Search Results (76)

Deep learning has achieved remarkable success in hyperspectral image (HSI) classification, attributed to its powerful feature extraction capabilities. However, existing methods face several challenges: Convolutional Neural Networks (CNNs) are limited in modeling long-range spectral dependencies because of their limited receptive fields; Transformers are constrained by their quadratic computational complexity; and Mamba-based methods fail to fully exploit spatial–spectral interactions when handling high-dimensional HSI data. To address these limitations, we propose MRFP-Mamba, a novel Multi-Receptive-Field Parallel Mamba architecture that integrates hierarchical spatial feature extraction with efficient modeling of spatial–spectral dependencies. The proposed MRFP-Mamba introduces two key innovation modules: (1) A multi-receptive-field convolutional module employing parallel $1\times 1$, $3\times 3$, $5\times 5$, and $7\times 7$ kernels to capture fine-to-coarse spatial features, thereby improving discriminability for multi-scale objects; and (2) a parameter-optimized Vision Mamba branch that models global spatial–spectral relationships through structured state space mechanisms. Experimental results demonstrate that the proposed MRFP-Mamba consistently surpasses existing CNN-, Transformer-, and state space model (SSM)-based approaches across four widely used hyperspectral image (HSI) benchmark datasets: PaviaU, Indian Pines, Houston 2013, and WHU-Hi-LongKou. Compared with MambaHSI, our MRFP-Mamba achieves improvements in Overall Accuracy (OA) by 0.69%, 0.30%, 0.40%, and 0.97%, respectively, thereby validating its superior classification capability and robustness. Full article

In recent years, deep learning techniques have become increasingly prominent in face recognition tasks, particularly through the extraction and classification of face vectors. These vectors enable the inference of demographic attributes such as gender, age, and ethnicity. This study introduces a gender classification approach based solely on face vectors, avoiding the use of traditional machine learning algorithms. Face embeddings were generated using three popular models: dlib, ArcFace, and FaceNet512. For classification, the Average Neural Face Embeddings (ANFE) technique was applied by calculating distances between vectors. To improve gender recognition performance for Asian individuals, a new dataset was created by scraping facial images and related metadata from AsianWiki. The experimental evaluations revealed that ANFE models based on ArcFace achieved classification accuracies of 93.1% for Asian women and 90.2% for Asian men. In contrast, the models utilizing dlib embeddings performed notably lower, with accuracies dropping to 76.4% for women and 74.3% for men. Among the tested models, FaceNet512 provided the best results, reaching 97.5% accuracy for female subjects and 94.2% for males. Furthermore, this study includes a comparative analysis between ANFE and other commonly used gender classification methods. Full article

ECG monitoring is central to the early detection of cardiac abnormalities. We compared 28 manually selected signal-specific features with 159 automatically extracted signal-independent descriptors from the MIT BIH Arrhythmia Database. ANOVA reduced features to the 10 most informative attributes, which were evaluated alone and in combination with the signal-specific features using Random Forest, SVM, and deep neural networks (CNN, RNN, ANN, LSTM) under an interpatient 80/20 split. Merging the two feature groups delivered the best results: a 128-layer CNN achieved 100% accuracy. Power profiling revealed that deeper models improve accuracy at the cost of runtime, memory, and CPU load, underscoring the trade-off faced in edge deployments. The proposed hybrid feature strategy provides beat-by-beat classification with a reduction in the number of features, enabling real-time ECG screening on wearable and IoT devices. Full article

With the growing complexity and frequency of network threats, anomaly detection in network traffic has become a vital task for ensuring cybersecurity. Traditional detection approaches typically rely on statistical features while overlooking the interaction patterns and structural dependencies among traffic flows. In addition, network traffic data are distributed across heterogeneous devices and domains, where centralized training methods face significant challenges such as data leakage and data silos. To address these issues, we propose a network traffic anomaly detection method based on cross-domain federated graph representation learning. In this method, network traffic is modeled as a graph, and a feature-structure decoupling design is adopted to separate the encoding and learning of graph topology and node attributes. Only structural information with minimal sensitive content is transmitted to the central server, whereas sensitive node attributes are preserved and processed locally to enhance privacy protection. Furthermore, a cross-gated feature fusion mechanism is introduced to enhance the expressive interaction between features and to generate graph-level embeddings for anomaly classification. To further improve the model’s generalization across domains, a cross-domain structural guidance mechanism is implemented on the server side, which integrates structural information from multiple domains to guide the training of local models. Comparative experiments with other methods demonstrate that the proposed approach achieves superior performance in distributed network traffic anomaly detection scenarios. Full article

As the frequency of cyberattacks rises, extracting actionable cyber threat intelligence (CTI) from diverse online sources has become critical for proactive threat detection and defense. Named entity recognition (NER) serves as a foundational task in CTI extraction, supporting downstream applications such as cybersecurity knowledge graph construction and attack attribution. However, existing NER methods face significant challenges in the cybersecurity domain, including the need to identify highly specialized entity types and adapt to rapidly evolving threats. These challenges are further exacerbated in few-shot scenarios with limited annotated data. In this work, we focus on few-shot NER for CTI extraction in general cyber environments. Our goal is to develop robust and adaptable methods that are not restricted to specific infrastructures (e.g., traditional IT systems), but instead can generalize across diverse cybersecurity contexts. Specifically, to address these issues, we propose CyberDualNER, a novel dual-stage framework for few-shot NER, which includes span detection and entity classification. In the first stage, we proposed a span detector that can utilize data from large-scale general domains to detect possible entity spans. Based on the detected spans, in the second stage, we propose a prompt-enhanced metric-based classifier. We use category descriptions to build prompt templates, extract category anchor representations, and classify entities based on similarity to span representations. By incorporating prior knowledge, we improve performance while reducing data dependency, which ensures generalizability in the face of emerging entities. Extensive experiments on real-world CTI datasets demonstrate the effectiveness of CyberDualNER, with significant performance improvements over baseline methods. Notably, the framework achieves robust results in scenarios with minimal annotated samples, highlighting its potential for practical applications in cybersecurity intelligence extraction. Full article

This paper explores the capability of large language models (LLMs) to accurately classify objects in challenging visual scenarios, focusing on two main tasks: differentiating real objects from artificial replicas and distinguishing human figures from human-like entities (e.g., mannequins, banners). We evaluate a diverse set of vision–language models (VLMs) ranging from large-scale architectures to parameter-efficient systems across multiple question prompts designed to probe object identification, authenticity verification, and multi-object reasoning. Our experiments reveal that while many models perform reasonably well in identifying single objects, their accuracy declines substantially under more complex conditions, such as multi-object scenes or tasks requiring fine-grained judgments of authenticity. Even top-tier models exhibit noticeable performance drops from around 100% to below 15% accuracy when forced to discern real from fake items among multiple candidates, and from 100% to 83.33% accuracy on providing positional details for human-like figures. We further discuss how these performance limitations indicate gaps in current LLM-based vision systems to highlight the need for more robust spatial reasoning and attribute analysis. Our findings underscore the significance of broadening these models’ multimodal understanding and refining prompts, with an eye toward improving real-world applications—from automated quality control to surveillance—where nuanced visual classification is crucial. By comparing a variety of architectures under consistent evaluation settings, this study offers insights into the barriers LLMs face when confronted with increasingly complex visual information. Full article

Cardiovascular diseases (CVDs) are among the leading causes of mortality worldwide, particularly in Mexico, where rural regions face challenges due to limited access to medical equipment. This preliminary study proposes a low-cost cardiovascular disease classifier, Buazduino-001, which integrates machine learning (ML) techniques with Arduino-based technology to provide accessible and non-invasive risk assessment. Three classical ML models—logistic regression, random forest, and support vector machine—were implemented and evaluated using a dataset of 303 patients from the UCI Machine Learning Repository. This study introduces a six-stage methodology, including a novel step that prioritizes non-invasive attributes to optimize diagnostic time and cost. The random forest model demonstrated the best performance, achieving 87% classification accuracy, with a reduced feature set of five attributes (sex, age, chest pain, heart rate, and exercise-induced angina). In this preliminary study, the system was validated experimentally with 30 patients, confirming an 85% accuracy and an 80% reduction in diagnostic time compared to traditional medical assessments. The results highlight the practicality of combining ML with low-cost electronics to address healthcare gaps in resource-limited settings. While this study is preliminary, the Buazduino-001 system demonstrates potential for early CVD risk detection and could serve as a screening tool in rural clinics, complementing conventional diagnostic methods. Full article

As 5G technology and 3D capture techniques have been rapidly developing, there has been a remarkable increase in the demand for effectively compressing dynamic 3D point cloud data. Video-based point cloud compression (V-PCC), which is an innovative method for 3D point cloud compression, makes use of High-Efficiency Video Coding (HEVC) to carry out the compression of 3D point clouds. This is accomplished through the projection of the point clouds onto two-dimensional video frames. However, V-PCC faces significant coding complexity, particularly for dynamic 3D point clouds, which can be up to four times more complex to process than a conventional video. To address this challenge, we propose an adaptive coding unit (CU) partitioning method that integrates occupancy graphs, convolutional neural networks (CNNs), and Bayesian optimization. In this approach, the coding units (CUs) are first divided into dense regions, sparse regions, and complex composite regions by calculating the occupancy rate R of the CUs, and then an initial classification decision is made using a convolutional neural network (CNN) framework. For regions where the CNN outputs low-confidence classifications, Bayesian optimization is employed to refine the partitioning and enhance accuracy. The findings from the experiments show that the suggested method can efficiently decrease the coding complexity of V-PCC, all the while maintaining a high level of coding quality. Specifically, the average coding time of the geometric graph is reduced by 57.37%, the attribute graph by 54.43%, and the overall coding time by 54.75%. Although the BD rate slightly increases compared with that of the baseline V-PCC method, the impact on video quality is negligible. Additionally, the proposed algorithm outperforms existing methods in terms of geometric compression efficiency and computational time savings. This study’s innovation lies in combining deep learning with Bayesian optimization to deliver an efficient CU partitioning strategy for V-PCC, improving coding speed and reducing computational resource consumption, thereby advancing the practical application of V-PCC. Full article

The global challenges that cities must face regarding sustainability, efficiency, integration, and resilience have found in the smart city concept a guideline of action as a model for urban development and transformation. The multidimensional nature of the smart city, along with the importance of identifying key urban stakeholders and ensuring their engagement, are two widely recognized characteristics within the scientific community. However, proposals for the identification, classification, and management of urban stakeholders are very scarce and almost non-existent when considered in conjunction with the holistic nature of smart cities. Thus, the significant importance attributed to stakeholder engagement contrasts with the lack of clear guidelines to develop it properly. Based on an iterative analysis of the scientific literature combined with the cross-referencing of smart city dimensions, statistical analysis tools, and multi-criteria analysis methods, this paper proposes a new methodology for the identification and management of urban stakeholders. The proposal includes a comprehensive classification and a new framework for developing urban stakeholder identification processes at their early stages or the monitoring and assessment of ongoing or completed processes, including tools for analyzing the extent and homogeneity achieved. The practical application of the methodology to a specific case study is also discussed. Full article

In the face of rapid progress in heritage preservation and cultural tourism integration, landscape planning in historic cities is pivotal to showcasing regional identities and disseminating cultural value. However, the complexity of cultural characteristic identification and the imbalance in planning often restrict the progress of urban development. Additionally, existing studies predominantly rely on subjective methods and focus on a single cultural attribute, highlighting the urgent need for research on diversified cultural perception. Using Jingdezhen, a renowned historic cultural city, as an example, this study introduces a multi-label deep learning approach to examine cultural perceptions in tourism heritage landscapes. Leveraging social media big data and an optimized ResNet-50 model, a framework encompassing artifacts, production, folk, and living culture was constructed and integrated with ArcGIS spatial analysis and diversity indices. The results show: (1) The multi-label classification model achieves 92.35% accuracy, validating its potential; (2) Heritage landscapes exhibit a “material-dominated, intangible-weak” structure, with artifacts culture as the main component; (3) Cultural perception intensity is unevenly distributed, with core areas demonstrating higher recognition and diversity; (4) Diversity indices suggest that comprehensive venues display stronger cultural balance, whereas specialized ones reveal marked cultural singularity, indicating a need for improved integration across sites. This research expands the use of multi-label deep learning in tourism heritage studies and offers practical guidance for global heritage sites tackling mass tourism. Full article

Existing online review research has not fully captured consumer purchasing behavior in complex decision-making environments, particularly in contexts involving multiple product comparisons and conflicting review perspectives. This study thoroughly investigates the impact on focal product purchase decisions when consumers compare multiple products and face information inconsistency. Based on online review data from JD.com, we propose a semi-supervised deep learning model to analyze consumers’ sentiment polarity toward product attributes. The method establishes implicit relationships between labeled and unlabeled data through consistency regularization. Subsequently, we conceptualize three types of online review dissimilarity factors, rating-sentiment dissimilarity, cross-review dissimilarity, and brand dissimilarity, and employ regression models to examine the impact of competing products’ online reviews on focal product sales. The results indicate that by employing a semi-supervised deep learning approach, unlabeled data are annotated with pseudo-labels and utilized for model training, achieving more accurate sentiment classification than using labeled data alone. Moreover, positive (negative) sentiment attributes of competing products have a significant negative (positive) effect on focal product purchases. Online review dissimilarity moderates the spillover effects of competing products. Notably, these spillover effects are more pronounced when competing products are from the same brand compared to different brands. The research findings not only highlight the heterogeneous effects of positive and negative sentiments but also provide a new perspective for examining dissimilarity, enriching the understanding of online review spillover effects and the role of dissimilarity, while offering practical guidance for resource allocation decisions by companies and platforms. Full article

Education is crucial for leading a productive life and obtaining necessary resources. Higher education institutions are progressively incorporating artificial intelligence into conventional teaching methods as a result of innovations in technology. As a high academic record raises a university’s ranking and increases student career chances, predicting learning success has been a central focus in education. Both performance analysis and providing high-quality instruction are challenges faced by modern schools. Maintaining high academic standards, juggling life and academics, and adjusting to technology are problems that students must overcome. In this study, we present a comprehensive dataset, SAPEx-D (Student Academic Performance Exploration), designed to predict student performance, encompassing a wide array of personal, familial, academic, and behavioral factors. Our data collection effort at Air University, Islamabad, Pakistan, involved both online and paper questionnaires completed by students across multiple departments, ensuring diverse representation. After meticulous preprocessing to remove duplicates and entries with significant missing values, we retained 494 valid responses. The dataset includes detailed attributes such as demographic information, parental education and occupation, study habits, reading frequencies, and transportation modes. To facilitate robust analysis, we encoded ordinal attributes using label encoding and nominal attributes using one-hot encoding, expanding our dataset from 38 to 88 attributes. Feature scaling was performed to standardize the range and distribution of data, using a normalization technique. Our analysis revealed that factors such as degree major, parental education, reading frequency, and scholarship type significantly influence student performance. The machine learning models applied to this dataset, including Gradient Boosting and Random Forest, demonstrated high accuracy and robustness, underscoring the dataset’s potential for insightful academic performance prediction. In terms of model performance, Gradient Boosting achieved an accuracy of 68.7% and an F1-score of 68% for the eight-class classification task. For the three-class classification, Random Forest outperformed other models, reaching an accuracy of 80.8% and an F1-score of 78%. These findings highlight the importance of comprehensive data in understanding and predicting academic outcomes, paving the way for more personalized and effective educational strategies. Full article

Traditional fault diagnosis methods for rolling bearings rely on nemerous labeled samples, which are difficult to obtain in engineering applications. Moreover, when unseen fault categories appear in the test set, these models fail to achieve accurate diagnoses, as the fault categories are not represented in the training data. To address these challenges, a zero-shot fault diagnosis model for rolling bearings is proposed, which realizes knowledge transfer from seen to unseen categories by constructing attribute information, thereby reducing the dependence on labeled samples. First, an attribute method Discrete Label Embedding Method (DLEM) based on word embedding and envelope analysis is designed to generate fault attributes. Fault features are extracted using the Swin Transformer model. Then, the attributes and features are input into the constructed model Distribution Consistency and Multi-modal Cross Alignment Variational Autoencoder (DCMCA-VAE), which is built on Convolutional Residual SE-Attention Variational Autoencoder (CRS-VAE). CRS-VAE replaces fully connected layers with convolutional layers and incorporates residual connections with the Squeeze-and-Excitation Joint Attention Mechanism (SE-JAM) for improved feature extraction. The DCMCA-VAE also incorporates a reconstruction alignment module with the proposed distribution consistency loss $L_{WT}$ and multi-modal cross alignment loss function $L_{MCA}$. The reconstruction alignment module is used to generate high-quality features with distinguishing information between different categories for classification. In the face of multiple noisy datasets, this model can effectively distinguish unseen categories and has stronger robustness than other models. The model can achieve 100% classification accuracy on the SQ dataset, and more than 85% on the CWRU dataset when unseen and seen categories appear simultaneously with noise interference. Full article

Hepatitis B remains a major public health issue in Vietnam. Mandatory reporting to the national electronic communicable disease surveillance system (eCDS) has been required since July 2016. We conducted an evaluation of the hepatitis B surveillance system in Ninh Binh, the province with the highest reported burden of hepatitis B in Northern Vietnam, between 2017 and 2022. Using the CDC’s guidelines for evaluating public health surveillance systems, we assessed four key attributes: simplicity, timeliness, data quality, and acceptability. This retrospective evaluation included document reviews, analysis of hepatitis B data, and in-depth interviews with provincial-level healthcare staff involved in the reporting of hepatitis B cases. The results showed that the eCDS improved reporting frequency, provided more detailed case information, and enhanced data accessibility compared to the previous paper-based system. However, the system faced several challenges, including unclear objectives, difficulties in distinguishing acute from chronic cases, insufficient training for staff, lack of supervision for data quality, and technical software issues. Despite these challenges, stakeholders found the system acceptable but emphasized the need for improvements, including revising the system’s objectives, automating case classification, enhancing training, securing funding for maintenance, and implementing regular data review processes. Full article

CO₂-soluble surfactant foam systems have gained significant attention for their potential to enhance oil recovery, particularly in tight oil reservoirs where conventional water-soluble surfactants face challenges such as poor injectability and high reservoir sensitivity. This review provides a comprehensive explanation of the basic theory of CO₂-soluble surfactant foam, its mechanism in enhanced oil recovery (EOR), and the classification and application of various CO₂-soluble surfactants. The application of these surfactants in tight oil reservoirs, where low permeability and high water sensitivity limit traditional methods, is highlighted as a promising solution to improve CO₂ mobility control and increase oil recovery. The mechanism of enhanced oil recovery by CO₂-soluble surfactant foam involves the effective reduction of CO₂ fluidity, the decrease in oil–gas flow ratio, and the stabilization of the displacement front. Foam plays a vital role in mitigating the issues of channeling and gravity separation often caused by simple CO₂ injection. The reduction in gas fluidity can be attributed to the increase in apparent viscosity and trapped gas fraction. Future research should prioritize the development of more efficient and environmentally friendly CO₂-soluble surfactants. It is essential to further explore the advantages and challenges associated with their practical applications in order to maximize their potential impact. Full article