• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.87-90
• /
• 2020

GPU의 발전과 함께 성장한 딥러닝(Deep Learning)은 영상 분류 문제에서 최고의 성능을 보이고 있다. 그러나 합성곱 신경망 기반의 모델을 깊게 쌓음에 따라 신경망의 표현력이 좋아짐과 동시에 때로는 학습이 잘되지 않고 성능이 저하되는 등의 부작용도 등장했다. 성능 향상을 방해하는 주요 요인 중 하나는, 차원감소 목적에 따라 필연적으로 정보 손실을 겪어야 하는 풀링 계층에 있다. 따라서 특성맵(Feature map)의 차원감소를 통해 얻게 되는 비용적 이득과 모델의 분류 성능 사이의 취사선택(Trade-off)이 존재한다. 그리고 이로부터 자유로워지기 위한 다양한 연구와 기법이 존재하는데 Spectral Pooling도 이 중 하나이다. 본 논문에서는 이산 푸리에 변환(Discrete Fourier Transform, DFT)을 이용한 Spectral Pooling에 대한 소개와, 해당 풀링의 성질을 통상적으로 사용되고 있는 Max Pooling과의 성능 비교를 통해 분석한다. 또한 영상 내 고주파수 부분에서 특히 더 강건하지 못하다는 맥스 풀링의 고질적인 문제점을, Spectral Pooling과의 하이브리드(Hybrid) 구조를 통해 어떻게 극복해나갈 것인지 그 가능성을 중심으로 실험을 수행했다.

• Journal of Forest and Environmental Science
• /
• v.13 no.1
• /
• pp.153-165
• /
• 1997

Hydroxyl radicals were detected and their qualitative yields were estimated by using chemiluminescence method and $\gamma$-irradiation technique in oxygen or chlorine dioxide radicals bleaching conditions. The correlation of hydroxyl radical formation and lignin model(Apocynol) or carbohydrate model($\alpha$-D-glucopyranose and methyl-$\beta$-D-glucopyranoside) degradation was studied in the presence of metal ion or without metal ion. The results showed that the presence of metal ions efficiently affected the formation of hydroxyl radicals in oxygen bleaching process, in the order of $Cu^{2+}$ > $Mn^{2+}$ > $Mg^{2+}$ > $Fe^{2+}$, and these metal gave also rise to the degradation of carbohydrate. But it was found that the addition of $100{\mu}m\;Mg^{2+}$ gave an efficient protection against carbohydrate degradation and suppressed the hydroxyl radical formation under oxygen bleaching conditions. And the presence of $Cu^{2+}$ had a detrimental effect on the stability of carbohydrates, whereas the addition of $3{\mu}m\;Mn^{2+}$ surprisingly had a small protective effect on methyl--$\beta$-D-glucopyranoside. In the $ClO_2$ radical bleaching conditions the hydroxyl radical expected to generate from water or substrates was not detected in the presence of metals.

• Journal of the Korea Convergence Society
• /
• v.12 no.2
• /
• pp.221-231
• /
• 2021

This paper identify factors of migration of employees' household who work for relocated public institutions. As a factors of migration, we consider individual and household characteristics, the gravity model of distance and population and so on. Considering discrete dependant variable and structure of data, we employ the logistic multilevel model and random intercept model. The result indicates employees' who are female, 30s and 40s, higher education level(PhD) and whose spouse are unemployed tend to transfer their residential registration to new city near relocated public institution. Regarding regional variable, the distance from employee's previous residential location and number of migration of prior year are statistically significant. Also the model indicate regional economy, educational and residential environment of new city influence employee's decision for transferring residential registration.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.21 no.1
• /
• pp.113-124
• /
• 2008

Distributed operation of overall structural design process, by which product optimization and process parallelization are simultaneously implemented, is presented in this paper. The database-interacted hybrid method, which selectively takes the accustomed procedure of the conventional method in the framework of the optimal design, is utilized here. The staged application of design constraints reduces the computational burden for large complex optimization problems. Two kinds of numeric and graphic processes are simultaneously implemented by concurrent engineering approach in the distributed environment of PC networks. The former is based on finite element optimization method and the latter is represented by AutoCAD using AutoLISP programming language. Numerical computation and database interaction on servers and graphic works on independent clients are communicated through message passing. The numerical experiments for some steel truss models show the validity and usability of the method. This study has sufficient adaptability and expandability, in that it is based on general methodologies and industry standard platforms.

• International Area Studies Review
• /
• v.12 no.3
• /
• pp.321-339
• /
• 2008

The purpose of this study to the analyze characteristics and purchasing activities of consumers by using the Multinomial Logit model, which is a well-known discrete selection model to explain and forecast consumers' selection activities(patterns). The study aims to determine the state of competition between National Brand and Private Band and how some demographic characters and marketing variables influence consumers' brand selections within the facial tissue market. Our analysis process includes reorganization of panel data(individuals' purchasing record at each point) to fit the purpose of our study as well as analysis of probability and influencing factors of consumers' brand selection at each point of purchases. The result showed that consumers at higher age and with higher income hold better probability to purchase National Brand. Likewise, locations also had considerable effect on selecting brand, and Private Brand was preferred in department store and discount stores. On the other hand, consumers loyal to National Brand reported higher probability to purchase if the product prices were higher while Private Brand buyers were more promoted the purchase under price discount.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.18 no.2
• /
• pp.117-123
• /
• 2018

In this study, we suggest a fast image reconstruction scheme using Poisson equation from image gradient domain. In this approach, using the Poisson equation, a guided vector field is created by employing source and target images within a selected region at the first step. Next, the guided vector is used in generating the result image. We analyze the problem of reconstructing a two-dimensional function that approximates a set of desired gradients and a data term. The joined data and gradients are able to work like modifying the image gradients while staying close to the original image. Starting with this formulation, we have a screened Poisson equation known in physics. This equation leads to an efficient solution to the problem in FFT domain. It represents the spatial filters that solve the two-dimensional screened Poisson model and shows gradient scaling to be a well-defined sharpen filter that generalizes Laplace sharpening. We demonstrate the results using a discrete cosine transformation based this Poisson model.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.7 no.4
• /
• pp.598-603
• /
• 2003

This paper presents a sub-band feature extraction approach in which the feature extraction method in the individual frequency sub-bands is determined in terms of speech recognition accuracy. As in the multi-band paradigm, features are extracted independently in frequency sub-regions of the speech signal. Since the spectral shape is well structured in the low frequency region, the all pole model is effective for feature extraction. But, in the high frequency region, the nonparametric transform, discrete cosine transform is effective for the extraction of cepstrum. Using the sub-band specific feature extraction method, the linguistic information in the individual frequency sub-bands can be extracted effectively for automatic speech recognition. The validity of the proposed method is shown by comparing the results of speech recognition experiments for our method with those obtained using a full-band feature extraction method.

• Journal of the Korean Institute of Telematics and Electronics
• /
• v.26 no.4
• /
• pp.98-105
• /
• 1989

Various models incorporating Human Visual System (HVS) with the Hadamard transform (HT) represented by Walsh functions are considered. Using the exact frequency components of HT basis functions, the optimum modulation transfer function (MTF) which has a higher peak frequency than DCT schemes is obtimum modulation transfer function (MTF) which has a higher peak frequency than DCT schemes is obtained analytically and visually. The main criterion, for error measurement, is errors at the block boundaries which is an important factor in transform coding. The scheme which has no inverse HVS is proposed. It causes some degradation of image data but it is insignigicant. Crossing area of 4 blocks is equalized by the HVS weighting coefficients. The HVS weighted coding results in perceptually higher quality images compared with the unweighted scheme.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2021.06a
• /
• pp.103-103
• /
• 2021

실험실에서의 파랑생성에 흔히 사용되는 피스톤형 조파장치는 수심에 따라 유속이 동일하게 생성된다는 제약이 있어 주로 천해파의 생성에 적합한 것으로 알려져 있다. 본 연구에서는 이러한 제약조건 없이 다양한 유속분포의 파형을 생성하는 수직 다열화된 조파장치가 개발되었다. 우선, 수심방향으로 이산화된 각 패들(paddle)의 스트로크에 대해 선형해석해가 유도되었다. 개발된 해석해는 패들의 수 및 유속분포에 따라 기존의 피스톤형 혹은 플랩형 조파장치 해석해로 근사함이 밝혀짐으로써 포괄적으로 활용될 수 있음이 확인되었다. 즉 개발된 해석해를 활용하면 선택적으로 피스톤형 및 플랩형 조파성능이 구현될 수 있다. 더불어 개발된 해석해는 다상유체의 내부파 생성에도 확정되어 적용가능함이 확인되었다. 다음으로, 개발된 조파장치를 수치적으로 구현하였다. 오픈소스 3차원 수치모형인 OpenFOAM 중, 두 개 이상의 불연속 및 비압축 유체에 대한 Navier-Stokes 방정식을 해결하는 수치 모듈을 사용하여 제안된 수직다열화된 조파장치의 성능이 평가되었다. 이때 동적격자모델(olaDyMFlow)을 결합함으로써 개발된 조파장치 움직임이 물리적 조파장치와 흡사하도록 수치적으로 구현하였다. 모의결과, 여러 개의 다열화된 패들이 층류 흐름 조건에서 심해파를 효율적으로 생성시키고, 중간수심 파랑조건에서는 제안된 조파장치가 상대적으로 덜 유리함을 확인할 수 있었다. 마지막으로 공기, 기름 및 물 등 3상의 흐름조건에서 단 두 개의 패들을 활용하여 각각 내부파 및 표면파를 생성하되었으며, 모의 결과는 해석해과 비교됨으로써 개발된 조파장치의 성능이 검증되었다.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.16 no.6
• /
• pp.1153-1160
• /
• 2021

This paper proposes a routing algorithm that determines the optimal path using deep reinforcement learning in software-defined networks. The deep reinforcement learning model for learning is based on DQN, the inputs are the current network state, source, and destination nodes, and the output returns a list of routes from source to destination. The routing task is defined as a discrete control problem, and the quality of service parameters for routing consider delay, bandwidth, and loss rate. The routing agent classifies the appropriate service class according to the user's quality of service profile, and converts the service class that can be provided for each link from the current network state collected from the SDN. Based on this converted information, it learns to select a route that satisfies the required service level from the source to the destination. The simulation results indicated that if the proposed algorithm proceeds with a certain episode, the correct path is selected and the learning is successfully performed.