• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.3
• /
• pp.289-293
• /
• 2005

Pattern classification extracts various types of pattern information expressing objects in the real world and decides their class. The top priority of pattern classification technologies is to improve the performance of classification and, for this, many researches have tried various approaches for the last 40 years. Classification methods used in pattern classification include base classifier based on the probabilistic inference of patterns, decision tree, method based on distance function, neural network and clustering but they are not efficient in analyzing a large amount of multi-dimensional data. Thus, there are active researches on multiple classifier systems, which improve the performance of classification by combining problems using a number of mutually compensatory classifiers. The present study identifies problems in previous researches on multiple SVM classifiers, and proposes BORSE, a model that, based on 1:M policy in order to expand SVM to a multiple class classifier, regards each SVM output as a signal with non-linear pattern, trains the neural network for the pattern and combine the final results of classification performance.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.25 no.3
• /
• pp.267-275
• /
• 2012

In this study, crack growth in a center-cracked plate is predicted under mode I variable amplitude loading, and the result is validated by experiment. Huang's model is employed to describe crack growth with acceleration and retardation due to the variable loading effect. Experiment is conducted with Al6016-T6 plate, in which the load is applied, and crack length is measured periodically. Particle Filter algorithm, which is based on the Bayesian approach, is used to estimate model parameters from the experimental data, and predict the crack growth of the future in the probabilistic way. The prediction is validated by the run-to-failure results, from which it is observed that the method predicts well the unique behavior of crack retardation and the more data are used, the closer prediction we get to the actual run-to-failure data.

• Journal of KIISE
• /
• v.43 no.7
• /
• pp.795-800
• /
• 2016

In image classification, recent CNNs compete with human performance. However, there are limitations in more general recognition. Herein we deal with indoor images that contain too much information to be directly processed and require information reduction before recognition. To reduce the amount of data processing, typically variational inference or variational Bayesian methods are suggested for object detection. However, these methods suffer from the difficulty of marginalizing over the given space. In this study, we propose an image-text integrated recognition system using active vision based on Spatial Transformer Networks. The system attempts to efficiently sample a partial region of a given image for a given language information. Our experimental results demonstrate a significant improvement over traditional approaches. We also discuss the results of qualitative analysis of sampled images, model characteristics, and its limitations.

• Journal of the Society of Disaster Information
• /
• v.17 no.3
• /
• pp.427-441
• /
• 2021

Purpose: This study proposes a method for developing a model that predicts the probability of traffic accidents in advance to prevent the most frequent traffic accidents in the military. Method: For this purpose, CRISP-DM (Cross Industry Standard Process for Data Mining) was applied in this study. The CRISP-DM process consists of 6 stages, and each stage is not unidirectional like the Waterfall Model, but improves the level of completeness through feedback between stages. Results: As a result of modeling the same data set as the previously constructed accident investigation data for the entire group, when the classification criterion was 0.5, Significant results were derived from the accuracy, specificity, sensitivity, and AUC of the model for predicting traffic accidents. Conclusion: In the process of designing the prediction model, it was confirmed that it was difficult to obtain a meaningful prediction value due to the lack of data. The methodology for designing a predictive model using the data set was proposed by reorganizing and expanding a data set capable of rational inference to solve the data shortage.

• KIPS Transactions on Software and Data Engineering
• /
• v.8 no.7
• /
• pp.275-288
• /
• 2019

In this paper, we consider a rumor source inference problem when sufficiently many nodes heard the rumor in the network. This is an important problem because information spread in networks is fast in many real-world phenomena such as diffusion of a new technology, computer virus/spam infection in the internet, and tweeting and retweeting of popular topics and some of this information is harmful to other nodes. This problem has been much studied, where it has been shown that the detection probability cannot be beyond 31% even for regular trees if the number of infected nodes is sufficiently large. Motivated by this, we study the impact of query that is asking some additional question to the candidate nodes of the source and propose budget assignment algorithms of a query when the network administrator has a finite budget. We perform various simulations for the proposed method and obtain the detection probability that outperforms to the existing prior works.

• The Journal of Korean Institute of Information Technology
• /
• v.17 no.6
• /
• pp.9-16
• /
• 2019

Companies that want to improve their manufacturing processes have recently introduced the smart factory, which is particularly noticeable. The ultimate goal is to maximize the area of the smart factory that performs the process of the production facility completely with minimal manual control and to minimize errors of reasoning. This research is a part of a project for unmanned production, management, packaging, and delivery management and the detection of the start point of rebars to perform the automatic calibration of the rollers through the tracking of the automated facilities of unmanned production. It must meet the requirement to accurately track the position from the start point to the end point. In order to improve the tracking performance, it is important to set the accurate start point. However, the probability of tracking errors is high depending on environments such as illumination and dust through the conventional time-based detection method. In this paper, we propose a starting point detection method using the average brightness change of high speed IR camera to reduce the errors according to the environments, As a result, its performance is improved by more than 15%.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.6
• /
• pp.381-392
• /
• 2021

In this study, we perform deep learning-based object detection analysis on eight types of buildings defined by the digital map topography standard code, leveraging images taken with UAV (Unmanned Aerial Vehicle). Image labeling was done for 509 images taken by UAVs and the YOLO (You Only Look Once) v5 model was applied to proceed with learning and inference. For experiments and analysis, data were analyzed by applying an open source-based analysis platform and algorithm, and as a result of the analysis, building objects were detected with a prediction probability of 88% to 98%. In addition, the learning method and model construction method necessary for the high accuracy of building object detection in the process of constructing and repetitive learning of training data were analyzed, and a method of applying the learned model to other images was sought. Through this study, a model in which high-efficiency deep neural networks and spatial information data are fused will be proposed, and the fusion of spatial information data and deep learning technology will provide a lot of help in improving the efficiency, analysis and prediction of spatial information data construction in the future.

• Korean Journal of Culture and Social Issue
• /
• v.29 no.4
• /
• pp.537-567
• /
• 2023

This study compared the nature of disgust caused by the crime scene with that by the stereotype of the sexual-minority defendant, and compared the effect of each type of disgust on evidence evaluation and legal judgment. A total of 600 participants (300 men, average age of 44.40) were randomly assigned to sources of disgust (crime scene, sexual minorities defendant, control condition), the existence of additional evidence of innocence (o/x), and the existence of judicial directives (o/x). As a result of the study, disgust under the condition of a cruel crime scene with strong physical disgust was significantly higher than that of the sexual minority defendant, interpreted the evidence in a more guilty direction, and was more prone to_evaluate that the defendant was guilty. It is noteworthy that evidence evaluation was a significant moderating variable between disgust and probability of guilt under conditions where the source of disgust was a sexual minority, but not under control conditions and crime scene condition. It means that the effect of disgust on legal judgment may not be direct when the defendant is a sexual minority. In addition, the existence of the judicial instruction had a significant inverse effect on the sentence. And simple effect analysis found that presenting judicial instruction lowered probability of guilt only under the control condition. This makes it reasonable to infer that disgust derived from the characteristics of the crime scene and the defendant can be recognized as integral emotions that are difficult to correct with instructions. Finally, pity for the defendant was significantly higher under the conditions of sexual minority which shows that an emotional response of sympathy may occur in addition to disgust for sexual minorities. After examining the nature of disgust (physical & moral), legal judgment according to the source and degree of disgust was reviewed. In addition, the meaning of disgust and sympathy for the sexual minority defendant was discussed.

• Journal of Korean Society of Transportation
• /
• v.26 no.3
• /
• pp.169-178
• /
• 2008

Level of Service (LOS) is one of ways to evaluate operational conditions. It is very important factor in evaluation especially for the facility of highways. However, some studies proved that ${\upsilon}/c$ ratio and accident rate is appeared like a second function which has a U-form. It means there is a gap between LOS and safety of highway facilities. Therefore, this study presents a method for evaluation of a signalized intersection which is considered both smooth traffic operation (delay) and traffic safety (accident). Firstly, as a result of our research, accident rates and EPDO are decreased when it has a big delay. In that reason, it is necessary to make a new Level of Service included traffic safety. Secondly, this study has developed a negative binominal regression model which is based on the relation between accident patterns and stream. Thirdly, standards of LOS are presented which is originated from calculation between annual delay costs and annual accident cost at each intersection. Lastly, worksheet form is presented as an expression to an estimation step of a signalized intersection with traffic accident prediction model and new LOS.

• Journal of Broadcast Engineering
• /
• v.17 no.2
• /
• pp.270-283
• /
• 2012

With the advent of multi-channel TV, IPTV and smart TV services, excessive amounts of TV program contents become available at users' sides, which makes it very difficult for TV viewers to easily find and consume their preferred TV programs. Therefore, the service of automatic TV recommendation is an important issue for TV users for future intelligent TV services, which allows to improve access to their preferred TV contents. In this paper, we present a recommendation model based on statistical machine learning using a collaborative filtering concept by taking in account both public and personal preferences on TV program contents. For this, users' preference on TV programs is modeled as a latent topic variable using LDA (Latent Dirichlet Allocation) which is recently applied in various application domains. To apply LDA for TV recommendation appropriately, TV viewers's interested topics is regarded as latent topics in LDA, and asymmetric Dirichlet distribution is applied on the LDA which can reveal the diversity of the TV viewers' interests on topics based on the analysis of the real TV usage history data. The experimental results show that the proposed LDA based TV recommendation method yields average 66.5% with top 5 ranked TV programs in weekly recommendation, average 77.9% precision in bimonthly recommendation with top 5 ranked TV programs for the TV usage history data of similar taste user groups.