• 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.

• Bulletin of the Korean Chemical Society
• /
• v.35 no.1
• /
• pp.165-172
• /
• 2014

The distribution of basic nitrogen-containing compounds in three vacuum gas oils (VGOs) with different boiling ranges and their dewaxed oils from the lube base oil refining unit of a refinery were characterized by positive-ion electrospray ionization (ESI) Fourier transform-ion cyclotron resonance mass spectrometry (FT-ICR MS). It turned out that the composition of basic nitrogen compounds in the samples varied significantly in DBE and carbon number, and the dominant basic N-containing compounds in these oil samples were N1 class species. $N_1O_1$, $N_1O_2$, and $N_2$ class species with much lower relative abundance were also identified. The composition of basic nitrogen compounds in VGOs and dewaxed VGOs were correlated with increased boiling point and varied in DBE and carbon numbers. The comparison of the analytical results between VGOs and dewaxed VGOs indicated that more basic N-containing compounds in VGO with low carbon number and small molecular weight tend to be removed by solvent refining in lube base oil processing.

• Journal of the Institute of Convergence Signal Processing
• /
• v.20 no.4
• /
• pp.199-204
• /
• 2019

In the boundary layer, vortical system has been analyzed by the statistical methods to identify the vortex interaction. However, there are the limitations in explaining vortices by the mean velocity or the standard deviation. This paper proposed a method to establish a frequency analysis by Fourier transform in order to simultaneously investigate various scale vortices. For this purpose, the flow visualization conducted to reveal a standing vortex, a hairpin vortex and a wake region around a hemisphere attached on a flat plate in a water channel. In addition, the velocity where the hairpin vortex was being generated in the wake region was measured by a hot-film anemometer. To observe changes in the vortex interaction, suction was applied through a hole in front of the hemisphere. For the evaluation of the proposed frequency analysis, the existing statistical results were compared to the frequency analysis that corresponds to the qualitative results of the flow visualization.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• v.9 no.2
• /
• pp.451-454
• /
• 2005

Recently, many methods to analyze signal have been proposed and representative methods are the Fourier transform and wavelet transform. In these methods, the Fourier transform represents signal with combination cosine and sine at all locations in the frequency domain. However, it doesn't provide time information that particular frequency occurs in signal and depends on only the global feature of the signal. So, to improve these points the wavelet transform which is capable of multiresolution analysis has been applied to many fields such as speech processing, image processing and computer vision. And the wavelet transform, which uses changing window according to scale parameter, presents time-frequency localization. In this paper, we proposed a new approach using a wavelet of cosine and sine type and analyzed features of signal in a limited point of frequency-time plane.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.2006-2009
• /
• 2005

This paper presents a method for registering images using phase correlation technique in fourier domain, hough transform and multi-resolution wavelet. To register images, source and input images are transformed to wavelet domain. An angular transition can be obtained by applying hough transform technique followed by phase correlation. Then we apply phase correlation technique to find x-axis and y-axis transition. We apply wavelet transform to reduce processing time and also use its coefficients as edge information instead of canny detector. With multi-resolution property of wavelet transform, registration time can be greatly reduced. After we get all transition parameters, we transform the input images according to these parameters. Then, we compose and blend all images into a new large image with details of all source images. From our experiment, we can find the accurate transition both x-y translation and angular transition with less error.

• Proceedings of the Korean Society of Marine Engineers Conference
• /
• 2002.05a
• /
• pp.219-223
• /
• 2002

Voltage and current signals during impulse tests on transformer are treated as non-stationary signals. A new method incorporating signal-processing method such as Wavelets and courier transform is proposed for failure identification. It is now possible to distinguish failure during impulse tests. The method is experimentally validated on a transformer winding. The wavelet transforms enables the detection of the time of occurrence of switching or failure events. After establishing the time of occurrence, the original waveform is split into two or more sections. The wavelet transform has ability to analysis the failure signal on time domain as well as frequency domain. Therefore, the wavelet transform is superior than courier transform to analysis the failure signal. In this paper, the fact was proved by real data which was achieved.

• Journal of Korea Multimedia Society
• /
• v.19 no.10
• /
• pp.1775-1781
• /
• 2016

Generally, in the underwater target recognition, feature vectors are extracted from the target signal utilizing spatial information according to target shape/material characteristics. In addition, various signal processing techniques have been studied to extract feature vectors which are less sensitive to the location of the receiver. In this paper, we synthesized active echo signals using 3-dimensional highlight distribution. Then, Fractional Fourier transform was applied to echo signals to extract signal features. For the performance verification, classification experiments were performed using backpropagation and probabilistic neural network classifiers based on single aspect and multi-aspect method. As a result, we obtained a better recognition result using proposed feature extraction and multi-aspect based method.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.27 no.11
• /
• pp.1534-1541
• /
• 2003

A new window-based, user-friendly fringe analysis system is developed, especially for analyzing noisy interferograms. The system integrates three major techniques, that is fringe tracking, Phase shifting, and Fourier transform, into a single shell by employing a unified procedure. Since the system is made in a modular fashion and all processing modules can be shared for any technique, a user can select necessary modules and easily edit the applying order of them based on the user's analysis strategy, which should be changed depending on the noise level of the image. The system provides a high-level GUI and a variety of image handling tools and therefore users can easily access the system and produce the optimal results without giving up in the middle of a process even for severely noise-contaminated interferometric images.

• Journal of the Korean Society of Safety
• /
• v.13 no.4
• /
• pp.256-263
• /
• 1998

In this paper, we propose a volume holographic fingerprint recognition system based on optical correlator for personal identification. Optical correlator has high speed and parallel processing characteristics of optics. Matched filters are recorded into a volume hologram that can store data with high density, transfer them with high speed, and select a randomly chosen data element. The multiple reference images of database are prerecorded in a photorefractive crystal in the form of Fourier transform images, simply by passing the image displayed in a spatial light modulator through a Fourier transform lens. The angular multiplexing method for multiple holograms of database can be achieved by rotating the crystal by use of a step motor. Experimental results show that the proposed system can be used for the security verification system.

• Electrical & Electronic Materials
• /
• v.11 no.10
• /
• pp.66-71
• /
• 1998

Plasma polymerized hexamethyldisiloxane (PPHMDSO) thin films were produced using an electrode capacitively coupled apparatus. Fourier transform infrared spectroscopy analysis indicated that the thin film spectra are composed not only of the corresponding monomer bands but also of several new bands. Auger electron spectroscopy analysis indicated that the permeation depth of aluminum into the films is ca. 30nm when top electrode is deposited by evaporation aluminum. The increase of relative dielectric constant and decrease of dielectric loss tangent with the discharge power is originated from high cross-link of the films.