• Proceedings of the Acoustical Society of Korea Conference
• /
• 1993.06a
• /
• pp.233-237
• /
• 1993

본 논문에서는 패턴 매칭 방법에 근거하여 인식 단위가 음소인 음소 기반 인식 시스템을 구성하였다. 선택한 신경망 구조는 생물학적 신경망인 코호넨(T. Kohonen)의 SOFM(Self-Organizing Feature Map)으로 패턴 매칭 과정 중 cluster로 사용하였다. SOFM 신경망은 신호 공간에 대해서 최적의 국소(局所) 해부적 사사에 의한 자기 조직화 과정을 수행하며, 그 결과 인식 문제에 있어서 상당히 높은 정확도를 나타낸다. 따라서 SOFM 신경망은 음소 인식에도 효과적으로 응용될 수 있다. 또한 음소 인식 시스템의 성능 향상을 위해 K-means 클러스터링 알고리즘이 결합된 학습 알고리즘을 제안하였다. 제안된 음소 인식 시스템의 성능을 평가하기 위해, 먼저, 우리말 음소들을 모음, 파열음, 마찰음, 파찰음, 유음 및 비음, 종성의 6개 음소군으로 분류하고 각 음소군에 대한 특징 지도를 구성하여 labeler의 기능을 수행하게 하였다. 화자 종속 인식실험 결과 87.2%의 인식률을 보였으며 제안한 학습법의 빠른 수렴성과 인식률 향상을 확인하였다.

• Proceedings of the IEEK Conference
• /
• 2003.07d
• /
• pp.1263-1266
• /
• 2003

One of the hardest problems in implementation of digital neural network are extension of synapses and programmability for relocating neurons. This paper Proposes a new hardware structure to solve these problems. The proposed structure can reconfigure network connections without alteration of basic design, and extend number of synapses attached to one neuron. Also, it is possible to extend the number of neurons and layers by connecting many MPUs(Modular Processing Unit). Generality and extensibility are verified by composing various kinds of Perceptorn and Kohonen networks using the architecture proposed in this paper and the verification performances compares well with HDL simulation results as well as the results of C modelling.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1998.06a
• /
• pp.134-138
• /
• 1998

Unification of Kohone SOM(Self-Organizing Maps) neural network with the branch-and-bound algorithm is presented for clustering large set of patterns. The branch-and-bound search technique is employed for designing coarse neural network learning paradaim. Those unification can be use for clustering or calssfication of large patterns. For classfication purposes further usefulness is possible, since only two clusters exists in the SOM neural network of each nodes. The result of experiments show the fast learning time, the fast recognition time and the compactness of clustering.

• Journal of Mechanical Science and Technology
• /
• v.20 no.12
• /
• pp.2013-2024
• /
• 2006

This paper proposes a condition classification system using wavelet transform, feature evaluation and artificial neural networks to detect faulty products on the production line of reciprocating compressors for refrigerators. The stationary features of vibration signals are extracted from statistical cumulants of the discrete wavelet coefficients and root mean square values of band-pass frequencies. The neural networks are trained by the sample data, including healthy or faulty compressors. Based on training, the proposed system can be used on the automatic mass production line to classify product quality instead of people inspection. The validity of this system is demonstrated by the on-site test at LG Electronics, Inc. for reciprocating compressors. According to different products, this system after some modification may be useful to increase productivity in different types of production lines.

• Korean Management Science Review
• /
• v.21 no.3
• /
• pp.55-69
• /
• 2004

This study analyzes customer buying-behavior patterns in a department store as time goes on, and predicts moving patterns of its customers. Through them, it suggests in this paper short-term and long-term marketing promotion strategies. RFM techniques are utilized for customer segmentation. Customers are clustered by using the Kohonen's Self Organizing Map as a method of data mining techniques. Then C5.0, a decision tree analysis technique, is used to predict moving patterns of customers. Using real world data, this study evaluates the prediction accuracy of predictive models.

• Journal of Korea Multimedia Society
• /
• v.10 no.6
• /
• pp.699-706
• /
• 2007

This article is an approach to improving the clustering of Web images by using high-level semantic features from text information relevant to Web images as well as low-level visual features of image itself. These high-level text features can be obtained from image URLs and file names, page titles, hyperlinks, and surrounding text. As a clustering algorithm, a self-organizing map (SOM) proposed by Kohonen is used. To evaluate the clustering efficiencies of SOMs, we propose a simple but effective measure indicating the accumulativeness of same class images and the perplexities of class distributions. Our approach is to advance the existing measures through defining and using new measures accumulativeness on the most superior clustering node and concentricity to evaluate clustering efficiencies of SOMs. The experimental results show that the high-level text features are more useful in SOM-based Web image clustering.

• Proceedings of the Korea Multimedia Society Conference
• /
• 1998.04a
• /
• pp.394-399
• /
• 1998

본 논문에서는 Kohonen SOM을 이용한 인식 학습 알고리즘인 LVQ를 이용하여 퍼지 규칙의 수를 줄이는 방안을 제안하였다. 많은 훈련 패턴을 입력하게 되면 그에 따른 퍼지 규칙 수가 증가하게 되고, 많은 기억용량과 분류에 긴 시간을 필요로 하는 문제점 있어 퍼지 규칙의 수를 줄이고자 한다. 그러나 퍼지 규칙의 수가 줄어듦으로서 발생하는 성능의 하락을 최소화하기 위하여 초기 참조 패턴이 입력 데이터에 근접하도록 훈련 된 후에 퍼지 규칙을 생성하였다. 생성된 퍼지 규칙은 LVQ를 이용하여 인식되기 바로 전에 가중치 벡터를 이용하여 근접하는 값 이내에 있는 가중치 벡터 값을 합하여 같은 퍼지 규칙을 부여하여 생성하였다. 그 결과로 5$\times$8 숫자 Gray scale를 이용하여 전체 146개의 가중치 벡터가 15개의 아주 적은 수의 퍼지 규칙으로 생성되었다.

• The Transactions of the Korean Institute of Electrical Engineers A
• /
• v.49 no.1
• /
• pp.31-36
• /
• 2000

To enhance the fault diagnosis ability for the dissolved gas analysis(DGA) of the power transformer, this paper proposes a learning vector quantization(LVQ) for the incipient fault recognition. LVQ is suitable expecially for pattern recognition such as fault diagnosis of power transformer using DGA because it improves the performance of Kohonen neural network by placing emphasis on the classification around the decision boundary. The capabilities of the proposed diagnosis system for the transformer DGA decision support have been extensively verified through the practical test data collected from Korea Electrical Power Corporation.

• Annual Conference of KIPS
• /
• 2003.11a
• /
• pp.291-294
• /
• 2003

많은 양의 유즈넷 뉴스 중에서 찾고자 하는 정확한 정보를 빠른 시간 안에 검색하고, 원하는 정보만 필터링 하는 것은 중요하다. 하지만 뉴스 문서는 이메일과 달라서 미리 자신에게 맞는 뉴스그룹을 등록해 주어야만 정보를 얻을 수 있다. 본 연구에서는 다양한 뉴스그룹들 중에서 사용자와 취향이 가장 유사한 뉴스그룹을 코호넨 신경망을 이용하여 분류하는 서비스를 제공한다. 신경망을 학습시키기 위한 뉴스 문서의 키워드들을 선택하기 위해 예제 문서들로부터 후보 용어들을 추출하고 퍼지 추론을 적용하여 대표 용어들을 선택한다. 뉴스 필터링 시스템의 분류 성능을 평가하기 위하여 유클리드 거리 면에서 비교한 결과, 제안한 방법의 유용성을 확인할 수 있었다.

• Transactions on Control, Automation and Systems Engineering
• /
• v.2 no.3
• /
• pp.182-188
• /
• 2000

High-frequency electric resistance welding (HERW) technique is one of the most productive manufacturing method currently available for pipe and tube production because of its high welding speed. In this process, a heat input is controlled by skilled operators observing color and shape of bead but such a manual control can not provide reliability and stability required for manufacturing pipes of high grade quality because of a variety of bead shapes and noisy environment. In this paper, in an effort to provide reliable quality inspection, we propose a neural network-based method for classification of bead shape. The proposed method utilizes the structure of Kohonen network and is designed to learn the skill of the expert operators and to provide a good solution to classify bead shapes according to their welding conditions. This proposed method is implemented on the real pipe manufacturing process, and a series of experiments are performed to show its effectiveness.