Earticle

This paper presents an improved approach to generate pseudo labels for unlabeled dataset. To properly train a network, large amount of dataset is required. The publicly available datasets are often not large enough or versatile. Although we can acquire a great deal of images from the internet, those images are not labeled. Conventionally, the generation of ground truth labels requires human effort which is very expensive and time-consuming. Recently, existing object detectors are being employed to automate the generation of labels, called pseudo labels. Such pseudo labels have poor accuracy, since most of the object detectors employ simplistic confidence thresholding, which tends to discard even good labels. This paper proposes an enhanced pseudo labeling technique that selects the predicted labels using a bi-directional tracking method instead of simplistic confidence thresholding. The proposed technique can recover many predicted labels that are actual good labels but would have been discarded due to their poor confidence. Our method can produce pseudo labels for new training dataset with higher accuracy than conventional pseudo labeling techniques, thus offering better training accuracy for object detector CNN models.

While recent convolutional neural networks (CNNs) for object detection have been substantially improved, they require a large amount of annotated data to further improve their accuracy to the level of human. Such annotated data is scarce. The generation of ground truth to annotate training data is a time consuming and resource expensive process. Researchers use traditional data augmentation techniques to increase the amount of training data. Recently, generative models are being employed to augment data which produces diverse training data. This leads to an increase in model performance. This paper presents a method to train a GAN network and generate augmented data of any domain of interest with the least compromise in the quality of generated images. The proposed method trains a GAN with vehicles images of different colors. Then it can change the color of vehicles in any given vehicle dataset to a set of specified colors.

For the anomaly detection task, previously presented deep learning approaches suffer from one potential issue in the testing stage, the resultant output image has noise and missing anomaly area. To deal with this issue, we present a novel two-stage convolutional neural network (CNN) for anomaly detection. In the training stage, the first model is trained by inserting pseudo-anomalies, while the second model is trained by a superpixel technique which segments the image refined by the first model. The superpixel technique can recover partially visible anomaly patterns and suppress noise outside the recovered anomaly patches. We trained the proposed model using an industrial dataset MVTec and compared its performance with state-of-the-art pseudo-anomalous method [11]. Our method shows comparable pixel based percentage area under the receiver operating characteristic (%AUROC) of 96.0% which is only 1.3% less than the performance of DRAEM. However, our model uses four times less number of parameters.

In recent years, the deployment of deep neural networks in real-time clinical settings has been considered vulnerable due to domain shifts, which lowers their performance. Polyp image has significant appearance shifts, which eventually impact the performance. So, a deep learning model that generalizes unseen images is in high demand. This paper introduced a practical approach to improving domain shift issues. Firstly, we unified the style transfer with the segmentation model into one framework to diminish the appearance shifts problems and do segmentation alongside. Secondly, with the help of Adaptive Instance Normalization, we transferred the style precisely and dynamically in the earlier layers of the segmentation model. Our solution shows better results on the 224*224 image input than other baseline models.

Data augmentation has been employed in neural networks for building robust models, not exclusively in the training phase but also in the testing stage, where the predictions of every transformed image are aggregated to a greater lustiness and upgraded accuracy. Furthermore, deep learning approaches applied in data augmentation, namely adversarial training, GANs, and Neural Style Transfer were applied while training the models, neither while testing them. In this work, we present a study of applying test-time Neural Style Transfer transformation in medical images as a method of augmentation in test time. Besides, we display the experiment's results of a classification task. Results reveal that the synthesized samples employed as modified images in the test time significantly improved the performance of the classification model.

Recently, due to the recent significant advances in machine learning and deep learning, it is being utilized in many fields. However, real-world data in the medical field significantly degrades the performance of machine learning algorithms due to problems that are heavily skewed to specific states or that the distribution of data is unbalanced. Therefore, this study solves the problem of not being learned by converting the dependent variable into a regression problem that predicts using a new dependent variable by pseudo labeling. Also, this study present ensemble methods to improve the performance of the model and prevent overfitting.

Neural machine translation has achieved great success in the last decade. It has been widely adopted in multiple commercial products and domains, facilitating communication and trade between countries. In this paper, we present our Chinese to Vietnamese neural machine translation system, which integrates recent advances in deep learning and is enhanced with an external named entity translation model.

Emergency exit signs are crucial during misfortunate events such as fire, earthquakes or even human caused events such as robbery and, bombing. However, these signs are of no use to the visually impaired people. During emergency scenarios, the blind people need to rely on other individuals and sometimes they may even be left helpless. This raises a need for some assistive device that could benefit the visually impaired people during the time of emergency. In this paper, we propose a concept of smart glasses that could be tremendously beneficial to the blind people. These glasses will have camera and headphone speakers embedded to them. The device will be capable of detecting emergency signs using modern deep learning techniques during the times of need and could notify the user regarding the direction where the exit is.

The TCP/IP-based wired communication network has very low transmission efficiency 0.32, which is actual data transfer speed divided by the maximum transferable speed. In order to improve the transmission efficiency, it is necessary to create a new protocol that replaces TCP/IP rather than to conduct a study for each TCP/IP layer. In this study, a coordinate-based protocol and router that broke away from TCP/IP is proposed. As a result of this routeerr experiment, the transmission efficiency was about 0.79 where the actual data transfer speed is about 2.2Gbps and maximum transferable speed is about 2.8Gbps.

This research work demonstrates surveillance of traffic on roads and streets which is used by private companies and public organizations and government institutions. The primary purpose is the well-organized management of the transport system and public safety on highways and in civil areas. This paper used the technique to well-structured localize the LP and segmentation of captured images is done by the ALPR system. We explained the localization of license plates by using the integrated segmentation method. ALPR system contains several well-observed skeletons like security administration, parking, vehicle identification, streets and road activity management, schedule of toll collecting framework, and so forth. There are various frameworks are present which are used for License plate capturing. The most important part of the ALPR framework is the accurate confinement of different number plates, recognition, and segmentation. By ALPR systems we can easily identify the number of vehicle plates. ANPR system also plays a crucial part in vehicle plate capturing and identification. This system helps in monitoring and tracking automobiles. In this paper, we have tried numerous techniques for traffic control and monitoring purposes which are works based on various techniques and methodologies. But ANPR primarily did their work for accuracy and template matching of vehicle number plates.

The growth rate of construction is increasing every year. The sustainable development of territories is becoming a problematic field for solving the strategic tasks of urban development; however, at the same time, the requirements for the standard of living, the safety of citizens, and the adaptability of tasks to the fast pace of urban life and a favorable environmental situation are also increasing. In addition, digitalization is rapidly penetrating all spheres of human life and can become the key to solving problems related to urban infrastructure development. The development and digitalization of regions require a special approach. During the study, artificial intelligence technologies are considered the central link in managing the digital processes of "smart cities" an analysis is made of experiences in implementing the concept of a "smart city" in life and the chains of interconnection between artificial intelligence technologies and relevant digital achievements of smart cities. Such an approach will form the main vector of the city design movement, where each person will feel safe and comfortable. The environmental situation against the background of intelligent systems will be protected from the negative anthropological factor.

The Common Vulnerability Scoring System (CVSS) is a well-known framework for providing the characteristics of a vulnerability and a numerical score reflecting its severity. Since the existing CVSS has been developed for an information technology (IT) environment, it is not suitable for an intelligent building (IB) environment where OT Operational Technology (OT) and Internet of Things (IoT) are integrated. For example, compared to IT systems that prioritize confidentiality, OT systems prioritize safety and have a very long lifespan. Therefore, if a vulnerable component is exploited in an intelligent building environment, the impact may be different from that in the IT environment. This paper identifies problems that may occur when applying the traditional CVSS to an intelligent building environment. We also propose an effective method to extend the CVSS for an intelligent building environment using threat intelligence and rubric.

Early warning is essential for reducing disaster damage. There is a need for an automated catastrophe classification model that can respond fast to local disaster damage by using the properties of social media, where information is exchanged swiftly. Recently, research on automatically identifying disasters using deep learning has begun, with the goal of supplementing the deep learning model's performance. Here, we propose a novel framework, Metamon-Disaster for deep learning model that automatically classifies disasters based on disaster-related keyword data collected from social media. To classify disasters by type, a learning model generated from NAS was employed, and when compared to other classification such as RF (Random Forest), SVC (Support Vector Classifier) or GBM (Gradient Boost Machine) to check the optimal performance of the model, the suggested model exhibited the best performance with an 0.8928 F1-Score. The model for disaster notification service will provide automated disaster notice and rapid reaction.

An artificial intelligence (AI) model can be used to determine a specific posture in a training scenario using virtual reality (VR) instead of training experts. One perspective that has been somewhat neglected is the consideration of left-handed people in modeling. In this paper, we propose a method to model both right- and left-handed users based on the data collected from right-handed people. With the models using the proposed features, we demonstrated the validity of our feature engineering through a user study with six participants. The left-handed participants displayed high satisfaction, showing a potential for applying the model to various VR scenarios.

The future tactical networks (FTNs) are expanding and evolving from the existing terrestrial space-oriented operational structure to multi-space architectures, including aerial and satellite spaces. This may lead to the complexity of interoperability to provide seamless connectivity among complex and numerous elements such as heterogeneous networks and devices constituting FTNs. Software Defined Networking (SDN) has been recognized as one of the most suitable solutions to solve interoperability problems, especially in terms of networking. In this paper, we consider the issues and challenges to be considered in applying SDN to FNTs.

the outbreak of COVID-19 helped to accelerate a digital transformation of government in countries throughout the world. Notably, digital technology has greatly contributed to maintaining effective quarantine and social distancing measures during the pandemic highlighting the vital importance of digital government. The purpose of this paper is to comparatively analyse the different digital government systems and key policies of two countries, Singapore and South Korea, which consistently rank in a top group among Asian countries in digital government development. This study may be helpful for other governments that strive to foster closer and more immediately responsive relationships to the public through the use of digital technologies such as AI.!!!

Internet of Things (IoT), Artificial Intelligence (AI), and Mixed Reality (MR) are recognized as ones of the most promising core technologies to lead the 4th industrial revolution. In this paper, we introduce a use case of MR-IoT/AI convergence Platform, which provide a general platform architecture to be utilized various filed of areas, to be applied for military surveillance system.

To increase the convenience in daily life activities, the Internet of Things (IoT) technology has been developed and used; it can achieve a hyper-connected society by connecting various devices and systems through the Internet. In building an IoT environment, the moving target defense (MTD) strategy is used as a method to construct an active defense strategy for a mission-critical system. However, there is a lack of indicators that can easily identify the various properties of the MTD strategy and enable in constructing and establishing a new MTD strategy-based protection plan. To solve this problem, in this paper, we survey various MTD strategy research results and analyze the research that has been conducted to derive the various properties of the three perspectives in the MTD strategy (When to move, What to move, How to move). In addition, based on the various derived properties, we propose a graph that can easily identify the various research results that appear. Additionally, it shows that the combination of various properties can be used as an indicator to understand the research direction of a new protection strategy research based on the software-defined MTD strategy for IoT device protection.

Approximately 40% of underground water and sewage pipes in Korea are more than 20 years old. Consequently, potential accidents related to water drainage systems are to be expected. In this study, using a special machine learning method that employs various available data, we developed a system that receives and analyzes data in smart pipes called the "digital twin-based smart pipe integrated management system" (DTMS-IM). This system presents an integrated approach for the efficient operation and monitoring of water pipes, allowing the innovative operation of groundwater pipes through smart decision-making. We trained the model using these data. This well-trained model has become able to predict the aging level of pipes. Similar artificial intelligence prediction models, widely used in various industrial applications, are also discussed.

The purpose of this study is to help the visually impaired lead a quality life by studying the effect of XR glasses on technology acceptability and sense of presence for the visually impaired. This study used the Technology Acceptance Model as a theoretical framework to verify the correlation of four variables: perceived immersion, information quality, XR presence, and digital informatization level for visually impaired people. As a result of the study, there were significant differences in technology acceptability in perceived ease of use, perceived usefulness, attitude toward use, and intention to use after the experiment on the visually impaired. There was no statistically significant difference in the variables related to the sense of presence, but a positive correlation could be confirmed. Based on the results of this study, it is expected to contribute to the technological advancement of XR glasses for the visually impaired.

From Covid-19 we have witnessed the destructive power of infectious diseases. To prevent such catastrophes from occurring, it is crucial to prevent an outbreak of any infectious disease. As it is well known for bacteria and viruses to cause such outbreaks, some fungal species also cause harmful reactions. In this paper, we attempt to classify toxic fungi protein sequences through the help of protBERT a BERT-based protein language model. Our experiment results reveal the effectiveness of our proposed approach as it shows 99% accuracy and F1 score of 0.9901 in the classification of toxic fungi protein sequences.

Android malware analysis systems examine the runtime behavior of apps in emulators for dynamic analysis. However, evasive Android malware can stop executing malicious activities to avoid detection by malware analysis systems if it finds itself executing on an emulator. Thus, it is important to investigate existing and potential techniques for detecting emulated environments. In this paper, we propose an effective technique to detect the latest Android emulators, Android Virtual Device (AVD), NoxPlayer, and BlueStacks. The proposed technique utilizes the properties of the build.prop file as well as android.os.Build class to detect the latest emulators. Experimental results show that emulators were effectively detected by checking the string values corresponding some specific properties.

The usage of Unmanned Aerial Vehicles (UAVs) has escalated widely in recent years. The mobility they provide and their capability to be used autonomously has made them applicable to a wide range of fields for both the civilians and the military. However, the flight safety is pivotal during UAV flights for any application and it is necessary that the recently ordained rules are complied with. Ensuring the compliance of such rules and regulations is especially difficult when using these UAVs are used in Autonomous mode. To establish the flight safety and compliance with the regulations, computer vision can be utilized while in autonomous mode. This is predominantly true when the flight is conducted in the proximity of crowds, animals, moving vehicles or when emergency and precision landings need to be performed. To that extent, we present a concept of drone flight control system which utilizes object detection and tracking algorithms.