• 한국멀티미디어학회논문지
• /
• 제22권6호
• /
• pp.655-664
• /
• 2019

In this paper, we propose a novel way of combining multiple deep convolutional neural network (DCNN) architectures which work well for accurate video face identification by adopting a serial combination of 3D and 2D DCNNs. The proposed method first divides an input video sequence (to be recognized) into a number of sub-video sequences. The resulting sub-video sequences are used as input to the 3D DCNN so as to obtain the class-confidence scores for a given input video sequence by considering both temporal and spatial face feature characteristics of input video sequence. The class-confidence scores obtained from corresponding sub-video sequences is combined by forming our proposed class-confidence matrix. The resulting class-confidence matrix is then used as an input for learning 2D DCNN learning which is serially linked to 3D DCNN. Finally, fine-tuned, serially combined DCNN framework is applied for recognizing the identity present in a given test video sequence. To verify the effectiveness of our proposed method, extensive and comparative experiments have been conducted to evaluate our method on COX face databases with their standard face identification protocols. Experimental results showed that our method can achieve better or comparable identification rate compared to other state-of-the-art video FR methods.

• 한국통신학회논문지
• /
• 제9권3호
• /
• pp.127-131
• /
• 1984

벡터-매트릭스 곱셈을 인코히어런트(incoherent)광원에 의해 빠른 속도로 대량의 정보를 처리할 수 있는 IOVMM(incoherent optical vector matrix multiplier)을 구성하고 실험결과와 이론치를 비교하였다. 입력 벡터 및 매트릭스의 원소들은 양의 실수로만 국한시키고 입력 벡터는 LED배열로 나타내었으며 매트릭스는 마스크상에 면적변조방식으로 부호화하였다. 이 두 곱셈의 결과는 렌즈계를 통하여 포토 다이오우드 배열로 검출하였으며 하나의 채널로 출력신호를 관찰하기 위하여 애널로그 멀티플렉스를 사용하였다.

• 전기전자학회논문지
• /
• 제23권4호
• /
• pp.1353-1359
• /
• 2019

It is presented in this paper that the static output feedback (SOF) pole-assignment problem of some linear time-invariant systems can be completely resolved by parametrization in real Grassmann space. For the real Grassmannian parametrization, the so-called Plucker matrix is utilized as a linear matrix formula formulated from the SOF variable's coefficients of a characteristic polynomial constrained in Grassmann space. It is found that the exact SOF pole assignability is determined by the linear independency of columns of Plucker sub-matrix and by full-rank of that sub-matrix. It is also presented that previous diverse pole-assignment methods and various computation algorithms of the real SOF gains for 2-input, 2-output, 4^th order systems are unified in a deterministic way within this real Grassmannian parametrization method.

• 전력전자학회:학술대회논문집
• /
• 전력전자학회 2010년도 하계학술대회 논문집
• /
• pp.21-22
• /
• 2010

Matrix converter is direct power conversion system. Matrix converter has many merits that possible bidirectional power flow, input power factor own control and system without DC-link. But matrix converter has some demerits that need many switching devices and switching loss. This paper suggest optimal matrix converter control scheme for improvement for switching loss part. Proposed control scheme verified that 10% improvement in efficiency, input current's harmonic loss and output voltage's EMI improvement.

• 한국정밀공학회지
• /
• 제23권6호
• /
• pp.88-95
• /
• 2006

In this paper, the gain-scheduled control design proposed in the previous paper has been applied to a target tracking system. In such system, it is needed to attenuate disturbance effectively as long as control input satisfies the given constraint on its magnitude. The scheduled gains are derived in the framework of linear matrix inequality(LMI) optimization by means of the MatLab toolbox. Its effectiveness is verified along with the simulation results compared with the conventional optimum constant gain and the scheduled gain control with constant Q matrix cases.

• Journal of Power Electronics
• /
• 제14권4호
• /
• pp.778-787
• /
• 2014

This study presents an input filter resonance mitigation method for an AC-DC matrix converter. This method combines the advantages of the one-cycle control strategy and the active damping technique. Unnecessary sensors are removed, and system cost is reduced by employing the grid-side input currents as feedback to damp out LC resonance. A model that includes the proposed method and the input filter is established with consideration of the delay caused by the actual controller. A zero-pole map is employed to analyze model stability and to investigate virtual resistor parameter design principles. Based on a double closed-loop control scheme, the one-cycle control strategy does not require any complex modulation index control. Thus, this strategy can be more easily implemented than traditional space vector-based methods. Experimental results demonstrate the veracity of theoretical analysis and the feasibility of the proposed approach.

• 한국정보통신학회논문지
• /
• 제25권11호
• /
• pp.1687-1695
• /
• 2021

본 논문에서는 버틀러 매트릭스를 이용한 빔포밍 안테나를 설계 및 분석하였다. 제안한 빔포밍 안테나의 동작 주파수는 2.4 GHz의 ISM 대역이며, 빔포밍 안테나의 구성 요소는 1 × 4 배열 안테나 및 4 × 4 버틀러 매트릭스로 구성된다. 4 × 4 버틀러 매트릭스의 출력포트에 서로 다른 위상차를 갖는 신호가 출력되며, 신호는 1 × 4 배열 안테나의 각각의 입력포트에 공급된다. 4개의 입력포트를 갖는 빔포밍 안테나는 총 4개의 빔을 형성한다. 빔포밍 안테나의 방사패턴을 분석하기 위해 각각의 입력포트에 신호를 스위칭하여 공급하였으며, 입력포트 1 ~ 4에 대한 개별적인 분석을 진행하였다. 제안한 빔포밍 안테나는 각각의 입력포트에 따라 각각 -12°, 40°, -40°, 12° 방향에서 주 빔이 형성되었다.

• 제어로봇시스템학회논문지
• /
• 제22권6호
• /
• pp.411-416
• /
• 2016

This paper presents the design methodology of an unknown input observer for Lipschitz nonlinear systems with unknown inputs in the framework of convex optimization. We use an unknown input observer (UIO) to consider both nonlinearity and disturbance. By deriving a sufficient condition for exponential stability in the linear matrix inequality (LMI) form, existence of a stabilizing observer gain matrix of UIO will be assured by checking whether the quadratic stability margin of the error dynamics is greater than the Lipschitz constant or not. If quadratic stability margin is less than a Lipschitz constant, the coordinate transformation may be used to reduce the Lipschitz constant in the new coordinates. Furthermore, to reduce the maximum singular value of the observer gain matrix elements, an object function to minimize it will be optimally designed by modifying its magnitude so that amplification of sensor measurement noise is minimized via multi-objective optimization algorithm. The performance of UIO is compared to a nonlinear observer (Luenberger-like) with an application to a flexible joint robot system considering a change of load and disturbance. Finally, it is validated via simulations that the estimated angular position and velocity provide true values even in the presence of unknown inputs.

• 대한전기학회논문지
• /
• 제32권11호
• /
• pp.394-403
• /
• 1983

본논문은 행렬비로 기미되는 다입.출력 연속시간 시스템에 대한 기준 모델형 적응제어에 관하여 고찰한다. 제어기는 monopoli - Narendra type 으로서 파라메타 적응칙에 시변리득행렬을 도입하였으며 가조정 제어기를 포함한 플랜트의 전달함수 행렬이 기준모델의 그것에 점차 따라가도록 한다. interactor 행렬에 대한 지식을 비롯한 약간의 가정하에서 단일 입.출력 시스템의 경우의 알고리즘이 적절하게 적용될 수 있음을 보인다. 적응칙의 수렴성은 안정도 이론을 이용하여 증명하며 전체 시스템의 안정성은 해석적인 고찰에 의해 보여 준다.

• 대한전기학회논문지
• /
• 제40권2호
• /
• pp.224-233
• /
• 1991

The multilayer expanson of single layer NN (Neural Network) was needed to solve the linear seperability problem as shown by the classic example using the XOR function. The EBP (Error Back Propagation ) learning rule is often used in multilayer Neural Networks, but it is not without its faults: 1)D.Rimmelhart expanded the Delta Rule but there is a problem in obtaining Ca from the linear combination of the Weight matrix N between the hidden layer and the output layer and H, wich is the result of another linear combination between the input pattern and the Weight matrix M between the input layer and the hidden layer. 2) Even if using the difference between Ca and Da to adjust the values of the Weight matrix N between the hidden layer and the output layer may be valid is correct, but using the same value to adjust the Weight matrixd M between the input layer and the hidden layer is wrong. Recognitron III was proposed to solve these faults. According to simulation results, since Recognitron III does not learn the three layer NN itself, but divides it into several single layer NNs and learns these with learning patterns, the learning time is 32.5 to 72.2 time faster than EBP NN one. The number of patterns learned in a EBP NN with n input and output cells and n+1 hidden cells are 2**n, but n in Recognitron III of the same size. [5] In the case of pattern generalization, however, EBP NN is less than Recognitron III.