Earticle

최근 빅데이터를 수용하기 위한 대용량 저장 장치가 필요한 엔터프라이즈 저장 시스템에서는 비용과 크기 대비 직접도가 높은 대용량의 플래시 메모리 기반 저장 장치를 많이 사용하고 있다. 본 논문에서는 엔터프라이즈 대용량 저장 장치의 신뢰도와 이용성에 직접적인 영향을 주는 플래시 메모리 미디어의 수명을 극대화 하기 위해 경사하강법을 적용한 고효율 수명 예측 방법을 제안한다. 이를 위해 본 논문에서는 불량 발생 빈도를 학습하기 위한 메타 데이터를 저장하는 매트릭스의 구조를 제안하고 메타데이터를 이용한 비용 모델을 제안한다. 또한 학습된 범위를 벗어난 불량이 발생 했을 때 예외 상황에서의 수명 예측 정책을 제안한다. 마지막으로 시뮬레이션을 통해 본 논문에서 제안하는 방법이 이전까지 플래시 메모리의 수명 예측을 위해 사용되어 온 고정 횟수 기반 수명 예측 방법과 예비 블록의 남은 비율을 기반으로 하는 수명 예측 방법 대비 수명을 극대화 할 수 있음을 증명하여 우수성을 확인했다.

Recently, enterprise storage systems that require large-capacity storage devices to accommodate big data have used large-capacity flash memory-based storage devices with high density compared to cost and size. This paper proposes a high-efficiency life prediction method with slope descent to maximize the life of flash memory media that directly affects the reliability and usability of large enterprise storage devices. To this end, this paper proposes the structure of a matrix for storing metadata for learning the frequency of defects and proposes a cost model using metadata. It also proposes a life expectancy prediction policy in exceptional situations when defects outside the learned range occur. Lastly, it was verified through simulation that a method proposed by this paper can maximize its life compared to a life prediction method based on the fixed number of times and the life prediction method based on the remaining ratio of spare blocks, which has been used to predict the life of flash memory.

본 논문은 그레이디언트 기반 눈 검출을 위한 비동공 영역 레이블링 방법에 관한 것이다. 제안된 방법은 비동공 영역 레이블링을 이용하여 기존의 그레이디언트 기반 눈 검출 방법보다 개선된 검출 정확도와 연산 속도를 제공함에 그 목적이 있다. 우선 Haar-like feature와 AdaBoost를 이용하여 얼굴 영역을 구한 뒤, 얼굴의 기하학적 특징을 이용하여 좌측과 우측의 눈 탐색 영역을 지정한다. 이후 눈 탐색 영역 내 에지 화소들의 그레이디언트 벡터와 정규 변위 벡터 간의 내적을 누적한 후 최대 누적값의 위치를 좌우 눈의 중심으로 검출한다. 제안된 방법은 동공의 중심은 저명도 평탄 영역에 위치함에 착안하여 비동공 영역으로 추정되는 부분을 내적 누적 연산에서 제외시킴으로써 검출 정확도와 연산 속도를 개선한 것이다. 눈 탐색 영역의 가우시안 필터링된 역 영상을 히스토그램 평활화하여 정규화한 후 임계처리를 통해 동공 후보 영역과 비동공 영역을 구분한다. 하지만 저명도의 안경테나 눈썹, 머리카락과 같은 영역이 후보 영역에 포함될 수 있으므로 비동공 영역 레이블링 기법을 이용해 동공 후보 영역에서 제거하고 전체 눈 탐색 영역이 아닌 후보 영역에서만 그레이디언트 기반 눈 검출을 수행한다. 시뮬레이션 결과에 따르면, 제안된 방법은 기존의 방법[9] 및 개선된 기존의 방법들[10][11]에 비해 우수한 검출 정확도와 연산 속도를 제공하는데, 특히 안경 착용 시 각각 약 51% 및 39.7%, 19% 정도의 연산 성능이 개선되는 장점이 있다.

For driver convenience, different types of transmission are being developed, such as AT(Automatic Transmission), AMT(Automated Manual Transmission), CVT(Continuously Variable Transmission) and DCT(Dual Clutch Transmission). To improve ride comfort and durability of the transmission, control system is important during launching and shifting process. For accurate control, vehicle mass and road gradient should be known. In this study, heavy duty vehicle’s mass and road gradient estimation method is developed. The method uses only signals from CAN(Controller Area Network) without applying extra sensors. Vehicle mass and gradient is estimated by LMS(Least Mean Square) method based on longitudinal vehicle dynamic model. To verify the estimation logic, test was conducted using a chassis dynamometer. The estimation results after test and test condition is compared. The error rate of vehicle mass estimation was 5 percent and gradient estimation result had 2 percent error.

An effective novel approach of detection and classification of real face image from tampered face image based on second order gradient is proposed in this paper. The intended purpose of proposed approach is to endorse the biometric authentication, by joining the vitality awareness with Facial Recognition Technology (FRT). The proposed method requires only one face image without requirement of additional equipment and easier to implement into existing face recognition technique. For this purpose, real (from own database and some publically available standard database) and tampered (own prepared databases of dummy, color imposed and masked faces) face image database are used here for verification and validation of our assertion. The technique is novel technique and obtained results are initial results which are obtained after applying gradient method and demonstrate that the methodology is very well suited for the discrimination of image of tampered face from the image of real face with accuracy ranges 82.7% 91.7%. This reliable way to detect the mala-fide attack is needed to robustness of the system and it will be able to solve very big real problems of the society when induced in automatic authentication system.

Ear identification is an important biometric identification technique and has been widely used in many applications. One of the most important components in ear identification system is image preprocessing. The performance of image preprocessing has a significant impact on the accuracy of ear identification system. This paper proposes a L0 gradient based image smoothing method for ear identification. First, ear images are filtered using L0 gradient based method. Then the contrast of the image is enhanced using histogram equalization method in order to make ear edges more discriminative. Comparative experiments with an existing algorithm demonstrate that our method has better performance and is more suitable for ear identification.

Protective measures can be rapidly and reliably taken by timely and correctly selecting the fault phase coping with the line protection, as well as the right operation of single-phase tripping and single phase re-closing. Traditional selection method is insensitive and insufficient in dealing with fault resistance, fault position, mutual inductance between closed lines, reactive effect and the in-sufficient for system parameters using power frequency component. This paper has proposed series multi-resolution morphological gradient (SMMG) filter, to extract the deviation of power frequency component of mode current, constructing a new phase selecting component of fault phase component, which provides a solution to fault phase selecting component of fault phase component.

In order to research the problem of the gray edge and leaving out some edges during detecting the color edges, the color contrast enhancement algorithm is designed that it dependences on decomposing the interrelated color component of the color images based on RGB color model. And the edge detection improved method is proposed based on the above algorithm. Namely, the components edges are computed by the gradient operators after the components are enhanced and the color image edges are composed. The experimental results show that the color edges are detected through the proposed method and the algorithm has the advantage of keeping the more edge details than the other ways. Because the algorithm is an improved method of pixel processing based on three components of RGB color model, it needn’t transform one color model to the other color model, so it is a more simple method than the other methods.

It is known that the steepest descent(SD) method and the conjugate gradient(CG) method [1, 2, 5, 6] converge when these methods are applied to solve linear systems of the form Ax = b, where A is symmetric and positive definite. For some finite difference discretizations of elliptic problems, one gets positive definite matrices that are almost symmetric. Practically, the SD method and the CG method work for these matrices. However, the convergence of these methods is not guaranteed theoretically. The SD method is also called Orthores(1) in iterative method papers. Elman [4] states that the convergence proof for Orthores($\kappa$), with $\kappa$ a positive integer, is not heard. In this paper, we prove that the SD method and the CG method converge when the $\iota$$^2$ matrix norm of the non-symmetric part of a positive definite matrix is less than some value related to the smallest and the largest eigenvalues of the symmetric part of the given matrix.(omitted)

We show how the minimization can be used to solve the quadratic matrix equattion. We then compare two different types of conjugate gradient method and show Polak and Ribire version converge more rapidly than Fletcher and Reeves version in several examples.

In this paper, we develop a new hybridization conjugate gradient method for solving the unconstrained optimization problem. Under mild assumptions, we get the sufficient descent property of the given method. The global convergence of the given method is also presented under the Wolfe-type line search and the general Wolfe line search. The numerical results show that the method is also efficient.

In this paper we consider the global convergence of a new super memory gradient method for unconstrained optimization problems. New trust region radius is proposed to make the new method converge stably and averagely, and it will be suitable to solve large scale minimization problems. Some global convergence results are obtained under some mild conditions. Numerical results show that this new method is effective and stable in practical computation.

A modification of the gradient method of convex programming is introduced. Also, we describe symbolic implementation of the gradient method and its modification by means of the programming language MATHEMATICA. A few numerical examples are reported.

A Conjugate Gradiant (CG) method is proposed for unconstained optimization which is invariant to a nonlinear scaling of a strictly convex quadratic function. The technique has the same properties as the classical CG-method when applied to a quadratic function. The algorithm derived here is based on a logarithmic model and is compared to the standard CG method of Fletcher and Reeves [3]. Numerical results are encouraging and indicate that nonlinear scaling is promising and deserves further investigation.

In this paper, we propose a new incremental gradient method for solving a regularized minimization problem whose objective is the sum of m smooth functions and a (possibly nonsmooth) convex function. This method uses an adaptive stepsize. Recently proposed incremental gradient methods for a regularized minimization problem need O(mn) storage, where n is the number of variables. This is the drawback of them. But, the proposed new incremental gradient method requires only O(n) storage.

A preconditioned conjugate gradient(PCG) scheme with the aid of deflation for computing a few of the smallest eigenvalues arid their corresponding eigenvectors of the large generalized eigenproblems is considered. Topically there are two types of deflation techniques, the deflation with partial shifts and an arthogonal deflation. The efficient way of determining partial shifts is suggested and the deflation-PCG schemes with various partial shifts are investigated. Comparisons of theme schemes are made with orthogonal deflation-PCG, and their asymptotic behaviors with restart operation are also discussed.