• Journal of KIISE:Software and Applications
• /
• v.28 no.9
• /
• pp.603-611
• /
• 2001

This paper proposes a novel fast layer-by-layer algorithm that has better generalization capability. In the proposed algorithm, the weights of the hidden layer are updated by the target vector of the hidden layer obtained by least squares method. The proposed algorithm improves the learning speed that can occur due to the small magnitude of the gradient vector in the hidden layer. This algorithm was tested in a handwritten digits recognition problem. The learning speed of the proposed algorithm was faster than those of error back propagation algorithm and modified error function algorithm, and similar to those of Ooyen's method and layer-by-layer algorithm. Moreover, the simulation results showed that the proposed algorithm had the best generalization capability among them regardless of the number of hidden nodes. The proposed algorithm has the advantages of the learning speed of layer-by-layer algorithm and the generalization capability of error back propagation algorithm and modified error function algorithm.

• Ocean and Polar Research
• /
• v.42 no.3
• /
• pp.195-210
• /
• 2020

The observed time series from the Korea Ocean Research Stations (KORS) in the Yellow and East China Seas (YECS) have various sources of noise, including bio-fouling on the underwater sensors, intermittent depletion of power, cable leakage, and interference between the sensors' signals. Besides these technical issues, intricate waves associated with background tidal currents tend to result in substantial oscillations in oceanic time series. Such technical and environmental issues require a regionally optimized automatic quality control (QC) procedure. Before the achievement of this ultimate goal, we examined the approach of the Ocean Observatories Initiative (OOI)'s standard QC to investigate whether this procedure is pertinent to the KORS. The OOI QC consists of three categorized tests of global/local range of data, temporal variation including spike and gradient, and sensor-related issues associated with its stuck and drift. These OOI QC algorithms have been applied to the water temperature time series from the Ieodo station, one of the KORS. Obvious outliers are flagged successfully by the global/local range checks and the spike check. Both stuck and drift checks barely detected sensor-related errors, owing to frequent sensor cleaning and maintenance. The gradient check, however, fails to flag the remained outliers that tend to stick together closely, as well as often tend to mark probably good data as wrong data, especially data characterized by considerable fluctuations near the thermocline. These results suggest that the gradient check might not be relevant to observations involving considerable natural fluctuations as well as technical issues. Our study highlights the necessity of a new algorithm such as a standard deviation-based outlier check using multiple moving windows to replace the gradient check and an additional algorithm of an inter-consistency check with a related variable to build a standard QC procedure for the KORS.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.40 no.9
• /
• pp.597-604
• /
• 2016

A parallel algorithm of bi-conjugate gradient method was developed based on CUDA for parallel computation of the incompressible Navier-Stokes equations. The governing equations were discretized using splitting P2P1 finite element method. Asymmetric stenotic flow problem was solved to validate the proposed algorithm, and then the parallel performance of the GPU was examined by measuring the elapsed times. Further, the GPU performance for sparse matrix-vector multiplication was also investigated with a matrix of fluid-structure interaction problem. A kernel was generated to simultaneously compute the inner product of each row of sparse matrix and a vector. In addition, the kernel was optimized to improve the performance by using both parallel reduction and memory coalescing. In the kernel construction, the effect of warp on the parallel performance of the present CUDA was also examined. The present GPU computation was more than 7 times faster than the single CPU by double precision.

• Geophysics and Geophysical Exploration
• /
• v.17 no.4
• /
• pp.187-201
• /
• 2014

Compared with the separated inversion of electromagnetic (EM) and seismic data, a joint inversion using both EM and seismic data reduces the uncertainty and gives the opportunity to use the advantage of each data. Seismic fullwaveform inversion allows velocity information with high resolution in complicated subsurface. However, it is an indirect survey which finds the structure containing oil and gas. On the other hand, marine controlled-source EM (mCSEM) inversion can directly indicate the oil and gas using different EM properties of hydrocarbon with marine sediments and cap rocks whereas it has poor resolution than seismic method. In this paper, we have developed a joint EM inversion algorithm using a cross-gradient technique. P-wave velocity structure obtained by full-waveform inversion using plane wave encoding is used as structure constraints to calculate the cross-gradient term in the joint inversion. When the jointinversion algorithm is applied to the synthetic data which are simulated for subsea reservoir exploration, images have been significantly improved over those obtained from separate EM inversion. The results indicate that the developed joint inversion scheme can be applied for detecting reservoir and calculating the accurate oil and gas reserves.

• Journal of the Optical Society of Korea
• /
• v.17 no.2
• /
• pp.188-199
• /
• 2013

This paper presents a novel algorithm for nighttime detection of the lane markers painted on a road at night. First of all, the proposed algorithm uses neighborhood average filtering, 8-directional Sobel operator and thresholding segmentation based on OTSU's to handle raw lane images taken from a digital CCD camera. Secondly, combining intensity map and gradient map, we analyze the distribution features of pixels on boundaries of lanes in the nighttime and construct 4 feature sets for these points, which are helpful to supply with sufficient data related to lane boundaries to detect lane markers much more robustly. Then, the searching method in multiple directions- horizontal, vertical and diagonal directions, is conducted to eliminate the noise points on lane boundaries. Adapted Hough transformation is utilized to obtain the feature parameters related to the lane edge. The proposed algorithm can not only significantly improve detection performance for the lane marker, but it requires less computational power. Finally, the algorithm is proved to be reliable and robust in lane detection in a nighttime scenario.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.10
• /
• pp.1829-1834
• /
• 2006

In this raper in order to recover sharp object boundaries we propose a new efficient stereo matching algorithm in which window size is varied to the distance from the boundaries of object. To this end, the processing region is divided into small subregions with a same area and the disparities of the center pixels in the subregions are calculated using a area-based algorithm with multiple windows. From the this disparity map we can find the edges of the contracted objects. The disparities of original image are obtained using the gradient constraint that means the disparity of the center pixel is similar to the ones of the remaining pixels in the subregion. from the experimental results it is found that the proposed algorithm is very good for recovering sharp object boundaries compared to the similar different algorithm.

• Journal of ILASS-Korea
• /
• v.6 no.1
• /
• pp.35-43
• /
• 2001

Previous studies on image processing techniques for panicle size measurement usually have focused on a single panicle or weakly overlapped particles. In the present work, the image processing algorithm for particle size measurement has been improved to process heavily-overlapped spherical-particle images. The algorithm consists of two steps; detection of boundaries which separate the images of the overlapped panicles from the background and the panicle identification process. For the first step, Sobel operator (using gray-level gradient) and the thinning process was adopted, and compared with the gray-level thresholding method that has been widely adopted. In the second, Hough transform was used. Hough transform is the detection algorithm of parametric curves such as straight lines or circles which can be described by several parameters. To reduce the measurement error, the process of finding the true center was added. The improved algorithm was tested by processing an image frame which contains heavily overlapped spherical panicles. The results showed that both the performances of detecting the overlapped images and separating the panicle from them were improved.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.21 no.3
• /
• pp.612-621
• /
• 1996

In this paper, we propose the low sidelobe digital FIR doppler filter bank synthesis algorithm through the Gradient Descent method and it can be practially appliable to coherent pulse doppler radar signal processing. This algorithm shows the appropriate calculation of tap coefficients or zeros for FIR transversal fiter which has been employed in radar signal processor. The span of the filters in the filter bank be selected at the desired position the designer want to locate, and the lower sidelobe level that has equal ripple property is achieved than one for which the conventional weithtedwindow is used. Especially, when we implemented filter zeros as design parameters it is possible to make null filter gain at zero frequency intensionally that would be very efficient for the eliminatio of ground clutter. For the example of 10 tap filter synthesis, when filter coefficients or zeros are selected as design parameters the corresponding sidelobelevel is reducedto -70db or -100db respectively and it has good convergent characteristics to the desired sidelobe reference value. The accuracy ofapproach to the reference value and the speed of convergence that show the performance measure of this algorithm are tuned out with some superiority and the fact that the bandwidth of filter appears small with respect to one which is made by conventional weighted window method is convinced. Since the filter which is synthesized by this algorithm can remove the clutter without loss of target signal it strongly contributes performance improvement with which detection capability would be concerned.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.12 no.1
• /
• pp.14-18
• /
• 2002

The perceptron algorithm, which is one of a class of gradient descent techniques, has been widely used in pattern recognition to determine linear decision boundaries. However, it may not give desirable results when pattern sets are nonlinerly separable. A fuzzy version was developed to male up for the weaknesses in the crisp perceptron algorithm. This was achieved by assigning memberships to the pattern sets. However, still another drawback exists in that the pattern memberships do not consider class typicality of the patterns. Therefore, we propose a possibilistic approach to the crisp perceptron algorithm. This algorithm combines the linearly separable property of the crisp version and the convergence property of the fuzzy version. Several examples are given to show the validity of the method.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.3
• /
• pp.79-89
• /
• 1998

The Error Back-Propagation(EBP) algorithm is widely applied to train a multi-layer perceptron, which is a neural network model frequently used to solve complex problems such as pattern recognition, adaptive control, and global optimization. However, the EBP is basically a gradient descent method, which may get stuck in a local minimum, leading to failure in finding the globally optimal solution. Moreover, a multi-layer perceptron suffers from locking a systematic determination of the network structure appropriate for a given problem. It is usually the case to determine the number of hidden nodes by trial and error. In this paper, we propose a new algorithm to efficiently train a multi-layer perceptron. OUr algorithm uses stochastic perturbation in the weight space to effectively escape from local minima in multi-layer perceptron learning. Stochastic perturbation probabilistically re-initializes weights associated with hidden nodes to escape a local minimum if the probabilistically re-initializes weights associated with hidden nodes to escape a local minimum if the EGP learning gets stuck to it. Addition of new hidden nodes also can be viewed asa special case of stochastic perturbation. Using stochastic perturbation we can solve the local minima problem and the network structure design in a unified way. The results of our experiments with several benchmark test problems including theparity problem, the two-spirals problem, andthe credit-screening data show that our algorithm is very efficient.