• Journal of Digital Contents Society
• /
• v.3 no.2
• /
• pp.197-209
• /
• 2002

If we reduce the number of vertexes to decrease bit rate in polygon-based shape coding, the distortion of approximated contour increases rapidly. On the other hand, if we reduce the distortion, the number of vertexes increases rapidly and many bits are required to encode the vertexes. To improve this problem, in this paper we propose the approximation vertex search method. The encoder in the proposed method searches the type of distortion patterns that is the most similar to the shape which polygon edge and contour segment form and then encodes it. And then, the decoder mathematically finds the approximated vertexes from decoded distortion pattern information. Therefore, the proposed algorithm results in encoding many vertexes at a low bit rate and having the smoother shape than conventional method. As shown in computer simulation, the proposed method has less distortion than conventional method. It costs less bit rate by $10{\sim}20%$ than conventional algorithm in same distortion.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.14 no.3
• /
• pp.339-348
• /
• 2001

More often a commercial package for the structural analysis is necessary in the structural optimum design. In this case the task of combining the package with an optimization program must be required, hut it is not so simple because interchanging some data between them is difficult. Sequential approximate optimization is currently used as a natural way to overcome the hard task. If sequential approximate optimization has wide side constraints that the lower limit of design variables is very small and their upper limit is very large, it is not so easy to obtain approximated functions accurately for the whole design domain. This paper proposes a sequential design domain method, which is very useful to carry out sequential approximate optimization in this case. In this paper, the response surface methodology is used to obtain approximated functions and the orthogonal array is used for design of experiments. The sequential approximate optimization of 3-bar and 10-bar trusses is demonstrated to verify the reliability of the sequential design domain method.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.31 no.3B
• /
• pp.293-303
• /
• 2011

An efficient diagonalized approximate factorization algorithm (DAF) is developed for the solution of three-dimensional incompressible viscous flows. The pressure-based, artificial compressibility (AC) method is used for calculating steady incompressible Navier-Stokes equations. The AC form of the governing equations is discretized in space using a second-order-accurate finite volume method. The present DAF method is applied to derive a second-order accurate splitting of the discrete system of equations. The primary objective of this study is to investigate the computational efficiency of the present DAF method. The solutions of the DAF method are evaluated relative to those of well-known four-stage Runge-Kutta (RK4) method for fully developed and developing laminar flows in curved square ducts and a laminar flow in a cavity. While converged solutions obtained by DAF and RK4 methods on the same computational meshes are essentially identical because of employing the same discrete schemes in space, both algorithms shows significant discrepancy in the computing efficiency. The results reveal that the DAF method requires substantially at least two times less computational time than RK4 to solve all applied flow fields. The increase in computational efficiency of the DAF methods is achieved with no increase in computational resources and coding complexity.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.32 no.1B
• /
• pp.21-27
• /
• 2012

The HLLL scheme, proposed by T. Linde, determines all the wave speeds from the initial states because the middle wave is evaluated by the introduction of a generalized entropy function. The scheme is considered a genuine successor to the original HLL scheme because it is completely separated form the Roe's linearization scheme unlike the HLLE scheme and does not rely on the exact solution unlike the HLLC scheme. In this study, a numerical model was configured by the HLLL scheme with the total energy as a generalized entropy function to solve governing equations, which are the one-dimensional shallow water equations without source terms and with an additional conserved variable relating a concentration. Despite the limitations of the first order solutions, results to three cases with the exact solutions were generally accurate. The HLLL scheme appeared to be superior in comparison with the other HLL-type schemes. In particular, the scheme gave fairly accurate results in capturing the front of wetting and drying. However, it revealed shortcomings of more time-consuming calculations compared to the other schemes.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.40 no.1
• /
• pp.130-138
• /
• 2003

The classical image reconstruction for stripmap-mode Synthetic Aperture Radar is the Range-Doppler algorithm. When the spotlight-mode SAR system was envisioned, Range-Doppler algorithm turned out to fail rapidly in this SAR imaging modality. Thus, what is referred to as Polar format algorithm, which is based on the Plane wave approximation, was introduced for imaging from spotlight-mode SAR raw- data. In this paper, we have studied for the raw data processing schemes in the spotlight-mode Synthetic Aperture Radar. We apply the Wavefront Reconstruction scheme that does not utilize the approximation in spotlight-mode SAR imaging modelity, and compare the performance of target imaging with the Polar format inversion scheme.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.34 no.7
• /
• pp.18-27
• /
• 2006

The conventional reliability based design optimization(RBDO) methods require high computational cost compared with the deterministic design optimization(DO) methods, therefore it is hard to apply directly to large-scaled problems such as an aerodynamic shape design optimization. In this study, to overcome this computational limitation the efficient RBDO procedure with the two-point approximation(TPA) and adjoint sensitivity analysis is proposed, that the computational requirement is nearly the same as DO and the reliability accuracy is good compared with that of RBDO. Using this, the 3-D aerodynamic shape design optimization is performed very efficiently.

• The KIPS Transactions:PartB
• /
• v.14B no.5
• /
• pp.377-382
• /
• 2007

This paper proposes a training algorithm for global optimization of the parameters of radial basis function networks. Since conventional training algorithms usually perform only local optimization, the performance of the network is limited and the final network significantly depends on the initial network parameters. The proposed hybrid simulated annealing algorithm performs global optimization of the network parameters by combining global search capability of simulated annealing and local optimization capability of gradient-based algorithms. Via experiments for function approximation problems, we demonstrate that the proposed algorithm can find networks showing better training and test performance and reduce effects of the initial network parameters on the final results.

• Journal of the Korea Computer Graphics Society
• /
• v.17 no.3
• /
• pp.33-42
• /
• 2011

We present a new technique for removing interior knots of parametric B-spline curves. An initial curve is constructed by continuous $L_{\infty}$ approximation proposed by Eck and Hadenfeld. We employ Hausdorff distance to measure the shape difference between the original curve and the initial one. The final curve is obtained by minimizing their Hausdorff distance. We demonstrate the effectiveness of our technique with experimental results on various types of planar and spatial curves.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2019.05a
• /
• pp.98-98
• /
• 2019

중력류 또는 밀도류는 주변 유체에 비해 상대적으로 밀도가 큰 유체가 밀도차에 의한 추진력으로 흐르는 것이다. 중력류의 수치모델링에는 두 가지 어려움이 있다. 즉, 적합한 지배방정식을 구성하여 적용하는 것 그리고 난류의 영향을 합리적으로 반영하는 것이다. 기존 중력류 해석을 위한 지배방정식들은 유체의 연속방정식과 운동량 방정식 그리고 밀도 또는 농도의 이송방정식을 조합하여 구성된다. 이들 지배방정식을 이용한 연구들은 대부분 두 유체 사이의 밀도차가 충분히 작아서 밀도 변동(variations)의 영향은 오로지 부력항에서만 유지된다는 Boussinesq 근사에 근거를 둔다. 그리고 이송방정식에서 밀도 또는 농도의 확산계수을 점성계수의 함수로 표현하기 위해서 Schmidt 수를 이용한다. 수치모델링에서 Schimdt 수는 상수값을 적용하지만, 이 값은 밀도의 연직방향 경사에 근거한 부력빈도(buoyancy frequency)와 난류량의 따라 큰 차이를 보이는 것으로 알려져있다. 한편, 표준 통계학적 난류모델과 벽함수를 적용한 수치모델링은 초기 중력에 의해서 무너지는(slumping) 단계를 넘어 관성력으로 추진되는 단계와 점성 효과가 지배적인 단계에서는 정확도에 현저히 낮아지기 때문에 대부분 큰와모의(large-eddy simulation, LES) 또는 DNS(direct numerical simulation)수준의 고해상도(high-resolution) 해석기법을 적용하여 공학적인 문제에 적용하는 데는 한계가 있다. 이 연구에서는 Boussinesq 근사와 Schmidt 수를 사용하지 않으며, LES 보다 적용이 용이한 DES (detached-eddy simulation)기법을 조합한 다상흐름 수치모델을 적용하여 중력류를 해석을 시도하였다. 수치해석결과를 실험값과 함께 기존 수치모델링 기법으로 구한 수치해와 비교분석하여 이 연구에서 개발 및 적용된 수치모델링 기법의 적용성을 평가한다.

• Journal of Broadcast Engineering
• /
• v.26 no.2
• /
• pp.125-131
• /
• 2021

In recent years, Convolutional Neural Networks (CNNs) have achieved outstanding performance in the fields of computer vision such as image classification, object detection, visual quality enhancement, etc. However, as huge amount of computation and memory are required in CNN models, there is a limitation in the application of CNN to low-power environments such as mobile or IoT devices. Therefore, the need for neural network compression to reduce the model size while keeping the task performance as much as possible has been emerging. In this paper, we propose a method to compress CNN models by combining matrix decomposition methods of LR (Low-Rank) approximation and CP (Canonical Polyadic) decomposition. Unlike conventional methods that apply one matrix decomposition method to CNN models, we selectively apply two decomposition methods depending on the layer types of CNN to enhance the compression performance. To evaluate the performance of the proposed method, we use the models for image classification such as VGG-16, RestNet50 and MobileNetV2 models. The experimental results show that the proposed method gives improved classification performance at the same range of 1.5 to 12.1 times compression ratio than the existing method that applies only the LR approximation.