In this paper, we present a new architecture for semantic segmentation. Semantic segmentation aims at a pixel-wise classification which is important to fully understand images. Previous semantic segmentation networks use features of multi-layers in the encoder to predict final results. However, they do not contain various receptive fields in the multi-layers features, which easily lead to inaccurate results for boundaries between different classes and small objects. To solve this problem, we propose a multi-path feature fusion module that allows for features of each layers to contain various receptive fields by use of a set of dilated convolutions with different dilatation rates. Various experiments demonstrate that our method outperforms previous methods in terms of mean intersection over unit (mIoU).

A Wasserstein GAN(WGAN), optimum in terms of minimizing Wasserstein distance, still suffers from inconsistent convergence or unexpected output due to inherent learning instability. It is widely known some kinds of restriction on the discriminative function should be considered to solve such problems, which implies the importance of Lipschitz continuity. Unfortunately, there are few known methods to satisfactorily maintain the Lipschitz continuity of the discriminative function. In this paper we propose techniques to stably maintain the Lipschitz continuity of the discriminative function by adding effective regularization terms to the objective function, which limit the magnitude of the gradient vectors of the discriminator to one or less. Extensive experiments are conducted to evaluate the performance of the proposed techniques, which shows the single-sided penalty improves convergence compared with the gradient penalty at the early learning process, while the proposed additional penalty increases inception scores by 0.18 after 100,000 number of learning.

With the recent increase in the number of households raising pets, various engineering studies have been underway for pets. The final purpose of this study is to automatically generate situation-sensitive captions that can express implicit intentions based on the behavior and sound of cats by embedding the already mature behavioral detection technology of pets as basic element technology in the video capturing research. As a pilot project to this end, this paper proposes a high-level capturing system using optical-flow, RGB, and sound information of cat videos. That is, the proposed system uses video datasets collected in an actual breeding environment to extract feature vectors from the video and sound, then through hierarchical LSTM encoder and decoder, to identify the cat's behavior and its implicit intentions, and to perform learning to create context-sensitive captions. The performance of the proposed system was verified experimentally by utilizing video data collected in the environment where actual cats are raised.

Research in the field of health lighting has continued to advance to reproduce the color temperature of natural light which periodically changes. However, most of this research could only reproduce a uniform circadian color temperature of natural light, therefore failing to realize the characteristics of the circadian cycle of color temperature difference by latitude and longitude. To reproduce the color temperature of natural light on which the characteristics of a region are reflected, the collection technology of real-time characteristics of natural light is needed. If the color temperatures which are not within a periodical pattern due to climate changes, etc., are measured, it will be difficult to judge the occurrence (presence) of the anomalies and to reproduce the circadian cycle of the color temperature of natural light. Therefore, this study proposes an algorithm for judging the anomalies in real time based on the sliding window to reproduce the color temperature of natural light. First, the natural light characteristics DB collected through the on-site measurement were analyzed, the differential values at a one-minute interval were calculated and examined, and then representative color temperature circadian patterns by solar terms were drawn. The anomalies were then detected by the application of the sliding window that calculated the deviation of the color temperature for the measured color temperature data set, which was collected through RGB sensors, while moving along the time sequence. In addition, the presence of anomalies was verified through the comparison study between the detection results and the representative circadian cycle of the color temperature by solar term. The judgment method for the anomalies from the measured color temperature of natural light was proposed for the first time, confirming that the proposed method was capable of detecting the anomalies with an average accuracy of 94.6%.

Science museum are dealing with exhibits on field of changing science and technology, and previous research suggested that exhibits replacement should carried out at least every 5 years. In order to efficiently replace exhibits within a limited budget, various studies analyzed visitors' preferences to exhibits. Recently, studies use various technologies to collect the data on visitors' preferences automatically, but almost of studies had a high dependency on their visitors such as visitors needed to carry specific sub-devices in the museums for gathering data. As complementing the limitations of previous research, this study introduces the improved method which is able to automatically collect and quantify visitors' preferences to exhibits using TensorFlow, a deep learning technology. By the proposed analysis method, it was possible to collect 2,520 data of visitors' experience on exhibits in totality. Based on collected data, attraction power and holding power indicating the preference of visitors on exhibits were able to be calculated. The result also confirmed antecedent research conclusion that the attraction power and holding power of the exhibit which consists of 3 dimensional structures work are higher than other exhibits. As a conclusion, the proposed method will provide more convenient data collection method for detecting visitors' preference.

In this paper, we propose a task distribution scheme in fog computing environment considering opportunistic fog computing nodes. As latency is one of the important performance metric for IoT(Internet of Things) applications, there have been lots of researches on the fog computing system. However, since the load can be concentrated to the specific fog computing nodes due to the spatial and temporal IoT characteristics, the load distribution should be considered to prevent the performance degradation. Therefore, this paper proposes a task distribution scheme which considers the static as well as opportunistic fog computing nodes according to their mobility feature. Especially, based on the task requirements, the proposed scheme supports the delay sensitive task processing at the static fog node and delay in-sensitive tasks by means of the opportunistic fog nodes for the task distribution. Based on the performance evaluation, the proposed scheme shows low service response time compared to the conventional schemes.

The user's password must be encrypted one-way through the hash function and stored in the database. Widely used hash functions such as MD5 and SHA-1 have also been found to have vulnerabilities, and hash functions that are considered safe can also have vulnerabilities over time. Salt enhances password security by adding it before or after the password before putting it to the hash function. In the case of the existing Salt, even if it is randomly assigned to each user, once it is assigned, it is a fixed value in a specific column of the database. If the database is exposed to an attacker, it poses a great threat to password cracking. In this paper, we suggest variable-dynamic Salt that dynamically changes according to the user's password during the login process. The variable-dynamic Salt can further enhance password security during login process by making it difficult to know what the Salt is, even if the database or source code is exposed.

As IoT(Internet Of Things) devices like a smart sensor have constrained power sources, a power strategy is critical in WSN(Wireless Sensor Networks). Therefore, it is necessary to figure out the residual power of each sensor node for managing power strategies in WSN, which, however, requires additional data transmission, leading to more power consumption. In this paper, a residual power estimation method was proposed, which uses ignorantly small amount of power consumption in the resource-constrained wireless networks including WSN. A residual power prediction is possible with the least data transmission by using Machine Learning method with some training data in this proposal. The performance of the proposed scheme was evaluated by machine learning method, simulation, and analysis.

The 4th industrial revolution needs a fusion of artificial intelligence, robotics, the Internet of Things (IoT), edge computing, and other technologies. For the fusion of technologies, cloud computing technology can provide flexible and high-performance computing resources so that cloud computing can be the foundation technology of new emerging services. The emerging services become a global-scale, and require much higher performance, availability, and reliability. Public cloud providers already provide global-scale services. However, their services, costs, performance, and policies are different. Enterprises/ developers to come out with a new inter-operable service are experiencing vendor lock-in problems. Therefore, multi-cloud technology that federatively resolves the limitations of single cloud providers is required. We propose a software platform, denoted as Cloud-Barista. Cloud-Barista is a multi-cloud service common platform for federating multiple clouds. It makes multiple cloud services as a single service. We explain the functional architecture of the proposed platform that consists of several frameworks, and then discuss the main design and implementation issues of each framework. To verify the feasibility of our proposal, we show a demonstration which is to create 18 virtual machines on several cloud providers, combine them as a single resource, and manage it.

In the case of laptops and tablet PC's that replace desktop, it uses the Windows operating system to provide various functions depending on operating system dependency, the Windows operating system does not support real-time processing because it uses multi-level feedback queue scheduling that extends round-robin scheduling. Also, since the initial value of Local APIC Counter can not be obtained from the Windows 8, the real-time processing function provided through the existing RTiK does not work. In this paper, we calculate Local APIC Counter value by using MSR_FSB_FREQ register to support real-time processing function on tablet PC's. We designed and implemented RTiK+, which provides real-time processing function to guarantee the periodicity by calculating the operation time of accurate timer. In order to verify and evaluate the performance of the implemented the RTiK+, the period was measured by using the Read Time-Stamp Counter(RDTSC) instruction and it was confirmed that it operates normally at 1ms and 0.1ms period.

To increase the quality of the system in the subjective-type question grading and document classification, we need the paragraph detection. But it is not easy because it is accompanied by semantic analysis. Many researches on the paragraph detection solve the detection problem using the word based clustering method. However, the word based method can not use the order and dependency relation between words. This paper suggests the paragraph detection system using syntactic-semantic relation between words with the Korean syntactic-semantic analysis. This system is the hybrid system of word based, concept based, and syntactic-semantic tree based detection. The experiment result of the system shows it has the better result than the word based system. This system will be utilized in Korean subjective question grading and document classification.

Korean education in China began at Peking University in the 1950s. At present, The Korean language education in China has made remarkable progress in many aspects such as scale, scope, level and achievement. In addition, with the increasingly frequent economic and cultural exchanges or cooperation between China and South Korea and the increasing trend of internationalization, the prosperity of Korean wave culture and the sound development of China-South Korea relations, the country's demand for Korean language talents is increasing day by day. However, with the rise of Korean education in China in recent years, some hidden problems also surfaced. In this paper, the Korean language proficiency test(TOPIK) is used to evaluate the Korean language proficiency of Korean learners, and Chinese juniors are used to evaluate the Korean language proficiency. In addition, a questionnaire survey was conducted to analyze the learning media of Chinese Korean learners at the present stage, and the relationship between learning media and learning outcomes was concluded. At the same time, deficiencies and problems existed in Korean education in colleges and universities were proposed and their own ideas were put forward.

Due to Covid-19, the un-contact culture has affected society as a whole, and the methods of education conducted offline has been greatly affected. In the private education of preparing for university entrance, the public official examinations and certification acquisition, the method of online education has been shown to have positive effects. While private class and school class which have offered in off-line to cope with rapid changes caused various problems such as decline in quality for education. Due to the characteristic of design class, practical training is important. As interactive feedback between students and educators is more important than one-way of delivering knowledge while class is conducted in online, educators have a challenge when they prepare for class. This study handles the methods of online education for the purpose of practical education methods in university nowadays, Especially, the non-face-to-face education methods for game animation production. Based on this study, I propose an effective educational method with non-face-to-face class that allows students to be satisfied and increases their knowledge, beyond face-to-face class.

The purpose of this study is to explore the acceptance intent of groups of experts and users for converged medical equipment using NeuroTech. This healthcare equipment is applied with technology that simultaneously enhances body function and cognitive function. Interviews were conducted with groups of experts and users to investigate the acceptance of equipment. The interviews include health problems for the elderly, the need to develop medical equipment, and the effectiveness. After recording the collected historical data, we performed the data analysis in accordance with the procedures of the phenomenological analysis method. The total number of people interviewed was 27, 16 experts and 11 users. As a result of the interview, we derived seven topic groups for each of the 12 topic groups and user groups. The implications were derived based on the results.

Conspicuous consumption is often investigated in Korea and consumers purchase a product to publicly display economic power or social status. The existing studies argue that brand loyalty and luxury brand preferences escalate conspicuous consumption, and it is important to identify different visual perception to understand irrational consumers' behavior. Car emblem as brands' visual identity represents the brand's image, mission and message. Irrational consumers thus are expected have different visual perception to the emblem or logo. To understand the impact of irrational consumer behavior on visual perception, this study explored how brand loyalty and luxury brand preferences impact on the visual perception of car emblems between Korean domestic brand and imported brand. This study used eye tracking technology to measure the respondents' attention on an emblem of the cars. Eye tracking tools analyze fixations and measure the attention on a specific area of interests (AOI). This study explored the maintaining time and frequency on AOI, as well as explored the size of pupil to identify the respondents' attention on the AOI. In this experiment, the logos of domestic cars (i.e., Hyundai and Kia) and foreign cars (i.e., Mercedes-Benz and BMW) were swapped to see how irrational consumer behavior impact on visual perception of car emblem.

With the continuous development and progress of science and technology, the field of mixed reality applications has involved scientific research, medicine, entertainment, education, information dissemination, and people's daily lives; This paper will focus on the application of mixed reality in the field of education and teaching, relying on the characteristics of mixed reality technology. This article will use the combination of mixed reality and smart glasses to fully consider the characteristics of photography darkroom teaching, design and produce a mixed reality photography darkroom teaching software, and explore the feasibility of the application of mixed reality in teaching. This paper will use literature research methods, practical research methods, and survey methods as the main methods of research topics to determine the importance, feasibility, and necessity of the relevant theories studied in this paper. The application of mixed reality technology in photography darkroom teaching can not only solve many problems in the existing darkroom teaching methods, but also develop and expand students' Independent learning ability. It is concluded from this that mixed reality teaching software has the feasibility of sustainable development, which brings particular application value to the development and research of other education and art discipline software.