• Journal of Korea Multimedia Society
• /
• v.24 no.11
• /
• pp.1552-1559
• /
• 2021

Convolution neural networks (CNNs) show notable performance in image processing and are used as representative core models. CNNs extract and learn features from large amounts of train dataset. In general, it has a structure in which a convolution layer and a fully connected layer are stacked. The core of CNN is the convolution layer. The size of the kernel used for feature extraction and the number that affect the depth of the feature map determine the amount of weight parameters of the CNN that can be learned. These parameters are the main causes of increasing the computational complexity and memory usage of the entire neural network. The most computationally expensive components in CNNs are fully connected and spatial convolution computations. In this paper, we propose a Fourier Convolution Neural Network that performs the operation of the convolution layer in the Fourier domain. We work on modifying and improving the amount of computation by applying the fast fourier transform method. Using the MNIST dataset, the performance was similar to that of the general CNN in terms of accuracy. In terms of operation speed, 7.2% faster operation speed was achieved. An average of 19% faster speed was achieved in experiments using 1024x1024 images and various sizes of kernels.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.65 no.10
• /
• pp.1731-1737
• /
• 2016

In this paper, a method is proposed to recognize human actions as nonverbal expression; the proposed method is composed of two steps which are action representation and action recognition. First, MHI(Motion History Image) is used in the action representation step. This method includes segmentation based on depth information and generates spatio-temporal templates to describe actions. Second, CNN(Convolution Neural Network) which includes feature extraction and classification is employed in the action recognition step. It extracts convolution feature vectors and then uses a classifier to recognize actions. The recognition performance of the proposed method is demonstrated by comparing other action recognition methods in experimental results.

• Proceedings of the KIEE Conference
• /
• 2005.10b
• /
• pp.532-535
• /
• 2005

These days, many image processing techniques have been studied for effective image compression. Among those, 2D image filtering is widely used for 2D image processing. The 2D image filtering can be implemented by performing ID linear filtering separately in the direction of horizontal and vertical. Efficiency of image compression depends on what filtering method is used. Generally, circular convolution is widely used in the 2D image filtering for image processing. However it doesn't consider correlations at the region of image boundary, therefore filtering can not be performed effectively. To solve this problem. I proposed new convolution technique using Symmetric-Mirroring convolution, satisfying the 'alias-free' and 'error-free' requirement in the reconstructed image. This method could provide more effective performance than former compression methods. Because it used very high correlative data when performed at the boundary region. In this paper, pre-processing filtering in H.264 codec was adopted to analyze efficiency of proposed filtering technique, and the simulator developed by Matlab language was used to examine the performance of the proposed method.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.41 no.12
• /
• pp.1692-1699
• /
• 2016

This paper deals with the problem of time delay estimation using de-convolution. Two approaches, conjugate gradient method and the total lease square method have been presented to solve the de-convolution problem. Numerical simulation demonstrates the superior performance of the proposed methods over the conventional GCC based algorithms and FIR filter method.

• Structural Engineering and Mechanics
• /
• v.29 no.4
• /
• pp.411-422
• /
• 2008

In the present study, free vibration analysis of thick annular plates is analyzed by discrete singular convolution method. The Mindlin plate theory is employed. The material is isotropic, homogeneous and obeys Hook's law. In this paper, discrete singular convolution method is used for discretization of equations of motion. Axisymmetric frequency values are presented illustrating the effect of radius ratio and thickness to radius ratio of the annular plate. The influence of boundary conditions on the frequency characteristics is also discussed. Comparing results with those in the literature validates the present analysis. It is shown that the obtained results are very accurate by this approach.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.16 no.5 s.96
• /
• pp.447-454
• /
• 2005

JEC method for Debye medium is required more memory resources and long calculation time than already well-known method such as RC method. It has been observed that JEC method would be converted to a memory effcient method by a change of discrete convolution integral range. The modified JEC method proposed here requires memory and calculation time similar to RC method, while it has a same or a smaller dispersion error than conventional methods, RC and JEC.

• Journal of Environmental Science International
• /
• v.24 no.4
• /
• pp.481-494
• /
• 2015

Groundwater level hydrographs from observation wells in Jeju island clearly illustrate distinctive features of recharge showing the time-delaying and dispersive process, mainly affected by the thickness and hydrogeologic properties of the unsaturated zone. Most groundwater flow models have limitations on delineating temporal variation of recharge, although it is a major component of the groundwater flow system. Recently, a convolution model was suggested as a mathematical technique to generate time series of recharge that incorporated the time-delaying and dispersive process. A groundwater flow model was developed to simulate transient groundwater level fluctuations in Pyoseon area of Jeju island. The model used the convolution technique to simulate temporal variations of groundwater levels. By making a series of trial-and-error adjustments, transient model calibration was conducted for various input parameters of both the groundwater flow model and the convolution model. The calibrated model could simulate water level fluctuations closely coinciding with measurements from 8 observation wells in the model area. Consequently, it is expected that, in transient groundwater flow models, the convolution technique can be effectively used to generate a time series of recharge.

• The Journal of the Acoustical Society of Korea
• /
• v.29 no.1
• /
• pp.1-9
• /
• 2010

The mutual radiation impedance was calculated using the spatial convolution in the cylindrical structure. The Cartesian coordinate was transformed into the cylindrical coordinate using the spatial convolution for the cylindrical array structure. This method cannot consider the cylindrical baffle, but can reduce the computation time. The error for not considering the cylindrical baffle was analyzed by the comparison of the spatial convolution method with the quadruple integration method in the cylindrical structure. The mutual radiation resistance in the cylindrical structure was compared with the one in the planar baffle. Based on two kinds of the comparison, we presented the error of the suggesting method in this paper, confirming that the spatial convolution method could be applied to compute the mutual radiation impedance in the cylindrical structure at certain conditions.

• Journal of IKEEE
• /
• v.20 no.3
• /
• pp.303-306
• /
• 2016

In this paper, we propose a method to accelerate convolutional neural network by utilizing a GPGPU. Convolutional neural network is a sort of the neural network learning features of images. Convolutional neural network is suitable for the image processing required to learn a lot of data such as images. The convolutional layer of the conventional CNN required a large number of multiplications and it is difficult to operate in the real-time on the embedded environment. In this paper, we reduce the number of multiplications through Winograd convolution operation and perform parallel processing of the convolution by utilizing SIMT-based GPGPU. The experiment was conducted using ModelSim and TestDrive, and the experimental results showed that the processing time was improved by about 17%, compared to the conventional convolution.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.24 no.11
• /
• pp.17-22
• /
• 2010

This paper presents a unique method for simulating permanent magnet motors without time-consuming numerical methods used in the conventional magnetic circuit method. The conventional method gives us average values like torque and power over specified periods of time, but it is usually very difficult and time-consuming to obtain instantaneous characteristics like cogging torques and torque ripples. The convolution operations method we present, however, considers relative angle variations of stator magnetic circuits and rotor magnetic circuits. As a result, it makes uses of instantaneous values possible. The authors compare the new method with the coventional method and verify that calculating cogging torque values and back-emf values is possible with the proposed new convolution method.