• Communications for Statistical Applications and Methods
• /
• 제18권6호
• /
• pp.871-886
• /
• 2011

This paper proposes a powerful SVM-ASM filter, the adaptive switching median(ASM) filter based on support vector machines(SVMs), to effectively reduce impulse noise in corrupted images while preserving image details and features. The proposed SVM-ASM filter is composed of two stages: SVM impulse detection and ASM filtering. SVM impulse detection determines whether the pixels are corrupted by noise or not according to an optimal discrimination function. ASM filtering implements the image filtering with a variable window size to effectively remove the noisy pixels determined by the SVM impulse detection. Experimental results show that the SVM-ASM filter performs significantly better than many other existing filters for denoising impulse noise even in highly corrupted images with regard to noise suppression and detail preservation. The SVM-ASM filter is also extremely robust with respect to various test images and various percentages of image noise.

• 한국컴퓨터정보학회논문지
• /
• 제18권7호
• /
• pp.37-45
• /
• 2013

문맥 종속형 융합(CDF, Context Dependent Fusion)은 여러 분류기의 결과를 종합하여 성능을 향상시키는 융합 방법으로 주어진 문제의 문맥을 균일한 여러 문맥으로 나누고 각 문맥에서 문맥 종속적인 융합을 시도함으로써 기존 융합 방법에 비해 향상된 성능을 보여주었다. 하지만 CDF는 학습해야할 파라미터의 개수가 많아 학습 데이터가 적은 경우 잡음에 민감한 문제점이 있으며, 선형 알고리듬이라는 한계로 인해 문맥 추출 및 지역적 융합 과정에서 성능 저하의 원인이 된다. 본 논문에서는 CDF의 문제점을 완화할 수 있는 방법으로 SVM(Support Vector Machine)과 커널 주성분 분석을 이용한 CDF-SVM을 제안하였다. 커널 주성분 분석은 입력 벡터에 비선형 변환을 가함으로써 타원형이 아닌 비정형의 클러스터 생성이 가능하도록 해주며, SVM은 융합과정에서 비선형 경계의 생성을 가능하게 해주어 CDF의 선형성 제약을 극복하도록 해준다. 또한 목적함수에 정규화 항을 추가함으로써 잡음 민감성을 줄이도록 하였다. 제안한 CDF-SVM은 기존 CDF 및 그 변형들에 비해 나은 성능을 보여주었으며 이는 실험 결과를 통해 확인할 수 있다.

• Communications for Statistical Applications and Methods
• /
• 제20권3호
• /
• pp.225-233
• /
• 2013

The linear support vector machine(SVM) is motivated by the maximal margin separating hyperplane and is a popular tool for binary classification tasks. Many studies exist on the consistency properties of SVM; however, it is unknown whether the linear SVM is consistent for estimating the optimal classification boundary even in the simple case of two Gaussian classes with a common covariance, where the optimal classification boundary is linear. In this paper we show that the linear SVM can be inconsistent in the univariate Gaussian classification problem with a common variance, even when the best tuning parameter is used.

• 한국경영정보학회:학술대회논문집
• /
• 한국경영정보학회 2007년도 International Conference
• /
• pp.608-613
• /
• 2007

Prediction of corporate bankruptcies has long been an important topic and has been studied extensively in the finance and management literature because it is an essential basis for the risk management of financial institutions. Recently, support vector machines (SVMs) are becoming popular as a tool for bankruptcy prediction because they use a risk function consisting of the empirical error and a regularized term which is derived from the structural risk minimization principle. In addition, they don't require huge training samples and have little possibility of overfitting. However. in order to Use SVM, a user should determine several factors such as the parameters ofa kernel function, appropriate feature subset, and proper instance subset by heuristics, which hinders accurate prediction results when using SVM In this study, we propose a novel hybrid SVM classifier with simultaneous optimization of feature subsets, instance subsets, and kernel parameters. This study introduces genetic algorithms (GAs) to optimize the feature selection, instance selection, and kernel parameters simultaneously. Our study applies the proposed model to the real-world case for bankruptcy prediction. Experimental results show that the prediction accuracy of conventional SVM may be improved significantly by using our model.

• 정보과학회 논문지
• /
• 제42권1호
• /
• pp.86-92
• /
• 2015

레벨 셋 트리는 다차원에 정의된 확률 밀도 함수를 표현하는데 유용하다. 복잡한 데이터의 구조를 트리 형태로 시각화하여 데이터의 형태를 효율적으로 파악할 수 있으며 클러스터링 분석에 효과적으로 이용할 수 있다. 본 논문에서는 미지의 확률 밀도 함수에서 생성된 데이터 샘플로부터 레벨 셋 트리를 생성하는 알고리즘을 제안한다. 제안된 알고리즘은 레벨을 0에서부터 무한대로 증가시키며 밀도 함수의 각 레벨 셋을 추정하고, 이로부터 레벨 셋 트리를 생성한다. 이를 위해 본 논문에서는 one-class 서포트 벡터 머신 (OC-SVM)을 이용하여 직접적으로 레벨 셋을 추정한다. 이때 다양한 레벨 값에 대해 OC-SVM 학습을 반복해야 하는데, OC-SVM 솔루션 path 알고리즘을 통해 빠른 시간 안에 모든 레벨값에 해당하는 레벨 셋를 추정할 수 있다.

• 한국소음진동공학회논문집
• /
• 제14권12호
• /
• pp.1233-1240
• /
• 2004

Condition monitoring and fault diagnosis of machines are gaining importance in the industry because of the need to increase reliability and to decrease possible loss of production due to machine breakdown. By comparing the nitration signals of a machine running in normal and faulty conditions, detection of faults like mass unbalance, shaft misalignment and bearing defects is possible. This paper presents a novel approach for applying the fault diagnosis of rotating machinery. To detect multiple faults in rotating machinery, a feature selection method and support vector machine (SVM) based multi-class classifier are constructed and used in the faults diagnosis. The results in experiments prove that fault types can be diagnosed by the above method.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2006년도 Proceedings of ISRS 2006 PORSEC Volume I
• /
• pp.114-117
• /
• 2006

One of the most important tasks of ocean color observations is to determine the distribution of phytoplankton primary production. A variety of bio-optical algorithms have been developed estimate primary production from these parameters. In this communication, we investigated the possibility of using a novel universal approximator-support vector machines (SVMs)-as the nonlinear transfer function between oceanic primary production and the information that can be directly retrieved from satellite data. The VGPM (Vertically Generalized Production Model) dataset was used to evaluate the proposed approach. The PPARR2 (Primary Production Algorithm Round Robin 2) dataset was used to further compare the precision between the VGPM model and the SVM model. Using this SVM model to calculate the global ocean primary production, the result is 45.5 PgC $yr^{-1}$, which is a little higher than the VGPM result.

• 충청수학회지
• /
• 제29권4호
• /
• pp.631-642
• /
• 2016

This article deals with one-dimension backward heat conduction problem (BHCP). A new approach based on least squares support vector machines (LS-SVM) is proposed for obtaining their approximate solutions. The approximate solution is presented in closed form by means of LS-SVM, whose parameters are adjusted to minimize an appropriate error function. The approximate solution consists of two parts. The first part is a known function that satisfies initial and boundary conditions. The other is a product of two terms. One term is known function which has zero boundary and initial conditions, another term is unknown which is related to kernel functions. This method has been successfully tested on practical examples and has yielded higher accuracy and stable solutions.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제9권3호
• /
• pp.213-218
• /
• 2009

This paper describes a classical algorithm carrying out dynamic 3D obstacle recognition for autonomous underwater vehicles (AUVs), Support Vector Machines (SVMs). SVM is an efficient algorithm that was developed for recognizing 3D object in recent years. A recognition system is designed using Support Vector Machines for applying the capabilities on appearance-based 3D obstacle recognition. All of the test data are taken from OpenGL Simulation. The OpenGL which draws dynamic obstacles environment is used to carry out the experiment for the situation of three-dimension. In order to verify the performance of proposed SVMs, it compares with Back-Propagation algorithm through OpenGL simulation in view of the obstacle recognition accuracy and the time efficiency.

• Communications for Statistical Applications and Methods
• /
• 제16권5호
• /
• pp.881-889
• /
• 2009

Support vector machines(SVMs) are known as flexible and efficient classifier of multivariate observations, producing a hyperplane or hyperdimensional curved surface in multidimensional feature space that best separates training samples by known groups. As various methodological extensions are made for SVM classifiers in recent years, it becomes more difficult to understand the constructed model intuitively. The aim of this paper is to visualize various SVM classifications tuned by several parameters in reduced dimensions, so that data analysts secure the tangible image of the products that the machine made.