence of location model based on deep learning, this paper proposes a fast license plate location algorithm based on statistical color features. The algorithm uses the HSV color space as the main processing channel, and quantifies the regional color distribution characteristics by constructing the hue histogram and calculating its standard deviation and other statistics, which significantly improves the discrimination and illumination adaptability of the license plate mask in complex background. Compared with the lightweight deep learning models such as “You Only Look Once Version 12 Nano”, this algorithm does not need GPU acceleration and model loading, eliminates the need for data training, significantly reduces the deployment cost and complexity, and can run efficiently on the general computing platform. The experimental results show that compared with the YOLOv12n model, the average processing time of this algorithm is shortened by 30.81% (when YOLOv12n is evaluated with GPU) or 48.42% (when YOLOv12n is evaluated with CPU) at the cost of sacrificing about 5.8% positioning accuracy. The positioning accuracy still reaches 93.7%, demonstrating high processing efficiency and excellent platform adaptability. The algorithm has the advantages of being lightweight, efficient and interpretable, and is especially suitable for intelligent parking lots, edge devices and other scenes sensitive to real time, cost and energy consumption. Full article

To address the rapid growth of software vulnerabilities, the latency of manual expert classification, and the limitations of existing methods restricted to fixed categories, this paper proposes V2W-LLM, an automated vulnerability-to-weakness mapping model based on Large Language Models (LLMs). First, a dataset of CVE-CWE description pairs is constructed based on established expert correlations from MITRE. Subsequently, the LLM is instruction-tuned on this dataset to leverage its reasoning capabilities in generating CWE-style descriptive text for newly disclosed, unmapped vulnerabilities. Finally, using a BAAI-based embedding model, the semantic representations of the generated text and official CWE descriptions are computed to identify the optimal mapping via cosine similarity (Top-1). Experimental results indicate that V2W-LLM achieves an accuracy of 90.18% and a Macro-F1 of 87.64% in common categories. Furthermore, on the public ChatGPT-VDMEval and the latest 2024 NVD datasets, the model attains F1 scores of 86.02% and 94.02% respectively, validating its effectiveness in automating the vulnerability-to-weakness mapping process. Full article

Dense retrieval pipelines often rely on large candidate pools before reranking, making candidate generation and downstream scoring a practical bottleneck. This paper studies training-free binary projection filtering as a lightweight pre-filter for reducing the candidate set before dense reranking. Rather than presenting it as a universally superior retrieval method or a validated speedup technique, we ask a narrower practical question: how far can the candidate pool be reduced before average top-rank quality, retained relevance, and query-level reliability begin to break down? We evaluate the approach on five BEIR datasets: SciFact, NFCorpus, FiQA, ArguAna, and TREC-COVID. The revised evaluation compares exhaustive Dense retrieval, FAISS-HNSW, FAISS-IVF-Flat, and Binary+Dense retrieval, and includes projection-dimension ablations over $D_b\in \{128,256,512,1000\}$, candidate-budget ablations over $K\in \{50,100,200,500\}$, five-seed robustness analysis, and typo-perturbed queries. In addition to MRR@10, nDCG@10, and Recall@100, we report filter-stage metrics including Retained@K, catastrophic failure rate, and Best Relevant Survival. Across datasets, Binary+Dense often remains close to exhaustive Dense retrieval in average top-rank metrics at representative operating points, but the filter-stage behavior is strongly collection-dependent. Larger $D_b$ and K generally improve retained relevance and reduce catastrophic failures, but they also increase filtering cost or reduce the degree of pruning. The latency results show that structural candidate reduction does not translate into consistent end-to-end wall-clock speedup in the current Python 3.16/NumPy implementation. Taken together, the results suggest that training-free binary projection filtering is best understood as a calibration-sensitive pre-filter and failure risk analysis mechanism rather than as a replacement for Dense or ANN retrieval. Full article

The Internet of Vehicles (IoV) is rapidly emerging as a core component of intelligent transportation systems, enabling real-time communication among vehicles, infrastructure, and cloud platforms. However, the increasing interconnectivity of vehicular systems and the advancement of quantum computing introduce significant security challenges to existing cryptographic mechanisms. Conventional schemes such as RSA and Elliptic Curve Cryptography (ECC) are vulnerable to quantum attacks and are computationally inefficient for resource-constrained vehicular environments. To address these limitations, this paper proposes a Double-Ring Hybrid Sparse NTRU (DRH-SNTRU) framework, a lightweight and quantum-resistant cryptographic scheme for secure IoV communication. The proposed framework introduces three key enhancements: (i) controlled-support sparse polynomial structures to reduce polynomial multiplication complexity while improving entropy distribution; (ii) a double-ring algebraic architecture that separates key operations from message processing to enhance structural security and minimize coefficient leakage; and (iii) hybrid ephemeral keys derived from contextual entropy to strengthen forward secrecy and adaptive security. An optional ciphertext evaluation mechanism is further incorporated to detect malformed and replayed ciphertexts prior to decryption. Security analysis demonstrates that the proposed framework achieves IND-CPA security under the hardness assumption of the NTRU lattice problem and can be extended to resist chosen-ciphertext attacks through the integrated validation mechanism. Experimental benchmarking across polynomial dimensions N = 512 to 8192 demonstrates that DRH-SNTRU achieves low setup overhead below 3 μs, efficient decryption latency of approximately 305.64 μs at N = 8192, and compact sparse private key representation of only 117 bytes at higher dimensions. Compared with Standard NTRUEncrypt, NTRU-HRSS, and Ring-LWE Encryption, the proposed framework demonstrates improved decryption efficiency, lightweight storage overhead, and enhanced ciphertext integrity protection while maintaining practical scalability for resource-constrained post-quantum IoV environments. Full article

Accurate short-term traffic prediction is a key requirement for modern traffic management systems, yet many existing approaches remain focused on offline evaluation and do not address the challenges of continuous real-time deployment. In this work, we present a context-aware spatiotemporal graph convolutional network (STGCN) framework designed for low-latency, scalable traffic forecasting under operational conditions. The proposed approach integrates structural information from the road network, temporal regularities derived from historical data, and a residual correction mechanism trained on systematic prediction errors observed during real-time operation. The framework is designed to remain lightweight, enabling continuous minute-level inference without computational overhead that would hinder long-term deployment. The methodology is evaluated in two real-world case studies of different scale and complexity. In Thessaloniki, Greece, multiple forecasting models are evaluated across different temporal resolutions using one-minute speed data, with the proposed STGCN selected for real-time deployment. A residual correction module trained on historical prediction errors further improves real-time forecasting accuracy compared to the baseline STGCN deployment. Scalability is further demonstrated in the South Holland region of the Netherlands, where the same architecture is applied to a larger network and extended to multi-horizon forecasting. Results show that the proposed framework achieves competitive predictive performance while maintaining low computational cost, and that incorporating residual error learning provides a robust and practical solution for improving forecasting accuracy in real-world deployments. These findings highlight the importance of combining domain-specific modeling with operational considerations in traffic prediction systems. Full article