http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Determining the fate of Shh-expressing cells in the diencephalon using a BAC transgenic reporter
이범휘,박민호,백광희,윤재승,정용수 한국유전학회 2010 Genes & Genomics Vol.32 No.6
The thalamus and prethalamus consist of multiple distinct nuclei and their boundary is demarcated by the zona limitans intrathalamica (ZLI). The development of the primordial thalamus and prethalamus proceed within the caudal diencephalon. Shh has been shown to be essential for diencephalic patterning and regionalization. To understand the role of Shh in the specification of distinct thalamic and prethalamic nuclei, we developed a lineage marker for diencephalic cells expressing Shh by using bacterial artificial chromosome (BAC) transgenesis. A genomic fragment containing ~210 kb of the mouse Shh locus was used to target enhanced green fluorescent protein (eGFP) in transgenic mice. This transgenic BAC reporter faithfully mimicked the pattern of endogenous Shh expression in the caudal diencephalon, including the ZLI. Fate mapping analysis at multiple developmental stages showed that descendents of Shh-expressing progenitor cells derived from ZLI contribute to a population of cells in the ventral lateral geniculate nucleus.
해상환경 및 수중타설이 고강도 그라우트의 역학적 성능에 미치는 영향
김범휘 ( Kim Beom-hwi ),손다솜 ( Son Da-som ),이종구 ( Yi Chong-ku ) 한국건축시공학회 2023 한국건축시공학회 학술발표대회 논문집 Vol.23 No.2
In this study, grout was poured into seawater to confirm the effect of similar marine environment and underwater erosion on the mechanical performance of domestically produced high-performance grout and compared with the existing strength. As a result of the compressive strength measurement, the specimen that simultaneously performed underwater drilling and seawater curing showed slow initial strength expression in both H1 and H2, and from the 7th day, it was confirmed to be within 2% of the existing intensity. It is believed that both grout were caused by disturbance with water during underwater drilling, and the same strength was subsequently shown as the existing strength.
양생환경 및 수중펌프압송이 고강도 그라우트의 강도에 미치는 영향
김범휘 ( Kim Beom-hwi ),손다솜 ( Son Da-som ),이종구 ( Yi Chong-ku ) 한국건축시공학회 2023 한국건축시공학회 학술발표대회 논문집 Vol.23 No.1
In recent years, the use of high-strength grout has gained popularity in offshore wind power generation complexes for facility foundations and bridges. These marine wind farms require support for horizontal loads from wind and waves. To ensure the strength of the grout produced in environments similar to the actual placing site, this study investigated the curing of high-strength grout discharged through pump pressure in various environments, and examined the difference in strength according to different variables. Compressive strength measurements revealed that the core specimen collected from the bottom (3cm) and uppermost (50cm) of the specimen exhibited lower strength compared to other height specimens, while the core specimen obtained from the corner exhibited lower strength compared to the center. These findings suggest that the strength difference between the center and the corner is more pronounced when curing at low temperatures. This effect is greater than the strength reduction that typically occurs during low-temperature curing, and thus, necessitates careful attention in similar construction environments.
최범휘(Beom-Hwi Choi),이재준(Jae-Jun Lee),한현택(Hyeon-Taek Han),최연웅(Yeon-Ung Choi),조우성(Woo-seong Cho),이해연(Hae-Yeoun Lee) 한국정보기술학회 2021 Proceedings of KIIT Conference Vol.2021 No.11
사람 음성 활동 감지는 스마트 홈이나 자동차 등 다양한 응용 분야에서 활용될 수 있으며, 딥러닝 기술을 이용한 연구들도 수행되고 있다. 본 논문에서는MobileNet 딥러닝 모델을 이용하여 사운드 세그먼트에 사람음성 활동이 있는지 검출하는 모델을 제안한다. 사운드 세그먼트의 MFCC 특징 추출을 위하여 MFCC 특징을 추출하였고, CNN 기반의 모델들보다 연산 복잡도가 최소화되고 의미있는 특징 데이터를 학습할 수 있는 MobileNet을 도입하여 최적화를 수행하였다. 이를 통하여 95.52% 정확도로 사람 음성 활동 여부를 검출하였다. Voice activity detection of human can be used in various applications such as smart homes and automobiles, and researches using deep learning technology are being conducted. In this paper, we propose a voice activity detection model of human in sound segments using MobileNet. The MFCC features were extracted as features of the sound segments. Also, MobileNet, which has lower computational complexity than CNN-based models and can learn meaningful features, is applied and optimized. As a result, voice activity detection was performed with 95.52% accuracy.
Residual 3D U-Net 기반 다발성 경화증 병변의 의미론적 분할
최범휘(Beom-Hwi Choi),이재준(Jae-Jun Lee),최연웅(Yeon-Ung Choi),한현택(Hyeon-Taek Han),이해연(Hae-Yeoun Lee) 한국정보기술학회 2021 한국정보기술학회논문지 Vol.19 No.11
Medical images in clinical practice are analyzed by pathologists, requiring human and time resources. Recently, deep learning methods that can predict the location of specific diseases in medical images has been developed. Also, regularization, constraints, and augmentation are introduced to solve data scarcity problems. This paper proposes an algorithm to detect multiple sclerosis lesions using Residual 3D U-Net. In particular, detection accuracy is improved through preprocessing and random patch enhancements. To verify the performance of the proposed algorithm, MS lesion datasets from 2008 MICCAI MS lesion segmentation challenge was used and compared with 3D U-Net. As a result, 70% DSC and 52% IoU were achieved, and 5.9% DSC and 6.9% IoU accuracy were improved. Quantitatively, detection results showed satisfactory performance except for local areas within the entire MRI image.
프라이빗 블록체인 기반 데이터 무결한 스마트 임상시험 관리 시스템
김범휘(Bumhwi Kim),김규형(Kyu Hyung Kim) 한국통신학회 2021 한국통신학회 학술대회논문집 Vol.2021 No.6
본 논문에서는 프라이빗 블록체인을 활용하여 임상시험에서 발생하는 데이터에 대해서 무결성을 확보할 수 있는 블록체인 기반 스마트 임상시험 관리 시스템을 구현하는 것에 대해 설명한다.
Perceiver 모델 기반의 음성 녹음 파일 분류 방법
최범휘,김준수,황현욱,이해연 한국디지털포렌식학회 2022 디지털 포렌식 연구 Vol.16 No.4
Most recent digital crimes are committed through data using devices such as PC and smartphones. Audio is also a subject of criminal investigations, and the demand for audio digital forensics technology is increasing. In this paper, we propose a method to classify general audio, public voice recording, and user voice recording from various audio files using deep learning. For training and evaluation of the proposed method, a dataset was constructed by defining the 3 classes of general audio, public voice recording, and user voice recording. After classifying the entire audio in segment units, the class was finally determined. Since the deep learning model based on the convolutional neural network has low accuracy for classifying voice recording in segment units, the Perceiver model, which is a variation of the transformer neural network used in natural language processing, was adapted. The input data was generated through MFCC feature extraction, attention and transform layers were modified to be suitable for voice recording classification. Also, parameters were optimized. In the experiment, the proposed method was trained as a constructed dataset and the classification performance was analyzed in the audio segment units and the entire audio file units. As a result, 83.47% accuracy was achieved for each segment unit and 95.19% accuracy was achieved for the entire audio file. 최근에 대부분의 디지털 범죄는 개인용 컴퓨터와 스마트폰을 이용한 데이터를 통해 발생하고 있다. 오디오도 범죄 수사의 대상이며 오디오에 대한 디지털 포렌식 기술의 수요가 증가하고 있다. 본 논문에서는 딥러닝을 이용하여 다양한 오디오 파일에서 일반 오디오 및 공개 음성 녹음, 사용자 음성 녹음의 3종에 대한 분류를 위한 방법을 제안한다. 제안한 방법의 학습과 평가를 위해 일반 오디오 및 공개 음성 녹음, 사용자 음성 녹음의 3종 클래스를 정의하여 데이터셋을 구축하였고, 전체 오디오에 대하여 세그먼트 단위로 분류를 수행한 후에 최종적으로 클래스를 결정하였다. 세그먼트 단위로 음성 녹음 분류에 있어서 컨볼루션 신경망 기반의 딥러닝 모델은 정확도가 낮아서, 자연어 처리에 사용되는 트랜스포머 신경망 기반의 Perceiver 모델을 도입하였다. MFCC 특징 추출을 통하여 입력 데이터를 생성하고, 음성 녹음 분류에 적합하도록 어텐션과 트랜스포머 계층을 수정하고 파라미터를 최적화하였다. 실험에서는 제안한 방법을 구축한 데이터셋으로 학습하고, 오디오 세그먼트 단위와 전체 오디오 파일 단위의 분류 성능을 분석하였다. 그 결과 세그먼트 단위로 83.47% 정확도를 달성하였고, 전체 오디오 파일에 대하여 95.19% 정확도를 달성하였다.
김범휘 ( Kim Beom-hwi ),이종구 ( Yi Chong-ku ) 한국건축시공학회 2022 한국건축시공학회 학술발표대회 논문집 Vol.22 No.2
The use of high-strength grout for facility foundations and bridges has recently been expanding in offshore wind farms. Offshore wind farms require a bearing capacity for horizontal loads such as wind, waves. Therefore, in this study, the strength of the high-strength grout discharged through pump pressure was measured and compared with the existing strength to secure the strength after the underwater pump pressure of the high-strength grout used in the offshore wind connection. The compressive strength measurement showed that the strength difference at each position of the core specimen was 1% higher than that of the other specimens, and there was almost no change in the strength according to the height. The strength of the core specimen decreased by 23% compared to the existing strength, which is similar to the result of this study because the strength of the core specimen decreased by approximately 25% compared to the general specimen according to related research. Therefore, it is believed that there is no decrease in strength due to underwater pumping.