• Journal of Korea Water Resources Association
• /
• v.56 no.4
• /
• pp.261-272
• /
• 2023

In recent years, natural disasters such as heavy rainfall and typhoons have occurred more frequently, and their severity has increased due to climate change. The Korea Meteorological Administration (KMA) currently uses the same criteria for all regions in Korea for watch and warning based on the maximum cumulative rainfall with durations of 3-hour and 12-hour to reduce damage. However, KMA's criteria do not consider the regional characteristics of damages caused by heavy rainfall and typhoon events. In this regard, it is necessary to develop new criteria considering regional characteristics of damage and cumulative rainfalls in durations, establishing four stages: blue, yellow, orange, and red. A classification model, called DSCM (Disaster Severity Classification Model), for the four-stage disaster severity was developed using four machine learning models (Decision Tree, Support Vector Machine, Random Forest, and XGBoost). This study applied DSCM to local governments of Seoul, Incheon, and Gyeonggi Province province. To develop DSCM, we used data on rainfall, cumulative rainfall, maximum rainfalls for durations of 3-hour and 12-hour, and antecedent rainfall as independent variables, and a 4-class damage scale for heavy rain damage and typhoon damage for each local government as dependent variables. As a result, the Decision Tree model had the highest accuracy with an F1-Score of 0.56. We believe that this developed DSCM can help identify disaster risk at each stage and contribute to reducing damage through efficient disaster management for local governments based on specific events.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.49 no.3
• /
• pp.15-26
• /
• 2012

Recently, due to applicability increase of vector data based digital map for geographic information and evolution of geographic measurement techniques, large volumed GIS(geographic information service) services having high resolution and large volumed data are flowing actively. This paper proposed an efficient vector map compression technique using the SEC(spatial energy compaction) based on classified bins for the vector map having 1cm detail and hugh range. We encoded polygon and polyline that are the main objects to express geographic information in the vector map. First, we classified 3 types of bins and allocated the number of bits for each bin using adjacencies among the objects. and then about each classified bin, energy compaction and or pre-defined VLC(variable length coding) were performed according to characteristics of classified bins. Finally, for same target map, while a vector simplification algorithm had about 13%, compression ratio in 1m resolution we confirmed our method having more than 80% encoding efficiencies about original vector map in the 1cm resolution. Also it has not only higher compression ratio but also faster computing speed than present SEC based compression algorithm through experimental results. Moreover, our algorithm presented much more high performances about accuracy and computing power than vector approximation algorithm on same data volume sizes.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.37 no.6
• /
• pp.405-416
• /
• 2019

Due to the development of information and communication technology, the production and processing speed of data is getting faster. To classify objects using machine learning, which is a field of artificial intelligence, data required for training can be easily collected due to the development of internet and geospatial information technology. In the field of geospatial information, machine learning is also being applied to classify or recognize objects using images and point clouds. In this study, the problem of manually constructing training data using existing digital map version 1.0 was improved, and the technique of classifying roads, buildings and vegetation using image and point clouds were proposed. Through experiments, it was possible to classify roads, buildings, and vegetation that could clearly distinguish colors when using true ortho-image with only RGB (Red, Green, Blue) bands. However, if the colors of the objects to be classified are similar, it was possible to identify the limitations of poor classification of the objects. To improve the limitations, random forest and support vector machine techniques were applied after band fusion of true ortho-image and normalized digital surface model, and roads, buildings, and vegetation were classified with more than 85% accuracy.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.13 no.3
• /
• pp.661-668
• /
• 2018

Biomedical signals using skin resistance have different characteristics according to stress diseases. Biological diagnostic devices for diagnosing stress diseases have been developed by using these characteristics, and devices have been developed so that the signals measured by the skin storage meter can be easily analyzed. Experts in the field will look directly at the output signal to determine the likelihood of any stress disorder. However, it is very difficult for a person to accurately determine whether a person to be measured has a stress disorder by analyzing a bio-signal measured by each person to be measured, and the result of the judgment is very likely to be wrong. In order to solve these problems, we implemented the function of determining the signal of a stress disorder by using the machine learning technique. SVM was used as a classification method in consideration of low computing ability of measurement equipment. Training data and test data were randomly generated for each disease using error range 5 based on 13 diseases. Simulation results showed more than 90% decision accuracy. In the future, if the measurement equipment is actually applied to the patients, we can retrain the classifier with the newly generated data.

• Journal of Intelligence and Information Systems
• /
• v.9 no.1
• /
• pp.227-249
• /
• 2003

Prediction of corporate failure using past financial data is a well-documented topic. Early studies of bankruptcy prediction used statistical techniques such as multiple discriminant analysis, logit and probit. Recently, however, numerous studies have demonstrated that artificial intelligence such as neural networks can be an alternative methodology for classification problems to which traditional statistical methods have long been applied. In building neural network model, the selection of independent and dependent variables should be approached with great care and should be treated as model construction process. Irrespective of the efficiency of a teaming procedure in terms of convergence, generalization and stability, the ultimate performance of the estimator will depend on the relevance of the selected input variables and the quality of the data used. Approaches developed in statistical methods such as correlation analysis and stepwise selection method are often very useful. These methods, however, may not be the optimal ones for the development of neural network model. In this paper, we propose a genetic algorithms approach to find an optimal or near optimal input variables fur neural network modeling. The proposed approach is demonstrated by applications to bankruptcy prediction modeling. Our experimental results show that this approach increases overall classification accuracy rate significantly.

• Korean Journal of Oriental Medicine
• /
• v.13 no.2 s.20
• /
• pp.47-52
• /
• 2007

In Oriental medicine, the status of tongue is the important indicator to diagnose one's health, because it represents physiological and clinicopathological changes of inner parts of the body. The method of tongue diagnosis is not only convenient but also non-invasive, therefore, tongue diagnosis is one of the most widely used in Oriental medicine. But tongue diagnosis is affected by examination circumstances a lot. It depends on a light source, degrees of an angle, doctor's condition and so on. So it is not easy to make an objective and standardized tongue diagnosis. As part of way to solve this problem, in this study, we tried to design a discriminant function for white and yellow coating with multi-dimensional color vectors. There were 62 subjects involved in this study, among them 48 subjects diagnosed as white-coated tongue and 14 subjects diagnosed as yellow-coated tongue by oriental doctors. And their tongue images were acquired by a well-made Digital Tongue Diagnosis System. From those acquired tongue images, each coating section were extracted by oriental doctors, and then mean values of multi -dimensional color vectors in each coating section were calculated. By statistical analysis, two significant vectors, R in RGB space and H in HSV space, were found that they were able to describe the difference between white coating section and yellow coating section very well. Using these two values, we designed the discriminant function for coating classification and examined how good it works. As a result, the overall accuracy of coating classification was 98.4%. We can expect that the discriminant function for other coatings can be obtained in a similar way. Furthermore, if an automated segmentation algorithm of tongue coating is combined with these discriminant functions, an automated tongue coating diagnosis can be accomplished.

• Journal of Broadcast Engineering
• /
• v.23 no.3
• /
• pp.351-360
• /
• 2018

Human emotion recognition is a research topic that is receiving continuous attention in computer vision and artificial intelligence domains. This paper proposes a method for classifying human emotions through multiple neural networks based on multi-modal signals which consist of image, landmark, and audio in a wild environment. The proposed method has the following features. First, the learning performance of the image-based network is greatly improved by employing both multi-task learning and semi-supervised learning using the spatio-temporal characteristic of videos. Second, a model for converting 1-dimensional (1D) landmark information of face into two-dimensional (2D) images, is newly proposed, and a CNN-LSTM network based on the model is proposed for better emotion recognition. Third, based on an observation that audio signals are often very effective for specific emotions, we propose an audio deep learning mechanism robust to the specific emotions. Finally, so-called emotion adaptive fusion is applied to enable synergy of multiple networks. The proposed network improves emotion classification performance by appropriately integrating existing supervised learning and semi-supervised learning networks. In the fifth attempt on the given test set in the EmotiW2017 challenge, the proposed method achieved a classification accuracy of 57.12%.

• Journal of Conservation Science
• /
• v.36 no.2
• /
• pp.73-81
• /
• 2020

Considering the difficulties associated with the creation of deterioration maps for stone cultural heritages, quantitative determination of chemical and biological contaminants in them is still challenging. Hyperspectral image analysis has been proposed to overcome this drawback. In this study, hyperspectral imaging was performed on Stone Buddhas Temple in Four Directions at Gulbulsa Temple Site(Treasure 121), and several surface contaminants were observed. Based on the color and shape, these chemical and biological contaminants were classified into ten categories. Additionally, a method for establishing each class as a reference image was suggested. Simultaneously, with the help of Spectral Angle Mapper algorithm, two classification methods were used to classify the surface contaminants. Method A focused on the region of interest, while method B involved the application of the spectral library prepared from the image. Comparison of the classified images with the reference image revealed that the accuracies and kappa coefficients of methods A and B were 52.07% and 63.61%, and 0.43 and 0.55, respectively. Additionally, misclassified pixels were distributed in the same contamination series.

• Journal of Nuclear Fuel Cycle and Waste Technology(JNFCWT)
• /
• v.13 no.1
• /
• pp.21-34
• /
• 2015

Source term analysis should be carried out to prepare the decommissioning of the nuclear power plant. In the planning phase of decommissioning, the classification of decommissioning wastes and the cost evaluation are performed based on the results of source term analysis. In this study, the verification of MCNP/ORIGEN-2 model is carried out for preliminary source term calculation for Wolsung Unit 1. The inventories of actinide nuclides and fission products in fuel bundles with different burn-up were obtained by the depletion calculation of MCNPX code modelling the single channel. Two factors affecting the accuracy of source terms were investigated. First, the neutron spectrum effect on neutron induced activation calculation was reflected in one-group microscopic cross-sections of relevant radio-isotopes using the results of MCNP simulation, and the activation source terms calculated by ORIGEN-2 using the neutron spectrum corrected library were compared with the results of the original ORIGEN-2 library (CANDUNAU.LIB) in ORIGEN-2 code package. Second, operation history effect on activation calculation was also investigated. The source terms on both pressure tubes and calandria tubes replaced in 2010 and calandria tank were evaluated using MCNP/ORIGEN-2 with the neutron spectrum corrected library if the decommissioning wastes can be classified as a low level waste.

• Journal of Intelligence and Information Systems
• /
• v.24 no.3
• /
• pp.45-66
• /
• 2018

Crowdfunding has become more popular than angel funding for fundraising by venture companies. Identification of success factors may be useful for fundraisers and investors to make decisions related to crowdfunding projects and predict a priori whether they will be successful or not. Recent studies have suggested several numeric factors, such as project goals and the number of associated SNS, studying how these affect the success of crowdfunding campaigns. However, prediction of the success of crowdfunding campaigns via non-numeric and unstructured data is not yet possible, especially through analysis of structural characteristics of documents introducing projects in need of funding. Analysis of these documents is promising because they are open and inexpensive to obtain. We propose a novel method to predict the success of a crowdfunding project based on the introductory text. To test the performance of the proposed method, in our study, texts related to 1,980 actual crowdfunding projects were collected and empirically analyzed. From the text data set, the following details about the projects were collected: category, number of replies, funding goal, fundraising method, reward, number of SNS followers, number of images and videos, and miscellaneous numeric data. These factors were identified as significant input features to be used in classification algorithms. The results suggest that the proposed method outperforms other recently proposed, non-text-based methods in terms of accuracy, F-score, and elapsed time.