Bài báo quốc tế
Lĩnh vực nổi bật
Trí tuệ nhân tạo
1.046
Internet vạn vật
23
An toàn bảo mật
11
Blockchain
3
Tìm kiếm theo
Năm xuất bản
Lĩnh vực bài báo
Kho tri thức
/
Bài báo quốc tế
Sắp xếp theo
Liên quan
Optimizing Detection Tasks with PySwarms: Applications of Paging Algorithms
Châu Văn VânEfficient detection in large-scale systems often requires balancing accuracy, computational cost, and memory management. This paper explores the integration of paging algorithms with particle swarm optimization (PSO), implemented via the PyCharm's library, to address detection tasks. By adopting paging-inspired strategies, candidate solutions are dynamically loaded and updated, reducing redundant computations and improving convergence efficiency in high-dimensional search spaces. Experimental results demonstrate that the paging–PSO hybrid framework enhances detection accuracy while lowering runtime overhead compared to standard PSO implementations. The proposed approach shows promise for resource-constrained environments, where optimization must be achieved under limited memory and processing capacity. This work highlights the potential of combining classical operating system principles with swarm intelligence, offering a novel pathway for efficient optimization in detection and classification problems.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
An Improvement of MVDR Beamformer under Adverse Situation
Quản Trọng ThếNowadays, the use of microphone array beamformer has been widely commonly due to its
convenience of steering the beampattern on the specified sound location and attenuating the
background noise field. Microphone array beamforming own the capability of suppressing
interference, third-party speakers and different adverse noise fields with high directivity index
of extracting the clean speech data without speech distortion. Minimum Variance Distortionless
Response beamformer bases on the constrained criteria of minimizing the total output noise
power while saving the target talker by ensurign the beampattern equals one at certain direction.
However, under realistic recording environments, due to the movement of speaker during
conversations, the error of start time recording, the different sensitivities of microphones, the
error of sampling frequency, the internal electrical acoustic equipments, the inaccurate
distribution of microphones, the overall beamformer’s performance often degraded. The
unacceptable surrounding noise level or speech distortion corrupt the speech intelligibility of
output signal. In the article, the author proposed two-stage - based method for adaptively
updating the smoothing parameters for increasing the beamformer’s evaluation in real-life
scenarios. The numerical simulation has shown the improvement of reducing the speech
distortion to 5 dB, removing the background noise level to 12.9 dB and increasing the speech
quality in the term of signal-to-noise ratio from 5.3 to 8.8 dB.
Năm:2026
|
Chủ đề: Kỹ thuật thông tin
|
Nhà xuất bản: Journal of Cyber Security
The impacts of optical display BaF2-Ce materials on solid-state lighting
Luu Hong QuanTransparent ceramic doped with barium fluorid cerium (BaF2-Ce) was created via a sintering method and its brightness and scintillation
characteristics were examined. The luminescence is associated with the 5d4f transitions in the Ce3+ ion and exhibits emitting maxima at 310 and 323 nm. For Na-22 radioisotopes, photo-maximum at 511 keV and 1274 keV were achieved using translucent ceramic BaF2-Ce. The translucent ceramic BaF2-Ce has been determined to have a power resolution of 13.5% at 662 keV. A luminescent production rate was measured for the BaF2-Ce (0.2%) ceramic, which is similar to sole crystal. Calculations of the scintillation degradation period beneath 662 keV gamma stimulation reveal a quick part of 58 ns and a somewhat sluggish part of 434 ns. The more gradual part in BaF2-Ce(0.2%) ceramic is linked to the dipole-dipole power transmission from the host structure to the Ce3+ luminous core and is quicker comparing to self-trapped excitons (STE) emitting in BaF2 host. BaF2-Ce offer various qualities, including significant illumination output, rapid degradation duration, and rapid scintillating reaction, which are desirable for many global fields such as medicine, radiation detection, industrial systems and
nuclear safety
Năm:2025
|
Chủ đề: Kỹ thuật điện
|
Nhà xuất bản: International Journal of Reconfigurable and Embedded Systems (IJRES)
Optimizing predictive accuracy in general medical exams using hybrid machine learning and metaheuristic optimization methods
Nguyễn Minh TuấnThis study presents a hybrid, metaheuristic-driven optimization framework for
power hyperparameter tuning in predictive modeling based on large-scale an
nual health examination data. Different from conventional grid and random
search strategies, the proposed method directly incorporates particle swarm
optimization, artificial bee colony, and gravitational search algorithm into the
training pipeline of multiple machine learning models, enabling adaptive ex
ploration of high-dimensional parameter spaces under clinical data constraints.
The approach was evaluated on a comprehensive dataset comprising 93 clin
ical attributes and 1,000 patient records, with a specific focus on ischemic
stroke risk prediction. Random Forest, decision tree, support vector machine,
and logistic regression models were optimized using the proposed hybrid struc
ture and benchmarked against baseline configurations. Experimental results
demonstrate consistent and statistically significant reductions in mean squared
error, mean absolute error, and root mean squared error, alongside improve
ments in R2 and classification accuracy exceeding 99% for optimized logistic
regression models, while maintaining computational efficiency suitable for rou
tine clinical deployment. Beyond performance gains, the study introduces a
stacked ensemble architecture guided by metaheuristic-tuned base learners,
enhancing model robustness and generalization across training and indepen
dent test sets. These findings demonstrate the practical novelty of integrating
swarm and numerical optimization into clinical predictive pipelines, providing
a scalable and domain-agnostic solution for high-accuracy risk decision support
in preventive healthcare and other data-intensive applications.
Năm:2026
|
Chủ đề: Khoa học máy tính và thông tin
|
Nhà xuất bản: International Journal of Optimization and Control: Theories & Applications
Covert Communication and Physical Layer Security in Hybrid Power-Frequency Multiple Access With Friendly Jamming: Performance Analysis and Deep Learning Evaluation
Nguyễn Quang SangCovert communication and physical layer security (PLS) have emerged as two essential pillars for safeguarding wireless networks against increasingly sophisticated eavesdropping and surveillance threats. This paper investigates a downlink hybrid power–frequency multiple access (HPMA) system aided by a friendly jammer, where a single source communicates with two legitimate users in the presence of a passive eavesdropper. By jointly exploiting the power and frequency domains, the proposed HPMA scheme achieves clear advantages in terms of covertness and secrecy performance compared to conventional orthogonal multiple access (OMA) and non-orthogonal multiple access (NOMA) schemes. For performance analysis, we derive closed-form and approximated expressions for key metrics, including detection error probability (DEP), secrecy outage probability (SOP), effective secrecy throughput (EST), and average secrecy capacity (ASC), which reveal important trade-offs among resource allocation, jammer power, and secrecy reliability. To further reduce the computational burden of conventional analysis, a regression-based deep neural network (DNN) framework is developed to efficiently predict the secrecy performance of the considered system. Extensive Monte Carlo simulations validate the analytical results and confirm the superiority of the proposed HPMA scheme with friendly jamming. These findings provide useful insights for the design of future secure and covert wireless systems with low-complexity and intelligent physical-layer security requirements.
Năm:2026
|
Chủ đề: Kỹ thuật điện tử và viễn thông
|
Nhà xuất bản: IEEE Internet of Things Journal
Intelligent UAV Positioning and Fair Power Allocation via DRL in Cloud-Impaired HAP-to-UAV FSO/RF Systems
Nguyễn Quốc HuyThis paper proposes an adaptive resource allocation strategy for a non-terrestrial network (NTN) architecture, where a high-altitude
platform (HAP) provides free-space optical (FSO) backhaul links, and unmanned aerial vehicles (UAVs) serve as radio-frequency (RF) access nodes for end users. The goal is to ensure a fair and efficient allocation of transmission capacity among users. However, atmospheric turbulence and cloud dynamics in FSO links can severely degrade system performance. To address this issue, this work integrates deep reinforcement learning (DRL) to optimize UAV positioning in real time, enabling UAVs to avoid cloud-blocked regions and maintain stable connectivity for users. Based on channel states and user rate requirements, the proposed method decouples subchannel assignment from power allocation and incorporates DRL-based UAV trajectory control, forming a comprehensive optimization framework. The numerical results demonstrate that the proposed approach significantly improves system throughput, maintains user fairness, and reduces transmit power requirements compared to conventional methods.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Digital transformation solution implementation risk in logistics and supply chain industry
Trần Thanh HươngThis study explores the risks associated with digital transformation (DT) solution implementation in the logistics and supply chain management industry, employing a hybrid PLS-SEM-ANN-fsQCA methodology. Despite growing DT adoption, limited research integrates hybrid methodologies to assess risks in emerging markets like Vietnam, where resource constraints amplify vulnerabilities. Anchored in the Technology-Organization-Environment framework with human factors, the research investigates how technological, organizational, environmental, and human elements shape DT solution adoption and associated risks, including financial, operational, cybersecurity, and reputational risks. Results indicate that TOE + H factors significantly impact DT implementation, yet risks are amplified by misalignment, ineffective management, volatile market conditions, and limited digital literacy. Notably, DT solution adoption heightens cybersecurity vulnerabilities. Moderation analyses reveal that high digital literacy, larger firm size, and regulatory compliance mitigate these risks. ANN analysis highlights non-linear relationships, identifying technological and human factors as primary drivers, while fsQCA uncovers configurations, such as strong technological and human factor alignment, linked to successful DT outcomes. Importance-Performance Map Analysis emphasizes prioritizing technological and human factors for efficient resource allocation, promoting sustainable DT by minimizing disruptions and resource inefficiencies. This study enriches the TOE + H framework, providing actionable insights for logistics SMEs to enhance performance, resilience, and environmental sustainability in supply chains.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản: Discover Sustainability
CONTROL SYSTEM USING MULTI-PERIOD DISTURBANCE OBSERVER FOR MINIMUM-PHASE SYSTEMS
Siripong SangsarpanBài báo này xem xét thiết kế hệ thống điều khiển cho các hệ thống pha tối thiểu nhằm đảm bảo đầu ra bám sát tín hiệu đầu vào tham chiếu không tuần hoàn và các nhiễu tuần hoàn được giảm thiểu mà không cần điều khiển lặp lại. Trong các ứng dụng thực tế, hệ thống điều khiển thường phải giảm thiểu nhiễu tuần hoàn và làm cho đầu ra bám sát tín hiệu đầu vào tham chiếu không tuần hoàn. Để đáp ứng các yêu cầu này, điều khiển lặp lại đã được đề xuất. Điều khiển lặp lại có thể giảm thiểu nhiễu tuần hoàn. Tuy nhiên, điều khiển lặp lại thường dẫn đến bộ điều khiển bậc cao. Để thiết kế bộ điều khiển bậc thấp có khả năng giảm thiểu nhiễu tuần hoàn, hệ thống điều khiển phải được thiết kế mà không sử dụng điều khiển lặp lại. Trong bài báo này, hệ thống điều khiển sử dụng bộ quan sát nhiễu đa chu kỳ để giảm thiểu nhiễu tuần hoàn được đề xuất. Bộ quan sát nhiễu đa chu kỳ sử dụng bộ lọc bậc thấp và giảm thiểu nhiễu tuần hoàn bằng cách sử dụng chu kỳ của nhiễu. Vì bộ quan sát nhiễu đa chu kỳ sử dụng bộ lọc bậc thấp, nên việc thiết kế bộ điều khiển bậc thấp trở nên khả thi. Tuy nhiên, chưa có nghiên cứu nào về phương pháp thiết kế hệ thống điều khiển sử dụng bộ quan sát nhiễu đa chu kỳ. Để thiết kế bộ điều khiển bậc thấp nhằm giảm thiểu nhiễu định kỳ, hàm truyền từ nhiễu đến đầu ra phải có số cực hữu hạn. Điều kiện để hàm truyền từ nhiễu đến đầu ra có số cực hữu hạn được làm rõ. Ngoài ra, điều kiện ổn định nội tại của hệ thống điều khiển sử dụng bộ quan sát nhiễu đa chu kỳ làm cho số cực hữu hạn cũng được làm rõ. Dựa trên các điều kiện trên, một phương pháp thiết kế hệ thống điều khiển sử dụng bộ quan sát nhiễu đa chu kỳ cũng được đề xuất.
Năm:2026
|
Chủ đề: Kỹ thuật điện tử và viễn thông
|
Nhà xuất bản: International Journal of Innovative Computing, Information and Control
RDD-SPA: An efficient visual recognition algorithm for robot-assisted rapeseed pest control systems
Quanshu SongRuộng bậc thang trồng cải dầu đặt ra những thách thức dai dẳng trong việc kiểm soát sâu bệnh do địa hình phức tạp, mức độ cơ giới hóa thấp và khả năng tính toán hạn chế của robot phun thuốc di động. Để giải quyết những hạn chế này, nghiên cứu này đề xuất RDD-SPA, một khung nhận thức-điều khiển nhẹ tích hợp bộ phát hiện đối tượng hướng triệu chứng (RDD-YOLO) với Bộ điều hợp điểm phun (SPA) để phun thuốc chính xác theo thời gian thực trong môi trường ruộng bậc thang. Một tập dữ liệu thực địa gồm 3309 hình ảnh được thu thập ở giai đoạn cây con và nụ đã được xây dựng, chứa 13.530 trường hợp được chú thích về cây cải dầu, lỗ lá và sự đổi màu lá, phản ánh tán cây dày đặc, hình thái lá không đều và sự che khuất thường xuyên. RDD-YOLO cải tiến YOLO11 thông qua việc giới thiệu các mô-đun Lite-PSConv, C3k2-IDC và hàm mất mát SIoU, đạt được mAP50:95 là 64,0% đồng thời giảm số tham số và chi phí tính toán lần lượt là 12,02% và 3,02%. Sau quá trình cắt tỉa và chắt lọc kiến thức, mô hình được nén xuống còn 1,12 triệu tham số và duy trì khả năng suy luận thời gian thực ổn định ở tốc độ 60 khung hình/giây trên nền tảng nhúng AX650N với tốc độ robot thực tế là 0,18 m/giây. Dựa trên ước lượng độ sâu bằng kính hai mắt và hiệu chuẩn hình học, SPA chuyển đổi kết quả phát hiện thành các lệnh phun thích ứng, tự động điều chỉnh vị trí vòi phun, góc phun và lượng phun. Các thí nghiệm thực địa trên 40 cây cải dầu non cho thấy hệ thống đề xuất làm tăng khả năng giữ nước trong lá lên 33,3% và giảm sự phát tán hóa chất xuống 27,1% so với phun thủ công. Những kết quả này cho thấy RDD-SPA cung cấp một giải pháp hiệu quả và khả thi cho việc ứng dụng thuốc trừ sâu chính xác trong nông nghiệp bậc thang hạn chế nguồn lực.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản: Industrial Crops and Products
Inexact projection methods for variational inquality problems and applications
Đỗ Duy ThànhIn this paper, we propose two new inexact projection algorithms, which can be easily implemented, for solving pseudomonotone variational inequality problems based on self-adaptive step sizes, viscosity technique, and inexact projections. We obtain two strongly convergent theorems of solutions in a real Hilbert space. Numerical experiments illustrate and compare the performances of the proposed algorithms with three other known results.
Năm:2026
|
Chủ đề: Toán học và thống kê
|
Nhà xuất bản: Applied Set-Valued Analysis and Optimization
Optimizing Mixed FSO-RF Downlink Systems With Active RIS and SLIPT-Enabled UAV-BSs
Vũ Bình MinhThis paper investigates the integration of active reconfigurable intelligent surfaces (ARISs) with uncrewed aerial vehicles (UAVs) in a mixed free-space optics radio frequency (FSO-RF) downlink communication system, enabling simultaneous lightwave information and power transfer (SLIPT). The proposed architecture addresses key challenges in UAV-based networks, including limited endurance and backhaul constraints, by allowing the UAV to harvest energy from the optical backhaul while transmitting RF signals enhanced via ARIS to ground users. The system design aims to maximize the minimum achievable rate among users by jointly optimizing the UAV’s beamforming strategy, 3D placement, ARIS reflection coefficients, optical ground station (OGS) transmit power and the power splitting (PS) ratio at the UAV. An alternating optimization framework is developed to decompose the resulting non-convex problem into efficiently solvable subproblems using inner approximation techniques. Simulation results confirm that the proposed approach significantly outperforms baseline schemes, such as passive RIS, fixed UAV deployment, and static PS configurations, delivering improved rate fairness and energy efficiency. These results demonstrate the potential of ARIS-assisted SLIPT-enabled UAVs to support robust and sustainable downlink communications in next-generation wireless networks.
Năm:2026
|
Chủ đề: Kỹ thuật điện tử và viễn thông
|
Nhà xuất bản: IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 74, 2026
3D Dynamic Radio Map Prediction Using Vision Transformers for Low-Altitude Wireless Networks
Nguyen Duc Minh QuangLow-altitude wireless networks (LAWN) are rapidly
expanding with the growing deployment of unmanned aerial vehicles (UAVs) for logistics, surveillance, and emergency response.
Reliable connectivity remains a critical yet challenging task due
to three-dimensional (3D) mobility, time-varying user density, and
limited power budgets. The transmit power of base stations (BSs)
fluctuates dynamically according to user locations and traffic
demands, leading to a highly non-stationary 3D radio environment. Radio maps (RMs) have emerged as an effective means to
characterize spatial power distributions and support radio-aware
network optimization. However, most existing works construct
static or offline RMs, overlooking real-time power variations and
spatio-temporal dependencies in multi-UAV networks. To overcome this limitation, we propose a 3D dynamic radio map (3DDRM) framework that learns and predicts the spatio-temporal
evolution of received power. Specially, a Vision Transformer (ViT)
encoder extracts high-dimensional spatial representations from
3D RMs, while a Transformer-based module models sequential
dependencies to predict future power distributions. Experiments
unveil that 3D-DRM accurately captures fast-varying power
dynamics and substantially outperforms baseline models in both
RM reconstruction and short-term prediction
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Towards Universal Segmentation for Log Parsing
Lê Văn HoàngLog parsing is a crucial step in log analysis, as it transforms unstructured log messages into structured data required by various downstream analysis tasks. The sheer volume of log data generated by modern software systems motivates the development of numerous log parsing techniques in the literature. However, existing log parsers still suffer from unsatisfactory accuracy, which may significantly affect the follow-up analysis such as log-based anomaly detection. We have identified two main limitations that hinder the effectiveness of existing log parsing methods: (1) under-segmentation: most log parsers leverage a fixed, predefined set of delimiters to separate a log message into a set of tokens, which may fail to split log messages correctly due to the heterogeneity of logging formats; (2) over-segmentation: using too many delimiters may lead to the over-segmentation issue, which fragments meaningful units in log messages and makes it difficult to accurately identify templates and parameters. To address these limitations, we propose SCLog, a novel syntax- and contextual-aware segmentation approach for log parsing. SCLog leverages a comprehensive set of syntax-based heuristics to segment log messages into coarse-grained tokens. To further tokenize log messages into fine-grained tokens, SCLog mines the structural patterns of tokens based on their surrounding contexts to identify the optimal delimiters for each token dynamically. We evaluate SCLog on widely-used, large-scale Loghub-2.0 datasets. The results demonstrate that SCLog significantly outperforms state-of-the-art log parsers in terms of parsing accuracy and robustness across diverse datasets.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Optimization of Mixed FSO-RF Downlink Systems With SLIPT-Enabled UAV Base Station and HAP-Mounted Optical RIS
Vũ Bình MinhIn this paper, we investigate a mixed free-space optics (FSO)-radio frequency (RF) downlink communication system enhanced by a simultaneous lightwave information and power transfer (SLIPT)-enabled uncrewed aerial vehicle base station (UAV-BS) and a high-altitude platform (HAP)-mounted optical reconfigurable intelligent surface (ORIS). We develop overall optimization frameworks for end-to-end systems aimed at minimizing transmitted power at optical ground station (OGS) for backhaul link or maximizing the multiuser fairness rate in the access link. This is achieved by jointly optimizing the UAV-BS placement, beamforming strategy, OGS power allocation, and power-splitting (PS) ratio under energy efficiency and quality of service (QoS) constraints. The formulated optimization problems are highly non-linear and non-convex, posing significant challenges for conventional methods. To address this, we employ inner approximation techniques and propose an iterative algorithm to achieve efficient solutions. Simulation results demonstrate substantial performance gains compared to conventional benchmarks, highlighting the efficacy of the proposed approach.
Năm:2026
|
Chủ đề: Kỹ thuật điện tử và viễn thông
|
Nhà xuất bản: IEEE ACCESS, VOLUME14, 2026
Joint Access Point Activation and Power Allocation for Cell-Free Massive MIMO Aided ISAC Systems
Nguyen Xuan TungCell-free massive multiple-input multiple-output (MIMO)-
aided integrated sensing and communication (ISAC) systems are investigated where distributed access points jointly serve users and sensing
targets. We demonstrate that only a subset of access points (APs) has
to be activated for both tasks, while deactivating redundant APs is essential for power savings. This motivates joint active AP selection and
power control for optimizing energy efficiency. The resultant problem is
a mixed-integer nonlinear program (MINLP). To address this, we propose a model-based Branch-and-Bound approach as a strong baseline to
guide a semi-supervised heterogeneous graph neural network (HetGNN)
for selecting the best active APs and the power allocation. Comprehensive
numerical results demonstrate that the proposed HetGNN reduces power
consumption by 20 − 25% and runs nearly 10, 000 times faster than
model-based benchmarks
Năm:2026
|
Chủ đề: Kỹ thuật điện tử và viễn thông
|
Nhà xuất bản: IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY
Survey data on digital competence assessment among pre-service teachers in Vietnam
Nguyễn Hà NamIn the context of digital transformation in education, digital competence is one of the significant essential requirements for future teachers. This study surveyed and analyzed the digital competence structure of 1439 pre-service teachers in different regions of Viet Nam. We utilized a self-assessment questionnaire based on the Digital Kids Asia-Pacific (DKAP) framework, with references to TPACK (Technological Pedagogical Content Knowledge), DigComp (Digital Competence Framework for Citizens), and DigCompEdu frameworks. The dataset provided detailed information on each participant’s self-evaluated digital proficiency in five categories, along with demographic variables such as gender and subject specialization. The core of this data file locates itself in such potential to inform teacher training programs and educational policy by offering evidence on prowess and weakness in future teachers’ digital competence.
Năm:2026
|
Chủ đề: Kỹ thuật điện tử và viễn thông
|
Nhà xuất bản: Data in Brief
Performance of Modified Fractional Frequency Reuse Algorithm in Random Ultra Dense Networks
Bach Hung LuuMitigating intercell interference by employing fractional frequency reuse algorithms is one of the important approaches to improving user performance in 5G and Beyond 5G cellular network systems, which typically have a high density of Base Stations (BSs). While most frequency reuse algorithms are based on the downlink Signal-to-Interference-plus-Noise Ratio (SINR) or the distance between the user and its serving BS to classify Cell-Edge Users (CEUs) and Cell-Center Users (CCUs), this paper discusses a modified algorithm that uses the power ratio between the signal strengths from the serving BS and the second nearest BS for user classification. Specifically, if the power ratio is below a predefined threshold, the user is classified as a CEU and is served with higher transmission power. Simulation results show that increasing transmission power is necessary to enhance CEU performance, but it also degrades the performance of typical users. The use of frequency reuse algorithms is particularly feasible in environments with a high density of obstacles, where intercell interference can be effectively suppressed.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
A Redundancy-Free Object-Aware Window Slicing Algorithm for Insect Detection
Van Quyet NguyenInsect pests on yellow sticky traps are notoriously difficult to detect automatically because each specimen occupies only a few dozen pixels and is often surrounded by a vast, feature-poor background. This study proposes a fully automated pipeline that combines (i) HSV-based trap isolation, (ii) self-bootstrapped pseudo-labelling with YOLOv8 and YOLOv11, and (iii) a slicing algorithm that converts every full-resolution frame into non-overlapping 640 × 640 patches, each containing a set of un-truncated insects. On the revised Yellow Sticky Traps dataset (8,114 annotations, three classes), this slice-based training strategy markedly improves performance. For YOLOv8, mAP@[0.5:0.95] rises from 0.41 to 0.77, with precision/recall increasing from 0.82/0.77 to 0.92/0.93. For YOLOv11, mAP@[0.5:0.95] increases from 0.46 to 0.89 (a 92% relative gain), while precision and recall reach 0.96. Confusion matrix analysis confirms that slice-based models virtually eliminate background bleed-through and inter-species misclassification, problems that severely affect full-image baselines. The pipeline requires no manual annotation beyond an initial calibration phase, making it well-suited for large-scale, real-time pest surveillance in commercial greenhouses and open-field agriculture.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Smarter Healthcare: Assessing Nurse Workload using Wearable Sensors and Machine Learning
Viet Hoan BuiNursing care activity recognition is vital for optimizing workflows and reducing documentation burden in elderly care. The SONAR dataset, recently released as a benchmark for nursing activities and recorded with wearable inertial sensors in a clinical environment, poses unique challenges due to noise, class imbalance, and fine-grained activity definitions. This study presents a systematic evaluation of classical ensemble learning methods on SONAR under deployment-oriented constraints for resource-limited microcontrollers. We assess the impact of sensor placement and model configuration on recognition performance. Among the evaluated algorithms, XGBoost consistently outperformed Extra Trees, LightGBM, and Random Forest, achieving 64.01 % accuracy and 61.21 % weighted F1. These findings highlight both the potential and the remaining challenges in developing reliable and resource-efficient nursing activity recognition systems.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
ECA-CBAM-MobileNetV3: An Enhanced Model for Rice Leaf Disease Classification
Thi Kim Cuc NguyenIn response to the growing demand for early and accurate detection of rice leaf diseases, this study proposes an improved deep learning model based on MobileNetV3-Small, enhanced with an integrated ECA-CBAM attention module. This module combines Efficient Channel Attention ECA to model inter-channel dependencies and a modified spatial attention mechanism using dilated convolutions to capture broader contextual information without increasing computational complexity. The model is trained on a dataset of 10,407 manually labeled rice leaf images, employing selective fine-tuning and agriculture-specific data augmentation strategies. Experimental results show that the proposed ECA-CBAM-MobileNetV3-Small model achieves an accuracy of 95.05% and an F1-score of 94.62%, significantly outperforming both the baseline MobileNetV3-Small and the CBAM-only variant. These findings highlight the effectiveness of combining lightweight attention mechanisms with dilation-based enhancements for improving plant disease classification performance.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Comparing XAI Methods for Identifying Critical Features and Expert Contributions in Multi-Expert Edge Computing Attack Detection
MINH HOANG NGUYENThis paper presents a comprehensive comparison of XAI (Explainable AI) techniques applied to analyze the influence of experts and features in a multi-expert feature selection system for intrusion detection systems deployed on edge computing platforms. The study focuses on four prominent XAI methods — SHAP, LIME, Integrated Gradients, and DeepLIFT — to evaluate their effectiveness in explaining feature importance and the contribution of component experts in the decision-making process of the multi-expert system. The comparison is based on quantitative metrics: Hamming Distance, Spearman Rank, Pearson Correlation, Jaccard Index, and Cosine Similarity, which help analyze the level of consistency and differences between explanations or selected feature sets. The experimental results provide deep insights into the advantages and limitations of each method, thereby suggesting the selection of appropriate explanation techniques that meet the specific requirements of machine learning-based intrusion detection systems in edge computing environments.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
RAC : Few-shot fruit recognition through CLIP-based ambiguity reduction
Trần Anh ĐạtCounterfeit agricultural products pose significant challenges for conventional classification models like CLIP, particularly under few-shot learning scenarios due to subtle visual ambiguities. This study introduces RAC (Reducing Ambiguity in CLIP for Counterfeit Detection), a novel framework designed to identify and mitigate these ambiguities through an active refinement process. Specifically, RAC employs a Multilayer Perceptron Plus (MLP-P) module to enhance discriminative power by effectively synthesizing multi-level visual representations derived from CLIP’s image encoder. Subsequently, the Inter-class Ambiguity Reduction (IRA) module suppresses specific visual patterns responsible for confusion between authentic and counterfeit products. Finally, a Network Fusion (NF) stage dynamically combines outputs from these enhanced pathways to ensure robust classification outcomes. Experimental validation demonstrates that RAC markedly improves detection accuracy, achieving an average of 76.95% in 16-shot learning across four benchmarks, surpassing the state-of-the-art by 2.50%. On the specialized TFS-Fruit dataset, the framework reaches 80.0% accuracy, confirming its efficacy in critical agricultural applications.
Năm:2026
|
Chủ đề: Khoa học dữ liệu và trí tuệ nhân tạo
|
Nhà xuất bản: Neural Networks
Advancing APT detection through transformer-driven feature learning and synthetic data generation
Le Tran Kim DanhAdvanced Persistent Threat (APT) detection based on artificial intelligence (AI) platforms has emerged as a dominant trend, has attracted increasing attention in cybersecurity. Nevertheless, two major challenges remain: (i) how to effectively extract discriminative features from complex network traffic flows, and (ii) how to address severe class imbalance caused by the rarity of APT attacks. To tackle these challenges, we propose an integrated pipeline/framework named ET-SDG. The ET-SDG model integrates Transformer-based Feature Learning with a Conditional Generative Model for Synthesis (CGMS). Specifically, the Transformer-based feature learning component combines the ExtraTrees algorithm with a Transformer architecture to select, aggregate, and encode informative flow-level features. To mitigate data imbalance, ET-SDG incorporates CGMS, a cGAN-based synthetic data generation module designed for data augmentation of minority APT traffic. By conditioning the generation process on class labels, CGMS synthesizes representative minority-class samples, aiming to improve the robustness and generalization of the downstream detection model under class imbalance. Across the evaluated benchmarks, ET-SDG shows competitive results and provides modest improvements (approximately 1–4% points, depending on the dataset and metric) relative to the compared baselines.
Năm:2026
|
Chủ đề: Khoa học kỹ thuật và công nghệ khác
|
Nhà xuất bản: Scientific Reports
A Novel Model for Detecting Website Defacements Based on BERT and BiLSTM
Hoàng Xuân DậuOver the past decade, web defacement and related forms of web-based attacks have emerged as significant security threats to enterprise and organizational systems. A single defacement incident can result in severe consequences for the affected party, including immediate disruption of website functionality, reputational damage, and substantial financial losses. To mitigate these risks, various monitoring and detection mechanisms have been developed. However, many of the existing approaches are constrained in scope: some are restricted to static web content, while others can process dynamic content but at the cost of substantial computational overhead. Furthermore, current solutions often suffer from limited detection accuracy and elevated false positive rates, largely due to the inadequate handling of key webpage elements such as plain text content. To address these limitations, this study introduces a novel web defacement detection model leveraging BERT and BiLSTM architectures, with a focus on features extracted from webpage plain text. Extensive experiments conducted on a dataset comprising nearly 70,000 webpages demonstrate that the proposed model achieves superior performance compared to prior approaches, attaining an overall accuracy of 98.61% and an F1-score of 98.70%.
Năm:2026
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản: