Mobile Networks and Applications最新文献

The capability to accurately detect web application attacks, especially in a timely fashion, is crucial but remains an ongoing challenge. This study provides an in-depth evaluation of 19 traditional machine learning techniques for detecting web application attacks. The evaluation was conducted across three distinct experiments on refined datasets derived from the HTTPCSIC 2010 dataset. The experiments investigated the performance of these algorithms in different scenarios (e.g., without Knowledge Graph integration, and with KG integration with node2vec feature enhancement). The experimental results revealed that neural network classifiers, notably the Multilayer Perceptron, consistently outperformed other models, achieving accuracy of above 0.90 and maintaining a balanced performance across various metrics. Furthermore, the findings demonstrated that certain algorithms, such as tree-based ensemble methods showed an increase of over 10% in accuracy and Gaussian Process models which exhibited a remarkable improvement in accuracy, rising from 0.84 to 0.99, and in AUC from 0.91 to 1.00, when integrated with the Knowledge Graph, effectively utilizing the additional contextual information. We also found that the KNN classifier demonstrated more than a 16% increase in accuracy. All classifiers showed significant improvements in AUC and other metrics mentioned in our study, indicating that KG integration not only enhances the detection capabilities but also enriches the overall analytical performance of the models. We also observed that linear classifiers and Naive Bayes models generally experienced a decline in performance, highlighting the importance of carefully evaluating the inherent characteristics and capabilities of each algorithm for the web attack detection task.

In basketball videos, the trajectories of a basketball changes rapidly. Since the visual features changes in a more homogeneous region, the frame difference method is a suitable basis for trajectory real-time tracking. However, traditional methods need a huge number of iterative calculations in a random image to find spatial feature differences to segment the basketball from to frame, resulting in tracking lag. Therefore, a real-time tracking method of basketball trajectory is designed based on an associative Markov Chain Monte Carlo (MCMC) model. From pixel illumination differences between two adjacent frames in basketball game videos, the basketball’s movement is determined, and the foreground and background of the basketball frame are separated. Then, coordinates of the basketball are detected by a Convolutional Neural Network (CNN), and the change of coordinates is used to construct a visual 2D mapping model, which calculates both angular and linear acceleration of the basketball. To solve the interaction problem of randomness and spatial variability, an associative MCMC model is designed to segment basketball images with simple conditions, and a Bayesian network is established to input parameters of the segmented basketball movement for the determination of trajectory deviation. Finally, basketball movement trends are calculated to achieve real-time tracking of the trajectory in the basketball video. The experimental results show that compared with the original running path, this method has the smallest difference in tracking trajectory error, and the estimation error does not exceed 0.2 when the false alarm rate is 100. The trajectory tracking time is always less than 2.2 seconds, indicating that it has good trajectory tracking ability.

In Vietnam, every year, several competitions for high school students in different disciplines are organized at multilevel, provincial and national, with highly valuable prizes for winners. The problem to be solved by educational managers is how to efficiently select students for their school teams. This study proposed the approach of winner-domain to solving the selection problem by applying two following strategies: (1) students with performance similar to the winners of previous prizes are more likely to win next prizes; (2) the selection of team members is based on the “False Leaving out Better Than False Selecting” rule. This study conceptually defined the winner-domain which is the domain of the performance of winners to select team members. The algorithms forming winner-domain and the process selecting team member were installed. The approach was experimentally applied and evaluated at a high school in Southern Vietnam. The initial achievement showed that the proposed approach outperformed the previous methods which choose team members based on learning or testing outcomes.

In this paper, we introduce a graphical method for modeling and reasoning with linguistic expressions. The former represents a graph called a conceptual graph, and the latter involves graph transformations. In our conceptual graphs, nodes represent linguistic concepts and edges links between these concepts. This model facilitates reasonining with linguistic concepts by making direct consequences easy to infer.

The development of 5G private networks harbors immense potential for technological innovation, significantly enhancing digital connectivity and performance. This advancement has unlocked opportunities for pioneering educational approaches within the metaverse in campus settings. Utilizing enhanced bandwidth, reduced latency, and improved security, 5G technology is pivotal in facilitating immersive and interactive educational experiences. This paper explores the practical application of 5G private networks at a university, demonstrating their role in revolutionizing traditional teaching methodologies. Leveraging the author’s experience in deploying a 5G network at their institution, the study underscores the transformative potential of this technology in creating engaging and dynamic learning environments in the metaverse. The findings provide valuable insights into the effective integration of 5G networks in education, highlighting their significance in evolving academic paradigms.

In recent years, virtualization platforms have not only been used to integrate data from traditional application systems but have also actively collected Internet of Things (IoT) data from various network transmissions. To address the challenges of real-time monitoring for key metrics on virtualization platforms, this study proposes an optimal machine learning training model that combines semi-supervised Self-Training algorithms with supervised ensemble algorithms. In the application of semi-supervised training learning algorithms, this study utilizes a Self-Training learning algorithm to label a large number of unlabeled virtual machine operational states with a small amount of labeled data, laying the foundation for subsequent model construction. Subsequently, an ensemble learning classification algorithm is introduced to further validate and identify learning models suitable for generalization. Empirical evaluations show that the RandomForest algorithm serves as the optimal base estimator for Self-Training, while the Bagging algorithm is the optimal choice for ensemble learning. The synergy of these two achieves an accuracy exceeding 99%, enabling the model to accurately differentiate between various operational states such as normal operation, resource insufficiency, and faults. Finally, the integrated training model is deployed to a dashboard, displaying the real-time operational status of virtual machines through different colored lights. Simultaneously, operational status information is communicated to stakeholders through various media, further improving coordination, decision-making, and resource allocation issues on the virtualization platform. This study provides an efficient and feasible solution for monitoring and managing virtualization platforms.

To solve the problems of poor estimation of full-body shape and inaccurate capture results in human motion capture in mobile sensor networks, a method of capturing image features of human full-body motion posture in mobile sensor networks is studied. The method uses Markov random fields to cooperate with sensors to extract human full-body motion foreground images and combines guided filtering to enhance the extraction effect of foreground images. Based on the foreground images, a human tree-structured model is established to simulate the actions of human movements. The extracted foreground images are used as input to the convolutional neural network to extract edge features and spatio-temporal features of human motion posture. After fusion, a human motion posture feature matrix is constructed. Based on the least squares method, a strong regression mapping model is constructed. According to the structure of the human tree model, multi-dimensional iterative mapping is performed from top to bottom between the human motion posture feature matrix and the human tree model. The joint positions corresponding to the human motion posture feature matrix in the human tree model are calculated, and the two-dimensional position information of all joint points of the moving human body is obtained. The capture of human full-body motion posture in mobile networks is completed. Experimental data show that the method has clear foreground image extraction, can effectively obtain human motion features, and has accurate capture results of human full-body motion posture.

Customized healthcare for cattle health monitoring is essential, which aims to optimize individual animal health, thereby enhancing productivity, minimizing illness-related risks, and improving overall welfare. Tailoring healthcare practices to individual requirements guarantees that individual animals receive proper attention and intervention, resulting in better health outcomes and sustainable cattle farming practices. In this regard, the manuscript proposes a visual cues-based region prediction methodology to design a customized cattle healthcare system. The proposed automated AI healthcare system uses resource-efficient deep learning-inspired architecture for computer vision applications like performing region-wise classification. The classification mechanism can be used further to identify a cattle and the regions it belongs. Extensive experimentation has been conducted on a redesigned image dataset to identify the best-suited deep-learning framework to perform region classification for livestock, such as cattle. MobileNetV2 outperforms the considered state-of-the-art frameworks by achieving an accuracy of 93% in identifying the regions of the cattle.

With the rapid development of virtual reality (VR) technology, the design and presentation of museum exhibitions are also undergoing changes. However, traditional VR technology has some limitations in terms of transmission speed and user experience, so new technologies need to be introduced to improve these issues. This study focuses on the perspective of optical communication to improve the problems existing in traditional VR technology. A wireless sensor network-based optical communication system is proposed, which utilizes fiber optic transmission to provide stable and high-speed data transmission. Using optical fiber as a transmission medium can effectively transmit a large amount of data with lower transmission delay and higher bandwidth, overcoming the problems of delay and lag caused by low data transmission speed in traditional VR technology. By utilizing the characteristics of optical communication, sensor nodes and VR devices are wirelessly connected. The sensor nodes are arranged in different areas of the museum and connected to the central server through fiber optics. VR devices establish wireless connections with sensor nodes, transmit data through optical signals, achieve high-speed data transmission, and provide more freedom of mobility and a more realistic interactive experience.