In recent years, Convolutional Neural Networks (CNNs) have been successfully implemented in different tasks of computer vision. Since CNN models are the representatives of supervised learning algorithms, they demand large amount of data in order to train the classifiers. Thus, obtaining data with correct labels is imperative to attain the state-of-the-art performance of the CNN models. However, labelling datasets is quite tedious and expensive process, therefore real-life datasets often exhibit incorrect labels. Although the issue of poorly labelled datasets has been studied before, we have noticed that the methods are very complex and hard to reproduce. Therefore, in this research work, we propose Deep CleanNet - a considerably simple system that achieves competitive results when compared to the existing methods. We use K-means clustering algorithm for selecting data with correct labels and train the new dataset using a deep CNN model. The technique achieves competitive results in both training and validation stages. We conducted experiments using MNIST database of handwritten digits with 50% corrupted labels and achieved up to 10 and 20% increase in training and validation sets accuracy scores, respectively.

Fonts have developed as a visual element, and their influence has rapidly increased around the world. Research on font automation is actively being conducted mainly in English because Hangul is a combination character and the structure is complicated. In the previous study to solve this problem, the stroke element of the character was automatically extracted by applying the object detection by component. However, the previous research was only for similarity, so it was tested on various print style fonts, but it has not been tested on other characters. In order to extract the stroke elements of all characters and fonts, we performed a performance analysis experiment according to the expansion character in the Hangul stroke element extraction training. The results were all high overall. In particular, in the font expansion type, the extraction success rate was high regardless of having done the training or not. In the character expansion type, the extraction success rate of trained characters was slightly higher than that of untrained characters. In conclusion, for the perfect Hangul stroke element extraction model, we will introduce Semi-Supervised Learning to increase the number of data and strengthen it.

This paper proposes a core technology for a micro river monitoring terminal device suitable for flood monitoring in small rivers and valleys. Our proposed device is basically equipped with a 77GHz radar, gyro and accelerometer sensors. To measure the flow velocity and water level, we proposed a signal processing technique that extracts pure water energy components from the observed Doppler velocity and reflection intensity from the radar. And to determine the stability of the river structure equipped with our device, we constantly monitor the displacement of the measured values of the gyro and accelerometer sensors. Experimental result verified that our method detects pure water energy in various river environments and distinguishes between flow velocity and water level well. And we verified that vibration and position change of structures can be determined through a gyro sensor. In future research, we will work to build a secure digital twin river network by lowering the cost of supplying RF-WAV devices. Also we expect our device to contribute to securing a preventive golden time in rivers.

Recently, a 4-col 7-row LED Screen that provides various information of major roads and local governments has been installed and operated. However, due to deterioration due to changes in temperature and humidity, deterioration due to static electricity, and mechanical stress, partial module failure of the display may occur, which is a major cause of missing information of vitally given to citizens. However, there have been frequent cases where the 4-col and 7-row LED Screen that have failed due to reasons such as installed location where the signboards are installed on the road and outdoor, the lack of monitoring means at all times, and the lack of manpower is often neglected for a long time. Following this flow, this paper proposes a method to detect defective modules by analyzing the images collected through the camera fixed to the front part of the LED display.

Daily monitoring of the heart rate can facilitate detection of heart-related diseases in their early stages. Therefore, providing an easy-to-use and noninvasive heart rate monitoring system has been a very popular research topic in the field of healthcare. One of good candidate methods is to use commonly available cameras and extract information that can help to estimate heart rate from a human face. Generally, such information can be retrieved using two different approaches: photoplethysmography (PPG) and ballistocardiography (BCG). PPG exploits slight color changes caused by blood volume variations during heartbeats; thus, it tends to be vulnerable to unstable lighting conditions. BCG exploits subtle head motions caused by pumped blood travelling through the carotid artery during heartbeats; thus, it is vulnerable to the voluntary head movements that are not related to heartbeats. Nevertheless, most related works use either to estimate the heart rate. In this paper, we propose to combine two approaches to be robust to challenging conditions. Specifically, we explore possible ways to combine raw signals obtained from two approaches and verify that the proposed combination shows better accuracies under challenging conditions, such as voluntary head movements and ambient lighting changes.

Cyber threats such as forced personal information collection and distribution of malicious codes using malicious URLs continue to occur. In order to cope with such cyber threats, a security technologies that quickly detects malicious URLs and prevents damage are required. In a web environment, malicious URLs have various forms and are created and deleted from time to time, so there is a limit to the response as a method of detecting or filtering by signature matching. Recently, researches on detecting and predicting malicious URLs using machine learning techniques have been actively conducted. Existing studies have proposed various features and machine learning algorithms for predicting malicious URLs, but most of them are only suggesting specialized algorithms by supplementing features and preprocessing, so it is difficult to sufficiently reflect the strengths of various machine learning algorithms. In this paper, a system for predicting malicious URLs using multiple machine learning algorithms was proposed, and an experiment was performed to combine the prediction results of multiple machine learning models to increase the accuracy of predicting malicious URLs. Through experiments, it was proved that the combination of multiple models is useful in improving the prediction performance compared to a single model.

Sounds that have been evaluated to be effective in inducing sleep are helpful to reduce sleep disorders. Generally, several sounds have been verified the effects by brainwave experiments, but it cannot be considered on all users because of individual variation for effects. Moreover, the effectiveness for inducing sleep is not known for all new sounds made by creative activities. Therefore, new classification system is required to collect new effective sounds with considering personal brainwave characteristics. In this paper, we propose a new sound classification method by applying improved MinHash cluster to brain waves. The proposed method will classify them through whether it is effective for sleep care by evaluation his brainwave during listening for each sound. In order to prove effectiveness of the proposed classification method, we conducted accuracy experiment for sleep sound classification using verified sleep induction sound. In addition, we have compared time for existing method and proposed method. The former is scored 85% accuracy in the experiment. We confirmed the latter one that the average processing time was reduced to 70%. It is expected to be one of method for pre-screening whether it is effective when a new sound is introduced as a sound for sleep induction.

The web has become indispensable in all of our daily lives. We communicate, study and get information with others through the web. This behavior also continues in the smart phone environment. The biggest problem is that the small display screen of a smart phone degrades the accuracy in selecting or manipulating content for people with low vision. To compensate for this, voice guidance services that combine touch and voice, such as VoiceOver and Talkback, are currently provided to smart phone devices. However, restrictions arise in GUI, TTS control problems, and content expansion and selection. In addition, unnecessary content is also output by voice, which causes fatigue for low vision people to use. In this study, we propose a mobile web browser interface that selects and enlarges a desired area from web browsers and contents, or outputs it as a voice so that people with low vision can easily use the mobile web browser. In this paper, we propose a context selective focusing function that enables selection for each element of web content. In addition, we intend to develop a mobile voice web browser that can enlarge the selected content or output it by voice.

Virtual reality (VR) and augmented reality (AR) technologies can interact intuitively using virtual 3D images. Information signs on buildings that are frequently used by the public play a very important role in conveying information. However, the current information sign has different understanding and interpretation, and it is difficult to communicate quickly and intuitively for interaction. The purpose of this study is to increase the efficiency of information signing through information provision using mobile augmented reality. The InfoAR app not only guides you to the destination of the building, but also acts as a marker for the logo image. The logo image created based on the feature points recognizes the information of the destination. It is converted to AR through a mobile app, allowing users to easily find their destination by obtaining specific information. In addition, it presents interactions that can enlarge, reduce, move, and rotate information about virtual objects augmented through a function called AR. In the future, if user convenience and usability of the interface are added, research is conducted, and the level of completion is improved, such as using VR contents and promotions and business cards, it is expected that highly useful contents will be produced.

Advertising is a communication that aims to persuade you through communication. Local governments are enforcing a lot of budget for outdoor advertisements and exposing advertisements to various media. However, in the case of outdoor advertisements, advertisements are executed without considering the specificity of the exposure environment, so it is inferior to the budget input. Therefore, this study aims to help local advertisement agents, producers, and related parties by classifying components by combining video analysis methodology and video ad analysis methodology for effective message delivery and immersion in local outdoor advertising. As a discussion to provide an efficient production methodology for outdoor advertising video production in the future, it is an extension of exploratory research through content analysis.