• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.6
• /
• pp.88-98
• /
• 1998

Subspace classifier is a popular pattern recognition method based on Karhunen-Loeve transform. This classifier describes a high dimensional pattern by using a reduced dimensional subspace. Because of the loss of information induced by dimensionality reduction, however, a subspace classifier sometimes shows unsatisfactory recognition performance to the patterns having quite similar principal components each other. In this paper, we propose the use of multiple novelty neural network classifiers constructed on novelty vectors to adopt minor components usually ignored and present a method of improving recognition performance through combining those with the subspace classifier. We develop the proposed classifier on handwritten numeral database and analyze its properties. Our proposed classifier shows better recognition performance compared with other classifiers, though it requires more weight links.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.1
• /
• pp.170-177
• /
• 1996

In this paper, we propose a neural network for extracting dashed lines from map images based on the feature extracting path be found in a cat and monkey's visual cortex. The proposed system converts a color image to a binary image and extracts the lines per orientation. Next, the system removes the lines longer than the dashed line and extracts the dashed lines by finding the consecutive short lines. The proposed system for extracting the dashed lines was tested using 1/25,000 scaled maps published by the National Geographic Institute of Korea. This paper shows that the proposed system can extract the dashed-lines from maps.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.38 no.2
• /
• pp.20-30
• /
• 2001

In the paper, we propose a predictive control scheme using multiple neural network-based prediction models. To construct the multiple models, we select several specific values of a parameter whose variation affects serious control performance in the plant. Among the multiple prediction models, we choose one that shows the best predictions for future outputs of the plant by a switching technique. Based on a nonlinear programming method, we calculate the current process input in the nonlinear predictive control system with multiple prediction models. The proposed control method is shown to be very effective when a parameter of the plant changes or the time delay, if it exists, varies. It is also shown that the proposed method is successfully applied for the control of suspension in a electro-magnetic levitation system.

• Proceedings of the KIEE Conference
• /
• 2006.04a
• /
• pp.318-320
• /
• 2006

Inaccurate localization exposes a robot to many dangerous conditions. It could make a robot be moved to wrong direction or damaged by collision with surrounding obstacles. There are numerous approaches to self-localization, and there are different modalities as well (vision, laser range finders, ultrasonic sonars). Since sensor information is generally uncertain and contains noise, there are many researches to reduce the noise. But, the correctness is limited because most researches are based on statistical approach. The goal of our research is to measure more exact robot location by matching between built VRML 3D model and real vision image. To determine the position of mobile robot, landmark-localitzation technique has been applied. Landmarks are any detectable structure in the physical environment. Some use vertical lines, others use specially designed markers, In this paper, specially designed markers are used as landmarks. Given known focal length and a single image of three landmarks it is possible to compute the angular separation between the lines of sight of the landmarks. The image-processing and neural network pattern matching techniques are employed to recognize landmarks placed in a robot working environment. After self-localization, the 2D scene of the vision is overlaid with the VRML scene.

• Proceedings of the KIEE Conference
• /
• 1995.11a
• /
• pp.175-177
• /
• 1995

The sensor signal validation and failure detection system must be able to detect, isolate, and identify sensor degradation as well as provide a reconstruction of the measurements. In this study, this is accomplished by combining the neural network, the Generalized Consistency Checking(GCC), and the Sequential Probability Ratio Test(SPRT) method in a decision estimator module. The GCC method is a computationally efficient system for redundant sensors, while the SPRT provides the ability to make decisions based on the degradation history of a sensor. The methodology is also extended to the detection of noise degradation. The acceptability of the proposed method is demonstration by using the simulation data in safety injection accident of nuclear power plants. The results show that the signal validation and sensor failure detection system is able to detect and isolate a bias failure and noise type failures under transient conditions. And also, the system is able to provide the validated signal by reconstructing the measurement signals in the failure conditions considered.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.64 no.7
• /
• pp.1064-1073
• /
• 2015

In this study, we introduce the pedestrian detection system by using the feature of HOG-PCA and RBFNNs pattern classifier. HOG(Histogram of Oriented Gradient) feature is extracted from input image to identify and recognize a object. And a dimension is reduced for improving performance as well as processing speed by using PCA which is a typical dimensional reduction algorithm. So, the feature of HOG-PCA through the dimensional reduction by using PCA leads to the improvement of the detection rate. FCM clustering algorithm is used instead of gaussian function to apply the characteristic of input data as well and connection weight is used by polynomial expression such as constant, linear, quadratic and modified quadratic. Finally, INRIA person database known as one of the benchmark dataset used for pedestrian detection is applied for the performance evaluation of the proposed classifier. The experimental result of the proposed classifier are compared with those studied by Dalal.

• Proceedings of the KIEE Conference
• /
• 2008.07a
• /
• pp.1374-1375
• /
• 2008

본 논문은 3상 일괄형 GIS의 부분방전 진단을 위해 새롭게 개발한 진단 알고리즘에 관한 것이다. 진단 알고리즘 개발을 위해, 먼저 실시간 부분방전 데이터를 행벡터 및 열벡터로 구성하고 각각의 벡터에서 통계 특징량 및 질감 특징량을 추출하였다. 다음으로 이들 특징량을 GA-NN(Genetic Algorithm - Neural Network) 학습에 적용하여 진단 알고리즘을 구성하였다. 또한 진단 알고리즘의 위상독립성은 부분방전 신호의 위상변화에 관계없이 진단결과가 일치하는 것을 확인함으로써 검증하였다. 개발한 진단알고리즘의 실증 평가를 위해, 부분방전이 발생되고 있는 국내 3상 일괄형 GIS 변전소에 적용하였다. 적용 결과, 위상에 관계없이 부분방전 발생원을 정확히 진단함을 확인하였고, 이를 통해 개발 알고리즘의 우수성을 입증하였다.

• Proceedings of the KIEE Conference
• /
• 2008.07a
• /
• pp.1857-1858
• /
• 2008

본 논문에서는 FCM 기반 RBF 뉴럴네트워크(FCM-RBFNN) 구조를 제안하고 PSO를 이용한 FCM-RBFNN의 구조 및 파라미터의 최적화 방법을 제시한다. 클러스터링 알고리즘은 퍼지 뉴럴 네트워크에서 멤버쉽함수의 중심점과 반경 등을 결정하는 학습에 일반적으로 사용된다. 제안된 FCM-RBFNN서는 방사기저함수로써 가우시안, 삼각형 타입 등의 정해진 형태를 사용하지 않고 데이터들 사이의 거리에 관계된 계산을 수행하는 FCM에 의해 결정된다. 기존의 RBFNN에서 후반부는 상수형태로써 방사기저함수의 선형결합으로써 표현되는 반면에 제안된 FCM-RBFNN의 후반부는 상수형, 선형, 2차식 등의 다양한 형태의 다항식으로 표현될 수 있으며 다항식의 계수는 WLSE를 이용하여 추정한다. FCM 기반 RBF 뉴럴 네트워크의 성능은 퍼지규칙의 수, 후반부 다항식의 차수 FCM의 퍼지화 계수에 의하여 결정기 때문에 FCM-RBFNN의 구조와 파라미터의 최적화가 요구된다. 본 논문에서는 PSO를 이용하여 FCM-RBFNN의 구조에 관련된 퍼지 규칙의 수, 후반부 다항식의 차수와 파라미터에 관련된 퍼지화 계수를 최적화한다. 또한 후반부 다항식의 계수는 WLSE를 사용하여 추정한다.

• Journal of Korean Institute of Industrial Engineers
• /
• v.31 no.3
• /
• pp.257-264
• /
• 2005

This paper suggests a heuristic algorithm for the clustering problem. Clustering involves grouping similar objects into a cluster. Clustering is used in a wide variety of fields including data mining, marketing, and biology. Until now there are a lot of approaches using Self-Organizing Feature Maps(SOFMs). But they have problems with a small output-layer nodes and initial weight. For example, one of them is a one-dimension map of k output-layer nodes, if they want to make k clusters. This approach has problems to classify elaboratively. This paper suggests one-dimensional output-layer nodes in SOFMs. The number of output-layer nodes is more than those of clusters intended to find and the order of output-layer nodes is ascending in the sum of the output-layer node's weight. We can find input data in SOFMs output node and classify input data in output nodes using Euclidean distance. We use the well known IRIS data as an experimental data. Unsupervised clustering of IRIS data typically results in 15 - 17 clustering error. However, the proposed algorithm has only six clustering errors.

• IE interfaces
• /
• v.22 no.3
• /
• pp.252-262
• /
• 2009

It is crucial to prevent low yields in the semiconductor industry. Since many factors affect variation in yield and they are deeply related, preventing low yield is difficult. There have been substantial researches in the field of yield prediction. Many researchers had used the statistical methods. Many studies have shown that artificial neural network (ANN) achieved better performance than traditional statistical methods. However, despite ANN's superior performance some problems such as over-fitting and poor explanatory power arise. In order to overcome these limitations, a relatively new machine learning technique, support vector machine (SVM), is introduced to classify the yield. SVM is simple enough to be analyzed mathematically, and it leads to high performances in practical applications. This study presents a new efficient classification methodology, Stepwise-SVM (SSVM), for detecting high and low yields. SSVM is step-by-step adjustment of parameters to be precisely the classification for actual high and low yield lot. The objective of this paper is to examine the feasibility of SVM and SSVM in the yield classification. The experimental results show that SVM and SSVM provides a promising alternative to yield classification for the field data.