• 한국컴퓨터정보학회논문지
• /
• 제10권5호
• /
• pp.353-362
• /
• 2005

최근 들어 바이모달 인식에 관한 연구가 활발히 진행되고 있다. 본 논문에서는 음성 정보와 얼굴정보를 이용하여 바이모달 시스템을 구현하였다. 얼굴인식은 얼굴 검출과 얼굴 인식 두 부분으로 나누어서 실험을 하였다. 얼굴 검출 단계에서는 AdaBoost를 이용하여 얼굴 후보 영역을 검출 한 뒤 PCA를 통해 특징 벡터 계수를 줄였다. PCA를 통해 추출된 특징 벡터를 객체 분류 기법인 SVM을 이용하여 얼굴을 검출 및 인식하였다. 음성인식은 MFCC를 이용하여 음성 특징 추출을 하였으며 HMM을 이용하여 음성인식을 하였다. 인식결과, 단일 인식을 사용하는 것보다 얼굴과 음성을 같이 사용하였을 때 인식률의 향상을 가져왔고, 잡음 환경에서는 더욱 높은 성능을 나타냈었다.

• 대한전자공학회논문지SP
• /
• 제43권3호
• /
• pp.61-67
• /
• 2006

ES 시스템에서 레이더 신호의 PRI 변조특성 분석은 고밀도 전자전 신호 환경에서의 레이더 식별 모호성 문제로 인해 그 중요성이 증가되고 있다. 본 논문에서는 ES를 위한 레이더 펄스 신호의 PRI 변조형태를 인식하는 새로운 기법을 제안한다. 제안한 기법은 PRI 시퀀스의 선형 자기상관관계에 나타나는 각 변조형태별 특징을 기반으로 정의한 형태 구분자들을 이용하여 PRI 변조형태를 인식한다. 또한 제안된 기법은 주기적인 변조특성을 갖는 PRI 변조형태에 대해서 변조주기를 추정한다. 제안한 기법의 성능을 입증하기 위해 다양한 모의신호에 대한 실험결과를 제시하였다.

• 한국정보통신학회논문지
• /
• 제14권1호
• /
• pp.63-69
• /
• 2010

최근 기계학습 분야에서 분류자의 정확도 개선을 위하여 제안된 다양한 방법들 중 가장 큰 주목을 받고 있는 학습방법 중 하나는 앙상블 학습이다. 그러나 앙상블 학습은 의사결정트리와 같이 불안정한 학습 알고리즘의 성과 개선 효과는 탁월한 반면, 인공신경망과 같이 안정적인 학습알고리즘의 성과 개선 효과는 응용 분야와 구현 방법에 따라 서로 상반된 결론들을 보여주고 있다. 본 연구에서는 국내 기업의 부실화 예측문제를 활용하여 인공신경 망 분류자 및 대표적 앙상블 학습기법인 부스팅 분류자를 적용한 결과 앙상블 학습은 기업부실 예측문제에 있어 전통적 인공신경망의 성과를 개선할 수 있음을 검증하였다.

• IEIE Transactions on Smart Processing and Computing
• /
• 제6권4호
• /
• pp.281-291
• /
• 2017

Nowadays, as traffic jams are a daily elementary problem in both developed and developing countries, systems to monitor, predict, and detect traffic conditions are playing an important role in research fields. Comparing them, researchers have been trying to solve problems by applying many kinds of technologies, especially roadside sensors, which still have some issues, and for that reason, any one particular method by itself could not generate sufficient traffic prediction results. However, these sensors have some issues that are not useful for research. Therefore, it may not be best to use them as stand-alone methods for a traffic prediction system. On that note, this paper mainly focuses on predicting traffic conditions based on a hybrid prediction approach, which stands on accuracy comparison of three prediction models: multinomial logistic regression, decision trees, and support vector machine (SVM) classifiers. This is aimed at selecting the most suitable approach by means of integrating proficiencies from these approaches. It was also experimentally confirmed, with test cases and simulations that showed the performance of this hybrid method is more effective than individual methods.

• 제어로봇시스템학회논문지
• /
• 제16권8호
• /
• pp.761-765
• /
• 2010

This paper presents the design of optimal fuzzy classifier for human detection by using genetic algorithms, one of the best-known meta-heuristic search methods. For this purpose, encoding scheme to search the optimal sequential intersection points between adjacent fuzzy membership functions is originally presented for the fuzzy classifier design for HOG (Histograms of Oriented Gradient) descriptors. The intersection points are sequentially encoded in the proposed encoding scheme to reduce the redundancy of search space occurred in the combinational problem. Furthermore, the fitness function is modified with the true-positive and true-negative of the confusion matrix instead of the total success rate. Experimental results show that the two proposed approaches give superior performance in HOG datasets.

• 한국생산제조학회지
• /
• 제26권3호
• /
• pp.278-284
• /
• 2017

Generally, classifiers have been used as machines to crush raw materials and classify suitable particle sizes in all industrial fields, such as food, chemical, and mineral. However, the technique for classifying micron-sized particles between 5 and $20{\mu}m$ is inferior. In particular, numerous experiments and considerable experiences are required to predict the particle size, because the classifier particle size is determined according to the internal flow. However, it is quite difficult to set the driving conditions so that the desired particle size can be classified only by experience and experimentation. Therefore, this study proposes a method of predicting the average particle size by employing multiphase flow analysis and the Taguchi method; this method is subsequently verified.

• 한국CDE학회논문집
• /
• 제5권2호
• /
• pp.127-135
• /
• 2000

Emergent computing paradigms such as genetic algorithms have found increased use in problems in engineering design. These computational tools have been shown to be applicable in the solution of generically difficult design optimization problems characterized by nonconvexities in the design space and the presence of discrete and integer design variables. Another aspect of these computational paradigms that have been lumped under the bread subject category of soft computing, is the domain of artificial intelligence, knowledge-based expert system, and machine learning. The paper explores a machine learning paradigm referred to as teaming classifier systems to construct the high-quality global function approximations between the design variables and a response function for subsequent use in design optimization. A classifier system is a machine teaming system which learns syntactically simple string rules, called classifiers for guiding the system's performance in an arbitrary environment. The capability of a learning classifier system facilitates the adaptive selection of the optimal number of training data according to the noise and multimodality in the design space of interest. The present study used the polynomial based response surface as global function approximation tools and showed its effectiveness in the improvement on the approximation performance.

• 전자공학회논문지CI
• /
• 제43권1호
• /
• pp.9-16
• /
• 2006

RDA(Relational Discriminant Analysis)는 패턴의 특징벡터 대신에 학습 패턴을 대표하는 프로토타입들과의 비유사도 벡터에 기반하여 식별기를 설계하는 방법이다. 따라서 RDA 식별기의 성능은 프로토타입을 선택하는 방법과 비유사도를 측정하는 방법에 따라 결정된다. 본 논문에서는 PRS(Prototype Reduction Schemes)를 이용하여 프로토타입을 추출한 다음, 샘플 벡터들간의 마할라노비스 거리에 의한 상관행렬로 RDA의 식별성능을 향상시키는 방법을 제안한다. 인공 데이터 및 실-생활 데이터를 대상으로 실험한 결과, 제안한 방법의 식별성능이 기존의 방법에 비하여 개선되었음을 확인하였다.

• 사상체질의학회지
• /
• 제21권3호
• /
• pp.39-47
• /
• 2009

1. Objectives: In order to classify an individual into four constitution type, an oriental medical doctor utilizes various information such as face, pulse, voice, and questionnaire. When only one type of information is used, one's constitution may not be decided correctly. 2. Methods: In this paper, we propose a novel four constitution types classifier using facial images which classifies subjects into indecision group as well as Taeumin, Soeumin, and Soyangin. 3. Results: Experimental results show that it increases the classification rate though the decision rate is rather decreased, which is more effective and reliable than conventional classifiers without indecision. 4. Conclusion: For the effective classification, we have found that it is more useful to add an indecision group which requires more information to be properly classified into one constitution type.

• ETRI Journal
• /
• 제34권5호
• /
• pp.743-752
• /
• 2012

Semantic interpretation of the relationship between noun compound (NC) elements has been a challenging issue due to the lack of contextual information, the unbounded number of combinations, and the absence of a universally accepted system for the categorization. The current models require a huge corpus of data to extract contextual information, which limits their usage in many situations. In this paper, a new semantic relations interpreter for NCs based on novel lightweight binary features is proposed. Some of the binary features used are novel. In addition, the interpreter uses a new feature selection method. By developing these new features and techniques, the proposed method removes the need for any huge corpuses. Implementing this method using a modular and plugin-based framework, and by training it using the largest and the most current fine-grained data set, shows that the accuracy is better than that of previously reported upon methods that utilize large corpuses. This improvement in accuracy and the provision of superior efficiency is achieved not only by improving the old features with such techniques as semantic scattering and sense collocation, but also by using various novel features and classifier max entropy. That the accuracy of the max entropy classifier is higher compared to that of other classifiers, such as a support vector machine, a Na$\ddot{i}$ve Bayes, and a decision tree, is also shown.