• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.33 no.6
• /
• pp.265-274
• /
• 2021

Outlier detection research in ocean data has traditionally been performed using statistical and distance-based machine learning algorithms. Recently, AI-based methods have received a lot of attention and so-called supervised learning methods that require classification information for data are mainly used. This supervised learning method requires a lot of time and costs because classification information (label) must be manually designated for all data required for learning. In this study, an autoencoder based on unsupervised learning was applied as an outlier detection to overcome this problem. For the experiment, two experiments were designed: one is univariate learning, in which only SST data was used among the observation data of Deokjeok Island and the other is multivariate learning, in which SST, air temperature, wind direction, wind speed, air pressure, and humidity were used. Period of data is 25 years from 1996 to 2020, and a pre-processing considering the characteristics of ocean data was applied to the data. An outlier detection of actual SST data was tried with a learned univariate and multivariate autoencoder. We tried to detect outliers in real SST data using trained univariate and multivariate autoencoders. To compare model performance, various outlier detection methods were applied to synthetic data with artificially inserted errors. As a result of quantitatively evaluating the performance of these methods, the multivariate/univariate accuracy was about 96%/91%, respectively, indicating that the multivariate autoencoder had better outlier detection performance. Outlier detection using an unsupervised learning-based autoencoder is expected to be used in various ways in that it can reduce subjective classification errors and cost and time required for data labeling.

• Journal of Intelligence and Information Systems
• /
• v.28 no.2
• /
• pp.191-206
• /
• 2022

Recently, it is a de-facto approach to utilize a pre-trained language model(PLM) to achieve the state-of-the-art performance for various natural language tasks(called downstream tasks) such as sentiment analysis and question answering. However, similar to any other machine learning method, PLM tends to depend on the data distribution seen during the training phase and shows worse performance on the unseen (Out-of-Distribution) domain. Due to the aforementioned reason, there have been many efforts to develop domain-specified PLM for various fields such as medical and legal industries. In this paper, we discuss the training of a finance domain-specified PLM for the Korean language and its applications. Our finance domain-specified PLM, KB-BERT, is trained on a carefully curated financial corpus that includes domain-specific documents such as financial reports. We provide extensive performance evaluation results on three natural language tasks, topic classification, sentiment analysis, and question answering. Compared to the state-of-the-art Korean PLM models such as KoELECTRA and KLUE-RoBERTa, KB-BERT shows comparable performance on general datasets based on common corpora like Wikipedia and news articles. Moreover, KB-BERT outperforms compared models on finance domain datasets that require finance-specific knowledge to solve given problems.

• Journal of the Korean Chemical Society
• /
• v.66 no.2
• /
• pp.78-85
• /
• 2022

Natural white clay was treated with 6 M of H₂SO₄ and heated at 80℃ for 6 h under mechanical stirring and the resulting acid active clay was used as an adsorbent for the removal of Cs⁺ in water. The physicochemical changes of natural white clay and acid active clay were observed by X-ray Fluorescence Spectrometry (XRF), BET Surface Area Analyser and Energy Dispersive X-line Spectrometer (EDX). While activating natural white clay with acid, the part of Al₂O₃, CaO, MgO, SO₃ and Fe₂O₃ was dissolved firstly from the crystal lattice, which bring about the increase in the specific surface area and the pore volume as well as active sites. The specific surface area and the pore volume of acid active clay were roughly twice as high compared with natural white clay. The adsorption of Cs⁺ on acid active clay was increased rapidly within 1 min and reached equilibrium at 60 min. At 25 mg L^- of Cs⁺ concentration, 96.88% of adsorption capacity was accomplished by acid active clay. The adsorption data of Cs⁺ were fitted to the adsorption isotherm and kinetic models. It was found that Langmuir isotherm was described well to the adsorption behavior of Cs⁺ on acid active clay rather than Freundlich isotherm. For adsorption Cs⁺ on acid active clay, the Langmuir isotherm coefficients, Q, was found to be 10.52 mg g^-1. In acid active clay/water system, the pseudo-second-order kinetic model was more suitable for adsorption of Cs⁺ than the pseudo-first-order kinetic model owing to the higher correlation coefficient R² and the more proximity value of the experimental value q_e,exp and the calculated value q_e,cal. The overall results of study showed that acid active clay could be used as an efficient adsorbent for the removal of Cs⁺ from water.

• Korean Journal of Remote Sensing
• /
• v.38 no.6_1
• /
• pp.1181-1189
• /
• 2022

Clouds are composed of tiny water droplets, ice crystals, or mixtures suspended in the atmosphere and cover about two-thirds of the Earth's surface. Cloud detection in satellite images is a very difficult task to separate clouds and non-cloud areas because of similar reflectance characteristics to some other ground objects or the ground surface. In contrast to thick clouds, which have distinct characteristics, thin transparent clouds have weak contrast between clouds and background in satellite images and appear mixed with the ground surface. In order to overcome the limitations of transparent clouds in cloud detection, this study conducted cloud detection focusing on transparent clouds using machine learning techniques (Random Forest [RF], Convolutional Neural Networks [CNN]). As reference data, Cloud Mask and Cirrus Mask were used in MOD35 data provided by MOderate Resolution Imaging Spectroradiometer (MODIS), and the pixel ratio of training data was configured to be about 1:1:1 for clouds, transparent clouds, and clear sky for model training considering transparent cloud pixels. As a result of the qualitative comparison of the study, bothRF and CNN successfully detected various types of clouds, including transparent clouds, and in the case of RF+CNN, which mixed the results of the RF model and the CNN model, the cloud detection was well performed, and was confirmed that the limitations of the model were improved. As a quantitative result of the study, the overall accuracy (OA) value of RF was 92%, CNN showed 94.11%, and RF+CNN showed 94.29% accuracy.

• Journal of the Korean Orthopaedic Association
• /
• v.55 no.5
• /
• pp.405-410
• /
• 2020

Purpose: Pedicle screw insertion has been traditionally used as a surgical treatment for degenerative lumbar spine disease. As an alternative, the cortical-bone trajectory screw allows less invasive posterior lumbar fixation and excellent mechanical stability, as reported in several biomechanical studies. This study evaluated the clinical and radiological results of a case of early failure of cortical-bone screw fixation in posterior fixation and union after posterior decompression. Materials and Methods: This study examined 311 patients who underwent surgical treatment from 2013 to 2018 using cortical orbital screws as an alternative to traditional pedicle screw fixation for degenerative spinal stenosis and anterior spine dislocation of the lumbar spine. Early fixation failure after surgery was defined as fixation failure, such as loosening, pull-out, and breakage of the screw on computed tomography (CT) and radiographs at a follow-up of six months. Results: Early fixation failure occurred in 46 out of 311 cases (14.8%), screw loosening in 46 cases (14.8%), pull-out in 12 cases (3.9%), and breakage in four cases (1.3%). An analysis of the site where the fixation failure occurred revealed the following, L1 in seven cases (15.2%), L2 in three cases (6.5%), L3 in four cases (8.7%), L4 in four cases (8.7%), L5 in four cases (8.7%), and S1 in 24 cases (52.2%). Among the distal cortical bone screws, fixation failures such as loosening, pull-out, and breakage occurred mainly in the S1 screws. Conclusion: Cortical-bone trajectory screw fixation may be an alternative with comparable clinical outcomes or fewer complications compared to conventional pedicle screw fixation. On the other hand, in case with osteoporosis and no anterior support structure particularly at L5-S1 fusion sites were observed to have result of premature fixation failures such as relaxation, pull-out, and breakage.

• Journal of Bio-Environment Control
• /
• v.31 no.2
• /
• pp.79-89
• /
• 2022

With the wide use of greenhouses, the working hours have been increasing inside the greenhouse for workers. In the closed ventilated greenhouse, the internal environment has less affected to external weather during making a suitable temperature for crop growth. Greenhouse workers are exposed to organic dust including soil dust, pollen, pesticide residues, microorganisms during tillage process, soil grading, fertilizing, and harvesting operations. Therefore, the health status and working environment exposed to workers should be considered inside the greenhouse. It is necessary to secure basic data on particulate matter (PM) concentrations in order to set up dust reduction and health safety plans. To understand the PM concentration of working environment in greenhouse, the PM concnentrations were monitored in the cut-rose and Hallabong greenhouses in terms of PM size, working type, and working period. Compare to no-work (move) period, a significant increase in PM concentration was found during tillage operation in Hallabong greenhouse by 4.94 times on TSP (total suspended particle), 2.71 times on PM-10 (particle size of 10 ㎛ or larger), and 1.53 times on PM-2.5, respectively. During pruning operation in cut-rose greenhouse, TSP concentration was 7.4 times higher and PM-10 concentration was 3.2 times higher than during no-work period. As a result of analysis of PM contribution ratio by particle sizes, it was shown that PM-10 constitute the largest percentage. There was a significant difference in the PM concentration between work and no-work periods, and the concentration of PM during work was significant higher (p < 0.001). It was found that workers were generally exposed to a high level of dust concentration from 2.5 ㎛ to 35.15 ㎛ during tillage operation.

• Journal of Korea Port Economic Association
• /
• v.38 no.1
• /
• pp.129-142
• /
• 2022

In the past, interest in air pollution was concentrated on greenhouse gases, but in recent years, interest in fine dust has been increasing. The media and environmental organizations continue to emphasize air pollution caused by fine dust. The awareness of fine dust is increasing, and air pollution generated at ports is analyzed to be serious as a domestic factor excluding foreign inflows. Recognizing this, in order to reduce air pollution generated at ports, special laws on improving air quality, such as port areas, have been enacted in Korea, and attempts are being made to curb air pollution caused by ports. In this law, it is a policy that regulates air pollutants generated not only by ships but also throughout ports such as vehicles and unloading machines, and representative are ECA, VSR, and AMP. This study attempted to analyze the effects of these eco-friendly policies at Incheon Port. First of all, a study was conducted to calculate emissions assuming that there was no policy, analyze each policy, and finally calculate and compare actual emissions reflecting all policies. The methodology presented by the European Environmental Administration and the U.S. Environmental Protection Agency was used, and pollutants to be analyzed were analyzed for sulfur oxides (SO_X), carbon monoxide (CO), nitrogen oxides (NO_X), total floating substances (TSP), fine dust and ultrafine dust (PM₁₀, PM_2.5) and ammonia (NH₃). As a result of the analysis, it was analyzed that the actual emission reflecting all policies was about 4,097 tons/year, which had an emission reduction effect of about 760 tons/year compared to about 4,857 tons/year when the policy was not reflected. When the effects of each policy were analyzed individually, it was found that ECA 4,111 tons/year, VSR 4,854 tons/year, and AMP 4,843 tons of air pollutant emissions occurred The results of this study can be used as basic data and evidence for policy establishment related to the atmospheric environment at Incheon Port.

• Korean Journal of Ecology and Environment
• /
• v.54 no.4
• /
• pp.272-279
• /
• 2021

A pump-type eDNA filtering system that can control voltage and hydraulic pressure respectively has been developed, and applied a filter case that can filter out without damaging the filter. The filtering performance of the developed system was evaluated by comparing the eDNA concentration with the conventional vacuum-pressured filtering method at the catchment conduit intake reservoir. The developed system was divided into a voltage control (manual pump system) method and a pressure control (automatic pump system) method, and the pressure was measured during filtering and the pressure change of each system was compared. The voltage control method started with 65 [KPa] at the beginning of the filtering, and as the filtering time elapsed, the amount of filtrate accumulated in the filter increased, so the pressure gradually increased. As a result of controlling the pressure control method to maintain a constant pressure according to the designed algorithm, there was a difference in the width of the hydraulic pressure fluctuation during the filtering process according to the feedback time of the hydraulic pressure sensor, and it was confirmed that the pressure was converged to the target pressure. The filtering performance of the developed system was confirmed by measuring the eDNA concentration and comparing the voltage control method and the hydraulic control method with the control group. The voltage control method obtained similar results to the control group, but the hydraulic control method showed lower results than the control group. It is considered that the low eDNA concentration in the hydraulic control method is due to the large pressure deviation during filtering and maintaining a constant pressure during the filtering process. Therefore, rather than maintaining a constant pressure during filtering, it was confirmed that a voltage control method in which the pressure is gradually increased as the filtrate increases with the lapse of filtering time is suitable for collecting eDNA. As a result of comparing the average concentration of eDNA in lentic zone and lotic zone as a control group, it was found to be 96.2 [ng µL^-1] and 88.4 [ng µL^-1l], respectively. The result of comparing the average concentration of eDNA by the pump method was also high in the lentic zone sample as 90.7 [ng µL^-1] and 74.8 [ng µL^-1] in the lentic zone and the lotic zone, respectively. The high eDNA concentration in the lentic zone is thought to be due to the influence of microorganisms including the remaining eDNA.

• Journal of Korea Port Economic Association
• /
• v.38 no.3
• /
• pp.133-149
• /
• 2022

The Port logistics is a pivotal industry in strengthening competitiveness for countries and companies and plays an important role in accounting for the majority of Korea's import and export volume, so its importance among Korea's industries is truly enormous. The port, which plays such an important role, is simultaneously unloaded, transferred into the dock, and stored, and it can be said that it is composed of a very complex flow. In particular, port unloading is an environment in which port unloading workers cannot escape the risk of disasters as large cargoes and cranes for container transport move, and numerous disasters are hidden and lead to serious disasters. Through literature studies on the Serious Accident Penalty Act, this study investigated how the Serious Accident Penalty Act affected port unloading workers and whether there was a change in satisfaction with safety education according to the enforcement of the Serious Accident Penalty Act. Through this study, it was found that port unloading workers were more aware of safety than managers in cognitive, attitude, and experience factors. Since the port unloading manager is not put into the unloading work, the perception of safety was analyzed to be lower than that of workers. Since managers should educate workers on safe work and be alert to safety, it seems that port unloading managers will also need to change their perception of safety. The Act on the Punishment of Severe Disasters took effect on January 27, 22, and in the early stages of the enforcement of the law, many confusing and ambiguous parts are found for those in charge of performing practical tasks on the front line until the law is settled. It seems necessary to revise the Act on the Punishment of Severe Disasters for the purpose of clear standards for the law and no safety accidents in the field to prevent confusion in work. In addition, in order to derive more meaningful research results in the future, it is expected that research on the perception of disasters by workers and managers will be continuously needed by deepening literature research to find various factors and using investigation methods such as interviews.

• Korean Journal of Mineralogy and Petrology
• /
• v.35 no.3
• /
• pp.313-331
• /
• 2022

Natural or native abiotic molecular hydrogen (H₂) is a major component in natural gas, however yet its importance in the global energy sector's usage as clean and renewable energy is underestimated. Here we review the occurrence and geological settings of native hydrogen to demonstrate the much widesprease H₂ occurrence in nature by comparison with previous estimations. Three main types of source rocks have been identified: (1) ultramafic rocks; (2) cratons comprising iron (Fe²⁺)-rich rocks; and (3) uranium-rich rocks. The rocks are closely associated with Precambrian crystalline basement and serpentinized ultramafic rocks from ophiolite and peridotite either at mid-ocean ridges or within continental margin(Zgonnik, 2020). Inorganic geological processes producing H₂ in the source rocks include (a) the reduction of water during the oxidation of Fe²⁺ in minerals (e.g., olivine), (b) water splitting due to radioactive decay, (c) degassing of magma at low pressure, and (d) the reaction of water with surface radicals during mechanical breaking (e.g., fault) of silicate rocks. Native hydrogen are found as a free gas (51%), fluid inclusions in various rock types (29%), and dissolved gas in underground water (20%) (Zgonnik, 2020). Although research on H₂ has not yet been carried out in Korea, the potential H₂ reservoirs in the Gyeongsang Basin are highly probable based on geological and geochemical characteristics including occurrence of ultramafic rocks, inter-bedded basaltic layers and iron-copper deposits within thick sedimentary basin and igneous activities at an active continental margin during the Permian-Paleogene. The native hydrogen is expected to be clean and renewable energy source in the near future. Therefore it is clear that the origin and exploration of the native hydrogen, not yet been revealed by an integrated studies of rock-fluid interaction studies, are a field of special interest, regardless of the presence of economic native hydrogen reservoirs in Korea.