• Journal of KIISE:Software and Applications
• /
• v.26 no.11
• /
• pp.1261-1269
• /
• 1999

다층 퍼셉트론은 다양한 응용 분야에 성공적으로 적용되고 있는 대표적인 신경회로망 모델이다. 그러나 다층 퍼셉트론의 학습에서 나타나는 플라토에 기인한 느린 학습 속도와 지역 극소는 실제 응용문제에 적용함에 있어서 가장 큰 문제로 지적되어왔다. 이 문제를 해결하기 위해 여러 가지 다양한 학습알고리즘들이 개발되어 왔으나, 계산의 비효율성으로 인해 실제 문제에는 적용하기 힘든 예가 많은 등, 현재까지 만족할 만한 해결책은 제시되지 못하고 있다. 본 논문에서는 다층퍼셉트론의 베이시스 함수로 사용되는 시그모이드 함수를 보다 일반화된 형태로 정의하여 사용함으로써 학습에 있어서의 플라토를 완화하고, 지역극소에 빠지는 것을 줄이는 접근방법을 소개한다. 본 방법은 기존의 변형된 가중치 수정식을 사용한 학습 속도 향상의 방법들과는 다른 접근 방법을 택함으로써 기존의 방법들과 함께 사용하는 것이 가능하다는 특징을 갖고 있다. 제안하는 방법의 성능을 확인하기 위하여 간단한 패턴 인식 문제들에의 적용 실험 및 기존의 학습 속도 향상 방법을 함께 사용하여 시계열 예측 문제에 적용한 실험을 수행하였고, 그 결과로부터 제안안 방법의 효율성을 확인할 수 있었다. Abstract A multilayer perceptron is the most well-known neural network model which has been successfully applied to various fields of application. Its slow learning caused by plateau and local minima of gradient descent learning, however, have been pointed as the biggest problems in its practical use. To solve such a problem, a number of researches on learning algorithms have been conducted, but it can be said that none of satisfying solutions have been presented so far because the problems such as computational inefficiency have still been existed in these algorithms. In this paper, we propose a new learning approach to minimize the effect of plateau and reduce the possibility of getting trapped in local minima by generalizing the sigmoidal function which is used as the basis function of a multilayer perceptron. Adapting a new approach that differs from the conventional methods with revised updating equation, the proposed method can be used together with the existing methods to improve the learning performance. We conducted some experiments to test the proposed method on simple problems of pattern recognition and a problem of time series prediction, compared our results with the results of the existing methods, and confirmed that the proposed method is efficient enough to apply to the real problems.

• KEPCO Journal on Electric Power and Energy
• /
• v.5 no.3
• /
• pp.141-147
• /
• 2019

Recently, the biggest issue in the electricity industry is the increase in renewable energy, and various technologies are being developed to ensure the capacity of the power system. In addition, super-grids linking power systems are being pushed to utilize eco-friendly energy between countries and regions worldwide. The HVDC transmission technology is required to link the power network between regions with different characteristics of the power system such as frequency and voltage. Until now, Korea has applied HVDC transmission technology that connects mainland and Jeju Island with submarine cables. But, the HVDC transmission technology is still developing for long-distance high-capacity power transmission from power parks on the east coast to load-tight areas near the metropolitan area. Considering the high population density and mountainous domestic environment, it is pushing for commercialization of the design technology of the ${\pm}500kV$ Double Bipole with metallic return wire transmission line to transmit large-scale power of 8 GW using minimal right of ways. In this paper, the insulation characteristics were studied for the design of double-bipole transmission tower with metallic return wire, which is the first time in the world. And the air insulation characteristics resistant to the various overvoltage phenomena occurring on transmission lines were verified through a full-scale impulse voltage test.

• Journal of Broadcast Engineering
• /
• v.26 no.4
• /
• pp.409-418
• /
• 2021

Conformal coating is a technology that protects PCB(Printed Circuit Board) and minimizes PCB failures. Since the defects in the coating are linked to failure of the PCB, the coating surface is examined for air bubbles to satisfy the successful conditions of the conformal coating. In this paper, we propose an algorithm for detecting problematic bubbles in high-risk groups by applying image signal processing. The algorithm consists of finding candidates for problematic bubbles and verifying candidates. Bubbles do not appear in visible light images, but can be visually distinguished from UV(Ultra Violet) light sources. In particular the center of the problematic bubble is dark in brightness and the border is high in brightness. In the paper, these brightness characteristics are called valley and mountain features, and the areas where both characteristics appear at the same time are candidates for problematic bubbles. However, it is necessary to verify candidates because there may be candidates who are not bubbles. In the candidate verification phase, we used convolutional neural network models, and ResNet performed best compared to other models. The algorithms presented in this paper showed the performance of precision 0.805, recall 0.763, and f1-score 0.767, and these results show sufficient potential for bubble test automation.

• Journal of Korea Water Resources Association
• /
• v.51 no.12
• /
• pp.1207-1216
• /
• 2018

Discharge or water level predictions at tidally affected river reaches are currently still a great challenge in hydrological practices. This research aims to predict water level of the tide dominated site, Jamsu bridge in the Han River downstream. Physics-based hydrodynamic approaches are sometimes not applicable for water level prediction in such a tidal river due to uncertainty sources like rainfall forecasting data. In this study, TensorFlow deep learning framework was used to build a deep neural network based LSTM model and its applications. The LSTM model was trained based on 3 data sets having 10-min temporal resolution: Paldang dam release, Jamsu bridge water level, predicted tidal level for 6 years (2011~2016) and then predict the water level time series given the six lead times: 1, 3, 6, 9, 12, 24 hours. The optimal hyper-parameters of LSTM model were set up as follows: 6 hidden layers number, 0.01 learning rate, 3000 iterations. In addition, we changed the key parameter of LSTM model, sequence length, ranging from 1 to 6 hours to test its affect to prediction results. The LSTM model with the 1 hr sequence length led to the best performing prediction results for the all cases. In particular, it resulted in very accurate prediction: RMSE (0.065 cm) and NSE (0.99) for the 1 hr lead time prediction case. However, as the lead time became longer, the RMSE increased from 0.08 m (1 hr lead time) to 0.28 m (24 hrs lead time) and the NSE decreased from 0.99 (1 hr lead time) to 0.74 (24 hrs lead time), respectively.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.18 no.3
• /
• pp.95-105
• /
• 2019

In recent years, image processing techniques for detecting road surface damaged spot have been actively researched. Especially, it is mainly used to acquire images through a smart phone or a black box that can be mounted in a vehicle and recognize the road surface damaged region in the image using several algorithms. In addition, in conjunction with the GPS module, the exact damaged location can be obtained. The most important technology is image processing algorithm. Recently, algorithms based on artificial intelligence have been attracting attention as research topics. In this paper, we will also discuss artificial intelligence image processing algorithms. Among them, an object detection method based on an region-based convolution neural networks method is used. To improve the recognition performance of road surface damage objects, 600 road surface damaged images and 1500 general road driving images are added to the learning database. Also, supervised learning using background object recognition method is performed to reduce false alarm and missing rate in road surface damage detection. As a result, we introduce a new method that improves the recognition performance of the algorithm to 8.66% based on average value of mAP through the same test database.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.36 no.6
• /
• pp.457-467
• /
• 2018

Currently, the regulation and rules for public surveying and the UCPs (Unified Control Points) adapts those of the triangulated traverse surveying. In addition, such regulations do not take account of the unique characteristics of GNSS (Global Navigation Satellite System) surveying, thus there are difficulties in field work and data processing afterwards. A detailed procesure of GNSS processing has not yet been described either, and the verification of accuracy does not follow the generic standards. In order to propose an appropriate procedure for field surveys, we processed a short session (30 minutes) based on the scenarios similar to actual situations. The reference network in Seoul was used to process the same data span for 3 days. The temporal variation during the day was evaluated as well. We analyzed the accuracy of the estimated coordinates depending on the parameterization of tropospheric delay, which was compared with the 24-hr static processing results. Estimating the tropospheric delay is advantageous for the accuracy and stability of the coordinates, resulting in about 5 mm and 10 mm of RMSE (Root Mean Squared Error) for horizontal and vertical components, respectively. Based on the test results, we propose a procedure to estimate the daily solution and then combine them to estimate the final solution by applying the minimum constraints (no-net-translation condition). It is necessary to develop a web-based processing system using a high-end softwares. Additionally, it is also required to standardize the ID of the public control points and the UCPs for the automatic GNSS processing.

• Journal of the Korean Data Analysis Society
• /
• v.20 no.6
• /
• pp.3087-3099
• /
• 2018

Due to the constant changes of companies' global networks, the expansion of global e-commerce as well as the market-oriented global supply chain management, global enterprises are strategically selecting and entering into viable countries able to become global footholds. Therefore, this study aims to scrutinize the trend of changes in the global networks of Korean companies by analyzing the current overseas countries over the past decade. From the analysis, it has been found that there is a significant difference in the priorities of targeting countries among shipping, manufacturing and logistics companies. Logistics companies preferred to enter Germany first while they attached to a lower priority to Singapore. Manufacturing companies had a lower priority to advance to India, while they preferred to advance to Mexico; however, shipping companies were analyzed to prefer to enter the US. In addition, all of these companies identified the importance of securing volume and network by entering overseas markets to achieve economies of scale and scope and to maintain global competitiveness. Joint overseas expansion of manufacturers with shipping and logistics companies can be recommended to facilitate the entry and thus, enhance global competitiveness and service capabilities and also secure new growth engines.

• Genes and Genomics
• /
• v.40 no.11
• /
• pp.1213-1223
• /
• 2018

Pelagic larval dispersal habits influence the population genetic structure of marine mollusk organisms via gene flow. The genetic information of the clam Gomphina aequilatera (short larval stage, 10 days) which is ecologically and economically important in the China coast is unknown. To determine the influence of planktonic larval duration on the genetic structure of G. aequilatera. Mitochondrial markers, cytochrome oxidase subunit i (COI) and 12S ribosomal RNA (12S rRNA), were used to investigate the population structure of wild G. aequilatera specimens from four China Sea coastal locations (Zhoushan, Nanji Island, Zhangpu and Beihai). Partial COI (685 bp) and 12S rRNA (350 bp) sequences were determined. High level and significant $F_{ST}$ values were obtained among the different localities, based on either COI ($F_{ST}=0.100-0.444$, P<0.05) or 12S rRNA ($F_{ST}=0.193-0.742$, P<0.05), indicating a high degree of genetic differentiation among the populations. The pairwise $N_m$ between Beihai and Zhoushan for COI was 0.626 and the other four pairwise $N_m$ values were >1, indicating extensive gene flow among them. The 12S rRNA showed the same pattern. AMOVA test results for COI and 12S rRNA indicated major genetic variation within the populations: 77.96% within and 22.04% among the populations for COI, 55.73% within and 44.27% among the populations for 12S rRNA. A median-joining network suggested obvious genetic differentiation between the Zhoushan and Beihai populations. This study revealed the extant population genetic structure of G. aequilatera and showed a strong population structure in a species with a short planktonic larval stage.

• Research in Plant Disease
• /
• v.25 no.3
• /
• pp.149-155
• /
• 2019

Nematode-trapping fungi develop trap and consume nematodes are an important part of the subsoil ecosystem and they share a special predator-prey relationship. Four nematode-trapping species, there with adhesive network, Arthrobotrys oligospora, A. sinensis, A. thaumasia and one with constricting ring, Drechslerella brochopaga were collected from soils in Korea and tested their predacity against 12 different nematode species. They were three feeding groups, plant-parasitic (Meloidogyne incognita and Pratylenchus penetrans), fungivorous (Aphelenchus avenae), bacteriovorous (Betlerius sp. and Diplogasteritus sp. in diplogasterid, Panagrolaimus labiatus, P. multidentatus in panagrolaimid, Mesorhabditis irregularis, Pelodera strongyloides and Rhabditis sp., in rhabditid, and Acrobeloides sp. in cephalobid). Results showed that nematode-trapping fungi successfully captured most of nematodes in Petri dish in the group of plant-parasitic nematodes and rhabditids, moderately and variably in other nematodes in 15 days. But it didn't captured A. avenae and Acrobeloides sp. both belongs to c-p group 2. Numbers of Acrobeloides sp. and A. avenae even increased during the test period. The results of this study indicated that nematode-trapping fungi may have specificity among nematode species.

• Journal of Pharmacopuncture
• /
• v.22 no.3
• /
• pp.166-170
• /
• 2019

Objectives: Attentional and memory functions are important aspects of neural plasticity that, theoretically, should be amenable to pharmacopuncture treatments. A previous study from our laboratory suggested that quantitative electroencephalographic (qEEG) measurements of theta/beta ratio (TBR), an index of attentional control, may be indicative of academic performance in a first-semester medical school course. The present study expands our prior report by extracting and analyzing data on frontal theta and beta asymmetries. We test the hypothesis that the amount of frontal theta and beta asymmetries (fTA, fBA), are correlated with TBR and academic performance, thereby providing novel targets for pharmacopuncture treatments to improve cognitive performance. Methods: Ten healthy male volunteers were subjected to 5-10 min of qEEG measurements under eyes-closed conditions. The qEEG measurements were performed 3 days before each of first two block examinations in anatomy-physiology, separated by five weeks. Amplitudes of the theta and beta waveforms, expressed in ${\mu}V$, were used to compute TBR, fTA and fBA. Significance of changes in theta and beta EEG wave amplitude was assessed by ANOVA with post-hoc t-testing. Correlations between TBR, fTA, fBA and the raw examination scores were evaluated by Pearson's product-moment coefficients and linear regression analysis. Results: fTA and fBA were found to be negatively correlated with TBR (P<0.03, P<0.05, respectively) and were positively correlated with the second examination score (P<0.03, P=0.1, respectively). Conclusion: Smaller fTA and fBA were associated with lower academic performance in the second of two first-semester medical school anatomy-physiology block examination. Future studies should determine whether these qEEG metrics are useful for monitoring changes associated with the brain's cognitive adaptations to academic challenges, for predicting academic performance and for targeting phamacopuncture treatments to improve cognitive performance.