Array最新文献

The COVID-19 pandemic has been infecting the entire world over the past years. To prevent the spread of COVID-19, people have acclimatised to the new normal, which includes working from home, communicating online, and maintaining personal cleanliness. There are numerous tools required to prepare to compact transmissions in the future. One of these elements for protecting individuals from fatal virus transmission is the mask. Studies have indicated that wearing a mask may help to reduce the risk of viral transmission of all kinds. It causes many public places to take efforts to ensure that its guests wear adequate face masks and keep a safe distance from one another. Screening systems need to be installed at the doors of businesses, schools, government buildings, private offices, and/or other important areas. A variety of face detection models have been designed using various algorithms and techniques. Most of the articles in the previously published research have not worked on dimensionality reduction in conjunction with depth-wise separable neural networks. The necessity of determining the identities of people who do not cover their faces when they are in public is the driving factor for the development of this methodology. This research work proposes a deep learning technique to determine if a person is wearing mask or not and identifies whether it is properly worn or not. Stacked Auto Encoder (SAE) technique is implemented by stacking the following components: Principal Component Analysis (PCA) and Depth-wise Separable Convolutional Neural Network (DWSC-NN). PCA is used to reduce the irrelevant features in the images and resulted high true positive rate in the detection of mask. We achieved an accuracy score of 94.16% and an F1 score of 96.009% by the application of the method described in this research.

The MQTT (Message Queue Telemetry Transport) protocol has recently been standardized to provide a lightweight open messaging service over low-bandwidth and resource-constrained communication environments. Hence, it is the primary messaging protocol used by Internet of Things (IoT) devices to disseminate telemetry data in a machine-to-machine approach. Despite its advantages in providing reliable, scalable, and timely delivery, the MQTT protocol is widely vulnerable to flooding and denial of service attacks, specifically, the low-rate distributed denial of services (LR-DDoS). Unlike conventional DDoS, the LR-DDoS attack tends to appear as normal traffic at a very slow rate, which makes it difficult to differentiate from legitimate packets, allowing the packets to move undetected by traditional detection policies. This paper presents an intelligent lightweight detection scheme that can capture LR-DDoS attacks based on MQTT protocol in a software-defined IoT environment. The proposed scheme examines the performance of four machine learning models on a modern dataset (LRDDoS-MQTT-2022) with a minimum feature set (i.e., two features only) and a balanced dataset, namely: decision tree classifier (DTC), multilayer perceptron (MLP), artificial neural networks (ANN), and naïve Bayes classifier (NBC). Our exploratory assessment demonstrates the arrogance of the DTC detection scheme achieving an accuracy of 99.5% with peak detection speed. Eventually, our best outcomes outdo existing models with higher prediction rates.

In the commercial and industrial sectors, automatic robotic control mechanisms, which include robots, end effectors, and anchors containing components, are often utilized to enhance service quality. Robotic systems must be installed in manufacturing lines for a variety of industrial purposes, which also increases the risk of a robot, end controller, and/or device malfunction. According to its automated regulation, this may hurt people and other items in the workplace in addition to resulting in a reduction in quality operation. With today's advanced systems and technology, security and stability are crucial. Hence, the system is equipped with fault management abilities for the identification of developing defects and assessment of their influence on the system's activity in the upcoming utilizing fault diagnostic methodologies. To provide adaptive control, fault detection, and state estimation for robotic automated systems intended to function dependably in complicated contexts, efficient techniques are described in this study. This paper proposed a fault detection and state estimation using Accelerated Gradient Descent based support vector machine (AGDSVM) and gaussian filter (GF) in automatic control systems. The Proposed system is called (AGDSVM + GF). The proposed system is evaluated with the following metrics accuracy, fault detection rate, state estimation rate, computation time, error rate, and energy consumption. The result shows that the proposed system is effective in fault detection and state estimation and provides intelligent control automatic control.

Semi-supervised clustering is a machine learning technique that was introduced to boost clustering performance when labeled data is available. Indeed, some labeled data are usually available in real use cases, and can be used to initialize the clustering process to guide it and to make it more efficient. Fuzzy ART is a clustering technique that is proved to be efficient in several real cases, but as an unsupervised algorithm, it cannot use available labeled data. This paper introduces a semi-supervised variant of the FuzzyART clustering algorithm (SSFuzzyART). The proposed solution uses the available labeled data to initialize clusters centers. In another hand, to deeply evaluate the characteristics of the proposed algorithm, a clustered binary data generation algorithm with controlled partitioning is also introduced in this paper. Indeed, the controlled generated clusters allows studying the characteristics of the proposed SSFuzzyART. Furthermore, a set of experiments is carried out on some available benchmarks. SSFuzzyART demonstrated better clustering prediction results than its classic counterpart.

In some sensitive domains where data imperfections are present, standard classification techniques reach their limits. To avoid misclassifications that have serious consequences, recent works propose cautious classification algorithms to handle this problem. Despite of the presence of uncertainty and/or imprecision, a point prediction classifier is forced to bet on a single class. While a cautious classifier proposes the appropriate subset of candidate classes that can be assigned to the sample in the presence of imperfect information. On the other hand, cautiousness should not be at the expense of precision and a trade-off has to be made between these two criteria. Among the existing cautious classifiers, two classifiers propose to manage this trade-off in the decision step by the mean of a parametrized objective function. The first one is the non-deterministic classifier (ndc) proposed within the framework of probability theory and the second one is “evidential classifier based on imprecise relabelling” (eclair) proposed within the framework of belief functions. The theoretical aim of the mentioned hyper-parameters is to control the size of predictions for both classifiers. This paper proposes to study this hyper-parameter in order to select the “best” value in a classification task. First the utility for each candidate subset is studied related to the values of the hyper-parameter and some thresholds are proposed to control the size of the predictions. Then two illustrations are proposed where a method to choose this hyper-parameters based on the calibration data is proposed. The first illustration concerns randomly generated data and the second one concerns the images data of fashion mnist. These illustrations show how to control the size of the predictions and give a comparison between the performances of the two classifiers for a tuning based on our proposition and the one based on grid search method.

Intrusion detection is a critical aspect of network security to protect computer systems from unauthorized access and attacks. The capacity of traditional intrusion detection systems (IDS) to identify unknown sophisticated threats is constrained by their reliance on signature-based detection. Approaches based on machine learning have shown promising results in identifying unknown malicious attacks. No learning algorithm-based model, however, is able to accurately and consistently detect all different kinds of attacks. Besides that, the existing models are tested for a specific dataset. In this research, a novel ensemble-based machine-learning technique for intrusion detection is presented. Numerous public datasets and multiple ensemble strategies, including Random Forest, Gradient Boosting, Adaboost, Gradient XGBoost, Bagging, and Simple Stacking, will be employed to evaluate the performance of the proposed approach. The most relevant features for the detection of intrusion are selected using correlation analysis, mutual information, and principal component analysis. Our research using different ensemble methods demonstrates that the proposed approach using the Random Forest technique outperforms existing approaches in terms of accuracy and FPR, typically exceeding 99% with better evaluation metrics like Precision, Recall, F1-score, Balanced Accuracy, Cohen's Kappa, etc. This strategy may be a useful tool for strengthening the safety of computer systems and networks against emerging cyber threats.

Semi-supervised clustering is a machine learning technique that was introduced to boost clustering performance when labeled data is available. Indeed, some labeled data are usually available in real use cases, and can be used to initialize the clustering process to guide it and to make it more efficient. Fuzzy ART is a clustering technique that is proved to be efficient in several real cases, but as an unsupervised algorithm, it cannot use available labeled data. This paper introduces a semi-supervised variant of the FuzzyART clustering algorithm (SSFuzzyART). The proposed solution uses the available labeled data to initialize clusters centers. In another hand, to deeply evaluate the characteristics of the proposed algorithm, a clustered binary data generation algorithm with controlled partitioning is also introduced in this paper. Indeed, the controlled generated clusters allows studying the characteristics of the proposed SSFuzzyART. Furthermore, a set of experiments is carried out on some available benchmarks. SSFuzzyART demonstrated better clustering prediction results than its classic counterpart.

Offline handwritten text recognition (HTR) typically relies on segmented text-line images for training and transcription. However, acquiring line-level position and transcript information can be challenging and time-consuming, while automatic line segmentation algorithms are prone to errors that impede the recognition phase. To address these issues, we introduce a state-of-the-art solution that integrates vision and language models using efficient split and multi-head attention neural networks, referred to as joint attention (ResneSt-Transformer), for end-to-end recognition of handwritten paragraphs. Our proposed novel one-stage, segmentation-free pipeline employs joint attention mechanisms to process paragraph images in an end-to-end trainable manner. This pipeline comprises three modules, with the output of one serving as the input for the next. Initially, a feature extraction module employing a CNN with a split attention mechanism (ResneSt50) is utilized. Subsequently, we develop an encoder module containing four transformer layers to generate robust representations of the entire paragraph image. Lastly, we designed a decoder module with six transformer layers to construct weighted masks. The encoder and decoder modules incorporate a multi-head self-attention mechanism and positional encoding, enabling the model to concentrate on specific feature maps at the current time step. By leveraging joint attention and a segmentation-free approach, our neural network calculates split attention weights on the visual representation, facilitating implicit line segmentation. This strategy signifies a substantial advancement toward achieving end-to-end transcription of entire paragraphs. Experiments conducted on paragraph-level benchmark datasets, including RIMES, IAM, and READ 2016 test datasets, demonstrate competitive results compared to recent paragraph-level models while maintaining reduced complexity. The code and pre-trained models are available on our GitHub repository here: HTTPS link.

The extensive use of images in many fields increased the demand for image compression algorithms to overcome the transfer bandwidth and storage limitations. With image compression, disk space, and transmission speed can be efficiently reduced. Some of the traditional techniques used for image compression are the JPEG and ZIP formats. The compression rate (CR) in JPEG can be high but to the detriment of the quality factor of the image. ZIP has a low compression rate, where the quality remains almost unaffected. Machine learning (ML) is considered an essential technique for image compression using different algorithms. The most widely used algorithm is Deep Learning (DL), which represents the features of the image at different scales by using different types of layers. In this research, an AutoEncoder (AE) deep learning-based compression algorithm is proposed for lossy image compression and experimented with using three standard dataset types: MNIST, Grayscale, and Color images datasets. A Stacked AE (SAE) for image compression and a binarized content-based image filter are used with a high compression rate while keeping the quality above 85% using structural similarity index metric (SSIM) compared to traditional techniques. In addition, a convolutional neural network (CNN) classification model has been utilized as SAEs compression model selector for each image class. Experimental results demonstrate that the proposed SAE image compression algorithm outperforms the JPEG-encoded algorithm in terms of compression rate (CR) and image quality. The CR that the proposed model achieved with an acceptable reconstruction accuracy was about 85%, which is almost 20% higher than the standard JPEG’s compression rate, with an accuracy of 94.63% SSIM score.

Edge-Based Online Learning (EBOL), a technique that combines the practical, hands-on approach of EBOL with the convenience of Online Learning (OL), is growing in popularity. But accurately monitoring student engagement to enhance teaching methodologies and learning outcomes is one of the difficulties of OL. To determine this challenge, this paper has put forth an Edge-Based Student Attentiveness Analysis System (EBSAAS) method, which uses a Face Detection (FD) algorithm and a Deep Learning (DL) model known as DLIP to extract eye and mouth landmark features. Images of the eye and mouth are used to extract landmarks using DLIP or Deep Learning Image Processing. Landmark Localization pre-trained models for Facial Landmark Localization (FLL) are one well-liked DL model for facial landmark recognition. The Visual Geometry Group-19 (VGG-19) learning model then uses these features to classify the student's level of attentiveness as fatigued or focused. Compared to a server-based model, the proposed model is developed to execute on an Edge Device (ED), enabling a swift and more effective analysis. The EBOL achieves 95.29% accuracy and attains 2.11% higher than existing model 1 and 4.41% higher than existing model 2. The study's findings have shown how successful the proposed method is at assisting teachers in changing their teaching methodologies to engage students better and enhance learning outcomes.