• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.479-487
• /
• 2021

In this paper, we propose an Sound Event Detection (SED) model using self-training based on a noisy student model. The proposed SED model consists of two stages. In the first stage, a mean-teacher model based on an Residual Convolutional Recurrent Neural Network (RCRNN) is constructed to provide target labels regarding weakly labeled or unlabeled data. In the second stage, a self-training-based noisy student model is constructed by applying different noise types. That is, feature noises, such as time-frequency shift, mixup, SpecAugment, and dropout-based model noise are used here. In addition, a semi-supervised loss function is applied to train the noisy student model, which acts as label noise injection. The performance of the proposed SED model is evaluated on the validation set of the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 Challenge Task 4. The experiments show that the single model and ensemble model of the proposed SED based on the noisy student model improve F1-score by 4.6 % and 3.4 % compared to the top-ranked model in DCASE 2020 challenge Task 4, respectively.

• Applied Chemistry for Engineering
• /
• v.30 no.4
• /
• pp.493-498
• /
• 2019

In polymer exchange membrane fuel cells, it is crucial to fabricate a highly active and thin Pt catalyst layer for the smooth mass transport of dissolved oxygen and water. Although a highly loaded platinum (Pt) catalyst based on the hydrothermal synthesis has been reported in several studies, its growing behaviors and kinetics were yet to be understood. In this study, we investigated the growth of Pt crystal in suspension after the reduction step depending on a stirring time and evaluated the electrochemical activity. For only a couple of hours in the early stage, Pt colloids were adsorbed on the Pt-carbon catalyst and the Pt crystal was grown. After that, the small Pt colloid was formed by another nucleation step, which did not involve the growth of Pt crystal. We reveal that the Pt-Carbon catalyst with stirring for 6 h showed a high activity toward the oxygen reduction reaction.

• Membrane Journal
• /
• v.28 no.6
• /
• pp.379-387
• /
• 2018

This study was conducted to improve the membrane characteristics and performance by increasing hydrophilicity by adding additives to the ultrafiltration polysulfone (PSf) hollow fiber membrane. The mixed matrix membranes (MMMs) were prepared by dispersing 15 nm of fumed silica (FS) in the spinning solution at 0.1, 0.3 and 0.5 wt%. SEM analysis was carried out to confirm the cross-section and surface condition. It was confirmed that mean pore radius of the hollow fiber increased by 4 nm as FS was added. In addition, contact angle measurement was carried out for the hydrophilicity analysis of hollow fiber membranes, and it was confirmed that the hydrophilicity of MMMs were increased by adding of FS. In the case of water permeability, the membrane including FS showed 91~96 LMH and showed 5~11% more increase than PSf membrane. In the antifouling performance test, relative flux reduction ratios of FS mixed hollow fiber membranes were lower than that of PSf membranes, and it was confirmed that increase of hydrophilicity hinders adsorption of hydrophobic BSA on the membrane surface.

• Korean Journal of Optics and Photonics
• /
• v.35 no.4
• /
• pp.157-163
• /
• 2024

We developed a 3-channel fiber laser with a common seed and a phase control system for laser beam combining through a diffractive optical element. Beam combining was performed by adjusting the angles of the beams incident on the diffractive optical elements, and the phase of each beam was controlled to maximize the intensity of the combined laser beam. The power of the 3-channel laser before passing through the diffractive optical elements is about 65 mW. The power of the combined beam varied between 2.9 mW and 48.3 mW depending on the phase change of each channel. Through phase control, the output of the combined beam can be maintained at 42 mW for more than 91.8% of the total time. It is expected that higher combining efficiency can be achieved by improving the transmittance of the diffractive optical elements and the performance of the phase control system.

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.4
• /
• pp.443-449
• /
• 2019

To protect the patient's internal organs when a patient with cardiovascular disease occurs, it is important to reduce the elapsed time by providing emergency medical services. Decisions for conducting cardiopulmonary resuscitation are mainly made using the carotid palpation method, which directs the pulse of the carotid artery, which can diagnose the patient's condition according to one's own subject and cause cerebral blood flow to be blocked by excessive pressure in the carotid due to the weaken cardiopulmonary function. In this study, we developed a multimodal patch-type probe based on multi-channel ultrasound Doppler pairs and oxygen saturation measurement modules which can monitor cardiopulmonary functions. From the in-vivo experiments, the developed probe can be utilized as a novel tool that can increase the survival rate of cardiovascular disease patients by objectively monitoring the cardiopulmonary function of the patient quantitatively and promptly in an emergency situation.

• Proceedings of the Korea Air Pollution Research Association Conference
• /
• 2003.05b
• /
• pp.417-418
• /
• 2003

2002년 10월과 11월에는 가을 추수 후 소각이 대기질에 미치는 영향을 조사하고자 광주과학기술원 내에서 에어로졸과 대기 복사 집중 측정 기간을 가졌다. 12시간 또는 일평균 자료만을 제공하는 에어로졸 화학적 특성의 측정과 달리 자외선 영역의 다파장 회전차폐판 복사계 (Ultraviolet Multi-filter Rotating Shadowband Radiometer)를 이용한 에어로졸 광학 깊이(aerosol optical depth)는 1분 간격의 직달 일사량 자료로부터 에어로졸 복사적 특성을 시간에 따른 변화를 볼 수 있다는 장점을 가진다. (중략)

• The Science & Technology
• /
• no.9 s.424
• /
• pp.20-22
• /
• 2004

"이제 미지의 영역은 뇌뿐입니다. 영상장치의 수준에 따라 각국의 연구성과는 달라질 겁니다." PET(양전자 방출 단층촬영 장치)를 세계 최초로 개발, 뇌영상 연구분야의 세계 3대 석학 가운데 한 사람으로 꼽히는 UC어바인 조장희(68) 교수는 쩌렁쩌렁한 목소리에 기백이 넘쳤다. 고희를 앞둔 나이가 무색할 정도였다. 한국과학기술원과 광주과학기술원의 초빙 석좌교수 시절(1978~98년)의 거침없는 성격은 변함이 없었고, 뇌과학의 미래에 대한 그의 확신은 여전했다. 조 교수는 자신의 마지막 연구 인생을 조국에서 불태울 수 있는 기회를 잡았다. 가천의대가 조 교수를 영입해 370억 원을 들여 최첨단 뇌과학연구소를 세우기로 한 것이다. 조 교수에게는 15년간 연봉 30만 달러를 보장하는 파격적인 조건을 제시했다. 조 교수가 맡게 될 뇌과학영상연구소는 뇌 속에서 벌어지는 미세한 현상을 손바닥 보듯 들여다볼 수 있는 방법과 장치를 개발하게 된다. 가장 우선적인 목표는 MRI(자기공명 영상장치)와 PET를 합친 영상장치의 세계 최초 개발이다.

• Journal of Korean Society of Water and Wastewater
• /
• v.32 no.6
• /
• pp.479-485
• /
• 2018

Large amounts of oily wastewater discharged from various industrial operations (petroleum refining, machinery industries and chemical industries) cause serious pollution in the aquatic environment. Although dissolved air flotation (DAF) separating oil pollutants using microbubbles represents current practice, bubble size cannot be selectively controlled, and lots of power is required to generate microbubbles. Therefore, to investigate performance of the DAF process, this study examined the distribution of different sizes of microbubbles resulting from changes in physical shear force via modifying shapes of a slit-nozzle without an additional power supply. Three types of slit-nozzles (different angle, shape and length of the slit-nozzle) were used to analyze the distribution of bubble size. At a slit angle of $60^{\circ}$, shear force was 4.29 times higher than a conventional slit, and particle size distribution (PSD) in the range between 2 and $20{\mu}m$ more than doubled. Treatment efficiency of synthetic oily wastewater through the coagulation-DAF process achieved 90% removal of COD by injecting $FeCl_3$ and PACl of 250 mg/L and 100 mg/L, respectively, and the same performance resulted using $FeCl_3$ of 200 mg/L and PACl of 80 mg/L employing a slit-nozzle angle of $60^{\circ}$. This study shows that a coagulation-DAF process using a modified slit-nozzle can improve the pre-treatment of oily wastewater.

• Atmosphere
• /
• v.28 no.4
• /
• pp.457-467
• /
• 2018

This study investigated the variability of spring (March-May) forest fire risk in Korea for the period 1991~2017 and analyzed its relationship with large-scale climate factors. The Forest Weather Index (FWI) representing the meteorological risk for forest fire occurrences calculated based on observational data and its relationship with large-scale climate factors were analyzed. We performed the empirical orthogonal function (EOF) analysis on the spring FWI. The leading EOF mode of FWI accounting for about 70% of total variability was found to be highly correlated with total number of forest fire occurrences in Korea. The high FWI, forest fire occurrence risk, in Korea, is associated with warmer atmosphere temperature in midwest Eurasia-China-Korea peninsula, cyclonic circulation anomaly in northeastern China-Korea peninsula-northwest pacific, westerly wind anomaly in central China-Korea peninsula, and low humidity in Korea. These are further related with warmer sea surface temperature and enhanced outgoing longwave radiation over Western Pacific, which represents a typical condition for a La $Ni\tilde{n}a$ episode. This suggests that large-scale climate factors over East Asia and ENSO could have a significant influence on the occurrence of spring forest fires in Korea.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.15 no.5
• /
• pp.93-103
• /
• 2005

In this paper, we propose a RST resistant robust logo embedding technique for multimedia copyright protection Geometric manipulations are challenging attacks in that they do not introduce the quality degradation very much but make the detection process very complex and difficult. Watermark embedding in the normalized image directly suffers from smoothing effect due to the interpolation during the image normalization. This can be avoided by estimating the transform parameters using an image normalization technique, instead of embedding in the normalized image. Conventional RST resistant schemes that use full frame transform suffer from the absence of effective perceptual masking methods. Thus, we adopt $8\times8$ block DCT and calculate masking using a spatio-frequency localization of the $8\times8$ block DCT coefficients. Simulation results show that the proposed algorithm is robust against various signal processing techniques, compression and geometrical manipulations.