Earticle

In the classifier which is constructed with a fully connected layer connected to an output layer the weight vector W creates a decision boundary between each. The learning of a feature vector influences the model to extract unique features of the class during the learning process which moves the distribution of the initial feature vector into the decision boundary of the class. When the similarity between the initial feature vector and the weight vector is high, it can be expected that the effect on the model is low because the loss value is small. On the other hand, if the similarity is low, it means that the interval between the feature vector and the weight vector is large and the loss value is high, which can be expected to have a higher effect on the model than when the similarity is high in the learning process of the model. In this paper, we verify how much the similarity between the initial feature vector and the weight vector before learning affects model learning. In order to confirm the effect of similarity, the model is learned by assigning an arbitrary fixed value so that the weight vectors W of the fully connected layer make different similarities. Both VGG16 and VGG19 models are used to compare the Recall and the Precision values of the class as a learning result of each model.

This paper focuses on the approach to solve the immune evolution optimization with interval number judgment matrix weight vector. In light of the features of interval number judgment matrix, we transform the solution problem of weights into the optimization one of nonlinear with restriction, design immune evolution algorithm on the base of immune mechanism, and use the convergence ability of immune system to search the optimal access to get interval numbers from matrix weight vector. Moreover, we construct immune operator and form vaccine and optimization strategies by priori knowledge of target problem, by which we can keep individual diversity and elite individuals and abandon useless individuals as soon as possible. In the evolution process, search can be dramatically improved by overcoming the problems of earliness and degradation during global search. Simulation results of experiments show the advantages of this algorithm in accuracy, convergence, convergence rate and so on.

In allusion to such problems as real-time requirement dissatisfaction and significant recognition difference caused by dimension difference existing in the imaging and recognition algorithm for pedestrian in dark scene, a fast head detection and recognition method for pedestrian at night based on fast support vector machine (FC-SVM) algorithm optimization and entropy weight is established in this paper according to relevant principle of statistics. Based on entropy weight, this method aims at improving the extraction process based on histogram gradient features in order to establish threebranch SVM for the deep recognition of pedestrian at night; meanwhile, FC-SVM algorithm is combined to optimize the recognition calculation overhead in order to ensure the real-time property of the recognition algorithm. Furthermore, the falsely detected pedestrians are evaluated on the basis of the head detection mode so as to improve pedestrian imaging matching accuracy. The simulation result shows that this method can not only effectively recognize FIR target of pedestrian at night, but also effectively adapt to such different application environments as urban and suburban areas on the basis of ensuring the real-time requirement for pedestrian recognition, thus presenting good practicability.

In gesture recognition, gesture is easy to be influenced by nonlinear factors such as illumination, wrists and motion blur, etc., based on the Analysis of the SVM classification method and its effect in practical application, this paper proposes a gesture method based on the combination of nonlinear support vector machine (SVM) and Linear Discriminant Analysis, LDA (SVM+LDA).First it works out gesture recognition system constituted by multiple-surface electrode sensor, microcontroller acquisition unit, and computer; second, to get the best classification feature of image through linear discriminant analysis with LDA feature extraction; finally, to carry out classification recognition on gesture characteristic vector by applying the nearest neighbor classifier, at the same time compared with other classification algorithms to get the advantages and disadvantages of the algorithm in this paper. The experimental results show that the SVM+LDA can gain higher recognition rate, so as to provide theoretical and data support for the researcher of electromyogram artificial hands control to choose the appropriate means of recognition.

본 논문은 기존 CMA(Constant Modulus Algorithm) 적응 등화기의 성능인 수렴 특성과 잔류 부호 간 간섭을 개선하기 위하여 가중치 벡터의 업데이트를 이중으로 적용한 DW-MCMA(Dual Weight vector Modified Constant Modulus Algorithm)에 대한 관한 것이다. 송신 시스템과 수신 시스템을 연결하는 무선과 유선의 대역 제한 채널에서 발생되는 부호간 간섭에 의한 찌그러짐을 줄이기 위하여 등화기가 사용되며, 이의 대표적인 알고리즘으로는 CMA가 널리 알려져 있다. 이와 같은 채널을 통하여 대량의 정보를 고속으로 전송하기 위해서는 등화에 필요한 오버헤드를 최소화할 수 있는 빠른 수렴 속도가 요구된다. 논문에서는 컴퓨터 시뮬레이션을 통하여 기존 CMA와 MCMA보다 등화기 입력단에서 정상 등화 상태의 오차를 줄일 수 있는 새로운 비용함수와 가중치 벡터를 이중으로 적용하는 제안 방식인 DW-MCMA가 잔류 부호간 간섭량과 빠른 속도를 가짐을 확인하였다.

This paper is concerned with the DW-MCMA(Dual Weight vector Modified Constant Modulus Algorithm) adaptive equalization algorithm using the dual weight vector in order to improve the convergence characteristic and residual inter-symbol interference which are used as the performance index for an adaptive equalizer. The equalizer is used to reduce the distortion caused by the inter-symbol interference on the wireless and the wired band-limited channel that connect the transmitting system and receiving system. The CMA is widely known as the representative algorithm for equalization. In order to transmitting the mass information with a high speed through the channels, a fast convergence speed in the equalizer performance that is able to minimize overhead needed for equalization is acquired. In this paper, By the computer simulation, we confirmed that the proposed DW-MCMA has the faster convergence speed and the smaller residual inter-symbol interference than the conventional CMA and MCMA.

본 연구는 공간상에서 전파를 이용하여 원하는 목표물의 도래 방향을 추정 한다. 도래방향 추정은 수신 배열 안테나들로 입사하는 신호들 중에서 원하는 목표물의 위치를 찾는 것이다. 본 연구에서는 도래방향 추정의 고 분해능 MUSIC알고리즘과 비용함수를 사용하여 목표물에 대한 도래방향을 추정하였고, 최적의 가중치를 계산하였다. 모의실험을 통하여 목표물 도래 방향 추정에서 기존 ESPRIT 알고리즘과 제안 알고리즘의 성능을 비교 분석 하였다. 목표물 도래 방향 추정에서 제안한 알고리즘이 기존의 ESPRIT 알고리즘보다 도래 방향 추정 능력이 향상되었다.

This paper estimates the direction of arrival of desired a target using propagation wave in spatial. Direction of arrival estimation is to find desired target position among received signal to receiver array antennas. In this paper, we estimated direction of arrival for target, by using cost function and high resolution MUSIC algorithm, in order to direction of arrival estimation, and calculated optimum weight vector. Through simulation, in regard to the estimation of the arrival direction of a target, the performances of the existing ESPRIT algorithm and the proposed algorithm were comparatively analyzed. In the estimation time of the arrival direction of a target object, the proposed algorithm showed an improvement of approximately as compared to the existing ESPRIT algorithm.

This paper develops a simple interactive procedure which can be efficiently used to solve a bicriteria linear programming problem. The procedure exploits the relatively simple structure of the bicriterion linear programming problem. Its application to a transportation problem is also presented. The results demonstrate that the method developed in this paper could be easily applicable to any bicriteria linear program in general.

Vector-valued Eisenstein series of weight 3/2 are often not holomorphic. In this paper we prove that, for an even lattice Ḻ, if there exists an odd prime p such that Ḻ is local p-maximal and the determinant of Ḻ is divisible by p2, then the Eisenstein series of weight 3/2 attached to the discriminant form of Ḻ is holomorphic.

Muilti-dimensional feature vector matching algorithm uses multiple features such as intensity, gradient, variance, first or second derivative of a pixel to find correspondence pixels in stereo images. In this paper, we proposed a new method for adjusting automatically the weight of feature in multi-dimensional feature vector matching considering sharpeness of a pixel in feature vector distance curve. The sharpeness consists of minimum and maximum vector distances of a small window mask. In the experiment we used IKONOS satellite stereo imagery and obtained accurate matching results comparable to the manual weight-adjusting method.

The maximum weight of single stock in mutual fund is limited by regulations to enforce diversification. Under incomplete information with added constraints on portfolio weights, enhanced performance had been reported in previous researches. We analyze a weight vector to examine the effects of additional constraints on the portfolio's performance by computing the Euclidean distance from the in-sample tangency portfolio, as opposed to previous researches which analyzed ex-post return only. Empirical experiment was performed on Mean-variance and Minimum-variance model with Fama French's 30 industry portfolio and 10 industry portfolio for the last 1,000 months from August 1932 to November 2015. We find that diversification-constrained portfolios have 7% to 26% smaller Euclidean distances with the benchmark portfolio compared to those of unconstrained portfolios and 3% to 11% greater Sharpe Ratio.

본 논문은 특징 가중치 벡터를 적용하여 능동형태 모델(Active Shape Model)기반에서 눈동자의 움직임 추적 속도를 향상시키는 방법을 제안한다. 일반적인 능동형태 모델에서는 객체 추적을 위한 PDM 구성을 위해 각 특징점 구성 벡터의 유클리디안 거리의 최소 값으로 Training Set정렬 과정을 수행한다. 이 과정에서 적절하지 못한 샘플 영상으로 인해 안정된 PDM을 구성하지 못하는 문제점이 발생한다. 이러한 문제점을 해결하기 위하여 본 논문에 서는 형태를 구성하는 특징점마다 가중치를 부여하는 벡터를 작성하고, 최소자승근사법으로 최유사 특징점 벡터를 산출하기 위한 선형방정식을 구상하였다. 이로 인해 안정된 PDM을 구성할 수 있었으며, 눈동자 추적실험을 통해 형태적 움직임을 보정하는 실험을 수행하였다. 실험결과 기존의 능동형태 모델에 비해 반복연산의 횟수가 줄어들고, 다양한 형태로 나타나는 눈동자의 움직임 추적에 보다 안정적인 결과를 얻을 수 있었다.

멀티미디어 압축 및 이동 통신 기술의 발전으로 다양한 형태의 멀티미디어 서비스가 이슈화되고 있다. 비디오를 전송하기 위해서는 많은 대역폭을 필요로 하지만, 모든 네트워크가 높은 수준의 대역 및 처리 능력을 가지는 것은 아니다. 이질적인 네트워크간의 멀티미디어를 서비스하기 위해서는 네트워크 상황 또는 수신자의 처리 능력에 맞도록 재 부호화해야 하지만 그 처리비용이 높다. 트랜스코딩 기법 중에서 시간당 요구된 프레임의 개수를 조절하면 제거된 프레임의 움직임 벡터를 재 사용하여 비트율을 감소시킬 수 있다. 본 논문에서는 기존의 기법보다 향상된 움직임과 중첩 영역의 정보를 적용한 WBVC(Weight-Based Vector Composition) 기법을 제안한다. 실험을 통한 기존의 기법과의 비교 분석 결과, 비슷한 계산 복잡도에서 제안한 WBVC 기법이 높은 성능을 보였다.

멀티미디어 압축 및 이동 통신 기술의 발전으로 다양한 형태의 멀티미디어 서비스가 이슈화되고 있다. 비디오를 전송하기 위해서는 많은 대역폭을 필요로 하지만, 모든 네트워크가 높은 수준의 대역 및 처리 능력을 가지는 것은 아니다. 이질적인 네트워크간의 멀티미디어를 서비스하기 위해서는 네트워크 상황 또는 수신자의 처리 능력에 맞도록 재 부호화해야 하지만 그 처리비용이 높다. 트랜스코딩 기법 중에서 시간당 요구된 프레임의 개수를 조절하면 제거된 프레임의 움직임 벡터를 재 사용하여 비트율을 감소시킬 수 있다. 본 논문에서는 기존의 기법보다 향상된 움직임과 중첩 영역의 정보를 적용한 WBVC(Weight-Based Vector Composition) 기법을 제안한다. 실험을 통한 기존의 기법과의 비교 분석 결과, 비슷한 계산 복잡도에서 제안한 WBVC 기법이 높은 성능을 보였다.