• Annual Conference on Human and Language Technology
• /
• 2012.10a
• /
• pp.181-184
• /
• 2012

본 논문에서는 한국전자통신연구원 언어처리연구팀에서 개발하고 있는 차량용 대화 시스템을 위한 Dialog Act 태깅 코퍼스 구축 방법에 대해 기술하는 것을 목표로 한다. 차량용 태깅 코퍼스 구축 방법은 크게 차량용 대화 코퍼스 수집과 수집된 대화 코퍼스에 Dialog Act를 반자동으로 태깅하는 방법으로 나눌 수 있다. 차량용 대화 코퍼스 수집은 1) 대화플랜 맵 구축, 2) 표준대화 구축, 3) 자유대화 구축, 4) 사용자 발화에 패러프래징 발화 구축의 순으로 구축되었다. Dialog Act 태깅은 수집된 대화코퍼스로부터 슬롯 후보를 추출하여 슬롯 체계를 구축한 후 반자동 슬롯 태깅을 실시하고, 슬롯 태깅 결과와 Dialog Act Type을 조합하여 Dialog Act 태깅 코퍼스를 구축하였다. 이렇게 구축된 Dialog Act 태깅 코퍼스는 차량 공조시스템(에어컨, 히터 등) 및 차량 응급 조치 정보 서비스와 같은 차량용 대화 시스템에 적용 중에 있다.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.91-94
• /
• 2018

감정분석은 텍스트에서 나타난 저자 혹은 발화자의 태도, 의견 등과 같은 주관적인 정보를 추출하는 기술이며, 여론 분석, 시장 동향 분석 등 다양한 분야에 두루 사용된다. 감정분석 방법은 사전 기반 방법, 기계학습 기반 방법 등이 있다. 본 논문은 사전 기반 감정분석에 필요한 한국어 감정사전 자동 구축 방법을 제안한다. 본 논문은 영어 감정사전으로부터 한국어 감정사전을 자동으로 구축하는 방법이며, 크게 세 단계로 구성된다. 첫 번째는 영한 병렬말뭉치를 이용한 영한사전을 구축하는 단계이고, 두 번째는 영한사전을 통한 이중언어 그래프를 생성하는 단계이며, 세 번째는 영어 단어의 감정값을 한국어 단어의 감정값으로 전파하는 단계이다. 본 논문에서는 제안된 방법의 유효성을 보이기 위해 사전 기반 한국어 감정분석 시스템을 구축하여 평가하였으며, 그 결과 제안된 방법이 합리적인 방법임을 확인할 수 있었으며 향후 연구를 통해 개선한다면 질 좋은 한국어 감정사전을 효과적인 방법으로 구축할 수 있을 것이다.

• Annual Conference of KIPS
• /
• 2004.05a
• /
• pp.873-876
• /
• 2004

사람의 동작을 인식하는 것에 대한 연구는 게임, 유비쿼터스 컴퓨팅 등의 발전에 따라 그 중요성이 증가하고 있다. 그러나, 대부분의 기존 연구에서는 극히 소수의 동작만을 정의하거나 특정 부위의 동작만을 다루므로 실제 응용에 적용하기에는 적합하지 않다. 본 논문에서는 특정 도메인의 사용 없이, 카메라 영상 입력으로 취득된 동작 패턴 정보만을 이용하여 40종 전신 연속 동작을 구분하는 동작인식 방법을 연구하였다. 인식에 사용된 입력 데이터는 동작자 관절들의 위치 및 회전 값들이며, 다수의 동작들을 인식하기 위해서는 기존의 인식 알고리즘들인 특징기반 인식, HMM, 신경망(Neural Network)등을 사용하여 복합적인 인식 엔진을 구성하여야 했다. 입력 데이터별로 적합한 인식 모듈을 거치게 하기 위해서는, 동작에 의한 입력 데이터에서 동작자 움직임의 주요 신체 부위를 추출함으로써 입력 데이터가 해당 그룹의 인식 모듈로 자동적으로 분류되게 하는 방법을 사용한다. 이는 다층의 인식 레이어 중 복잡도가 증가하는 하위 레이어일수록 자동 분류에 의해 걸러진 데이터만을 취급하게 되므로 효과적이다. 전체 실험 결과 단계별로 약 79~97%의 인식률을 보였다. 이는 향후 특정 컨텍스트 정보와 결합할 때 매우 높은 인식률을 기대할 수 있게 하는 수치이다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 1999.05a
• /
• pp.203-207
• /
• 1999

In this paper we propose the automated generation algorithm of penetration scenario using association mining technique. Until now known intrusion detections are classified into anomaly detection and misuse detection. The former uses statistical method, features selection, neural network method in order to decide intrusion, the latter uses conditional probability, expert system, state transition analysis, pattern matching for deciding intrusion. In proposed many intrusion detection algorithms unknown penetrations are created and updated by security experts. Our algorithm automatically generates penetration scenarios applying association mining technique to state transition technique. Association mining technique discovers efficient and useful unknown information in existing data. In this paper the algorithm we propose can automatically generate penetration scenarios to have been produced by security experts and is easy to cope with intrusions when it is compared to existing intrusion algorithms. Also It has advantage that maintenance cost is not high.

• Journal of the Korea Society for Simulation
• /
• v.17 no.4
• /
• pp.1-9
• /
• 2008

Proposed in the paper is an automated procedure to construct a plant model for PLC simulation. Since PLC programs only contain the control logic without the information on the plant model, it is necessary to build the corresponding plant model to perform simulation. Conventionally, a plant model for PLC simulation has been constructed manually, and it requires much efforts as well as the in-depth knowledge of simulation. As a remedy for the problem, we propose an automated procedure to generate a plant model from the symbol table of a PLC program. To do so, we propose a naming rule for PLC symbols so that the symbol names include enough information on the plant model. By analyzing such symbol names, we extract a plant model automatically. The proposed methodology has been implemented, and test runs were made.

• KIPS Transactions on Software and Data Engineering
• /
• v.6 no.5
• /
• pp.267-278
• /
• 2017

As the same letter can be pronounced differently depending on word contexts, one should refer to a lexicon in order to pronounce a word correctly. Phonetic alphabets that lexicons adopt as well as pronunciations that lexicons describe for the same word can be different from lexicon to lexicon. In this paper, we use a sequence-to-sequence model that is widely used in deep learning research area in order to convert automatically from one pronunciation to another. The 12 seq2seq models are implemented based on pronunciation training data collected from 4 different lexicons. The exact accuracy of the models ranges from 74.5% to 89.6%. The aim of this study is the following two things. One is to comprehend a property of phonetic alphabets and pronunciations used in various lexicons. The other is to understand characteristics of seq2seq models by analyzing an error.

• Journal of the Korea Society of Computer and Information
• /
• v.10 no.6 s.38
• /
• pp.65-74
• /
• 2005

This Paper Propose a robust scene change detection technique that use the weighted chi-square test and the automated threshold-decision algorithm. The weighted chi-test can subdivide the difference values of individual color channels by calculating the color intensities according to mSC standard, and it can detect the scene change by joining the weighted color intensities to the predefined chi-test which emphasize the comparative color difference values. The automated decision algorithm uses the difference values of frame-to-frame that was obtained by the weighted chi-test. In the first step, The average of total difference value and standard deviation value is calculated and then, subtract the mean value from the each difference values. In the next step, the same process is performed on the remained difference value. The propose method is tested on various sources and in the experimental results, it is shown that the Proposed method is efficiently estimates the thresholds and reliably detects scene changes.

• Journal of the Society of Disaster Information
• /
• v.17 no.3
• /
• pp.556-567
• /
• 2021

Purpose: In this study, various methods of deep learning-based automatic damage analysis technology were reviewed based on images taken through Unmanned Aerial Vehicle to more efficiently and reliably inspect the exterior inspection and inspection of railway bridges using Unmanned Aerial Vehicle. Method: A deep learning analysis model was created by defining damage items based on the acquired images and extracting deep learning data. In addition, the model that learned the damage images for cracks, concrete and paint scaling·spalling, leakage, and Reinforcement exposure among damage of railway bridges was applied and tested with the results of automatic damage analysis. Result: As a result of the analysis, a method with an average detection recall of 95% or more was confirmed. This analysis technology enables more objective and accurate damage detection compared to the existing visual inspection results. Conclusion: through the developed technology in this study, it is expected that it will be possible to analysis more accurate results, shorter time and reduce costs by using the automatic damage analysis technology using Unmanned Aerial Vehicle in railway maintenance.

• Journal of the Korea Society of Computer and Information
• /
• v.17 no.3
• /
• pp.51-57
• /
• 2012

Hepatic vessel tree is the key structure for hepatic disease diagnosis and liver surgery planning. Especially, it is used to evaluate the donors' and recipients' liver for the LDLT(Living Donors Liver Transplantation) and estimate the volumes of left and right hepatic lobes for securing their life in the LDLT. In this study, we propose a method to apply canny edge detection that is not affected by noise to the liver images for automatic segmentation of hepatic vessels tree in contrast abdominal MDCT image. Using histograms and average pixel values of the various liver CT images, optimized parameters of the Canny algorithm are determined. It is more time-efficient to use the common parameters than to change parameters manually according to CT images. Candidates of hepatic vessels are extracted by threshold filtering around the detected the vessel edge. Finally, using a system which detects the true-negatives and the false-positives in horizontal and vertical direction, the true-negatives are added in candidate of hepatic vessels and the false-positives are removed. As a result of the process, the various hepatic vessel trees of patients are accurately reconstructed in 3D.

• Journal of the Korean Geotechnical Society
• /
• v.37 no.11
• /
• pp.119-129
• /
• 2021

The process of constructing the DB in the current geotechnical information DB system needs a lot of human and time resource consumption. In addition, it causes accuracy problems frequently because the current input method is a person viewing the PDF and directly inputting the results. Therefore, this study proposes building an automatic digital DB using AI (artificial intelligence) of boring logs. In order to automatically construct DB for various boring log formats without exception, the boring log forms were classified using the deep learning model ResNet 34 for a total of 6 boring log forms. As a result, the overall accuracy was 99.7, and the ROC_AUC score was 1.0, which separated the boring log forms with very high performance. After that, the text in the PDF is automatically read using the robotic processing automation technique fine-tuned for each form. Furthermore, the general information, strata information, and standard penetration test information were extracted, separated, and saved in the same format provided by the geotechnical information DB system. Finally, the information in the boring log was automatically converted into a DB at a speed of 140 pages per second.