• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.06a
• /
• pp.190-192
• /
• 2018

최근 딥 러닝 기술의 발전이 가속화됨에 따라, 기존의 알고리즘과 융합하여 뛰어난 성능 향상을 보이는 연구가 급격히 증가하고 있다. 본 논문에서는 딥 러닝을 이용하여 HEVC 로 압축된 동영상의 일부 프레임의 삭제여부를 검출하는 알고리즘을 제안한다. 영상의 삭제 정보가 포함되어 있는 HEVC 의 부호화 파라미터를 추출하여 간단한 전 처리 과정을 통해 데이터의 크기를 효과적으로 압축한 뒤, 동영상의 시간적 특성을 고려할 수 있도록 CNN 네트워크를 구성한다. 실험 결과, 효과적으로 다양한 압축 환경에 강인한 영상 삭제 검출 성능을 보이는 것을 확인하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.11a
• /
• pp.123-125
• /
• 2022

본 논문에서는 판별자를 활용하여 Image to Image translation(I2I) 분야에서 사용되는 적대적 생성 신경망(GAN)을 압축하는 방법을 제시한다. 우선, 잘 학습된 판별자와 생성자 사이의 adversarial loss 를 활용하여 생성자 내 필터들의 중요도 점수를 매겨준다. 그리고 생성자 내의 필터들을 중요도 점수를 기준으로 나열한 후 점수가 낮은 필터들을 제거하는 필터 프루닝을 한번 수행하여 적은 시간 비용으로 생성자를 압축한다. 마지막으로 지식 증류를 활용해 압축된 생성자를 학습시켜 기존의 생성자와 유사한 성능을 보이도록 하였다. 이 과정들을 통해 효과적이고 빠르게 GAN 모델을 압축할 수 있음을 확인하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.190-191
• /
• 2020

본 논문은 모델 분할 기법과 중복성 제거 기법을 통한 대용량 3차원 메쉬 모델의 고속 압축 기술에 관한 내용이다. 대용량 3차원 메쉬 모델의 비실시간 압축은 실시간 스트리밍 응용 시나리오에서 제약점으로 작용하고 있고, 본 논문에서는 인코딩 시간을 줄이기 위해 경량 메쉬 분할 방법을 통해 대용량 메쉬를 여러 개의 작은 메쉬로 분할하고, 각각의 분할된 메쉬를 병렬적으로 인코딩하여 처리 속도를 개선하였다. 또한, 메쉬 모델 내의 같은 기하학적 정보를 가진 중복된 정점들이 존재할 수 있으며, 중복된 정보를 제거하고 제거된 정점과 삼각형 표면 간의 연결 정보를 갱신하는 과정을 통해 메쉬 모델의 기하학적 정보를 유지하면서 압축 성능을 확보하였다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.107-109
• /
• 2022

본 논문은 건설구조물 안전의 가장 중요한 요소 중 하나인 콘크리트 압축강도 안정성을 확보하기 위한 시스템으로, 콘크리트 구조물을 만들 재료인 레미콘의 수분 감소율에 따른 압축강도 감소 리스크를 관리할 수 있는 모델을 제시하였다. 동일한 물,시멘트비(W/C)로 생산된 레미콘은 현장타설시까지 교통환경으로 인한 도착시간 지연 및 강우, 강설 등 외부적인 요건으로 감수율이 발생하는 리스크가 발생한다. 이로 인해 콘크리트의 압축강도가 저감하는 중대한 문제가 발생한다. 본 연구에서 제시한 알고리즘을 이용하여 현장 타설전 콘크리트 시료의 함수율을 측정하여 감수율이 발생한 제품 발견시, 실시간으로 Operator로 GCM 기반의 Push Alarm을 전송하여 감수율이 반영된 제품을 제공함으로써 구조물의 안전성을 확보할 수 있는 시스템을 모델링하였다. 본 연구는 기존시스템의 문제점을 실시간으로 개선할 수 있는 것으로 건설현장의 구조물 안정성 확보에 효과가 클 것으로 기대된다.

• Journal of the Korean GEO-environmental Society
• /
• v.13 no.11
• /
• pp.43-50
• /
• 2012

In this research, the value of consolidation index was investigated. The range of the investigated standard deviation was analyzed and the deviation based settlement was calculated. Also, the compression index, which is the effect of the uncertainty in the ground was analyzed using the flimsy ground construction method. The settlement behavior in each embankment compaction stage was analyzed by applying the precompression load method, drainage expediting method, and displacement method through numerical analysis. In addition to the above, the settlement behavior was studied by analyzing the Piled Raft method which is stable for long term settlement. As a result, the final settlement amount based on average analysis results was that the settlement based on each of the average interpretation value, mean value of the maximum and minimum value and average compression index was different. The result of the comparison shows the difference in variation coefficient by the difference in time. Amongst them, the Piled Raft method shows the most consistent variation coefficient regardless of time and it also was least affected by the compression index of uncertainty.

• Journal of Korea Game Society
• /
• v.13 no.5
• /
• pp.55-62
• /
• 2013

Since large terrain data can not be loaded on the GPU or CPU memory at once, out-of-core methods which read necessary part from the secondary storage such as a hard disk are commonly used. However, long delay may occur due to limited bandwidth while loading the data from the hard disk to memory. We propose efficient rendering method of large terrain data, which compresses the data with wavelet technique and save its coefficients in RGBA channel of an image us, then decompresses that in rendering stage. Entire process is performed in GPU using Direct Compute. By reducing the amount of data transfer, performing wavelet computations in parallel and doing decompression quickly on the GPU, our method can reduce rendering time effectively.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.2
• /
• pp.40-48
• /
• 1996

In this paper, a 40 dimensional segment vector with 4 frame and 7 frame width in every monosyllable interval was compressed into a 10, 14, 20 dimensional vector using K-L expansion and neural networks, and these was used to speech recognition feature parameter for CHMM. And we also compared them with CHMM added as feature parameter to the discrete duration time, the regression coefficients and the mixture distribution. In recognition test at 100 monosyllable, recognition rates of CHMM +${\bigtriangleup}$MCEP, CHMM +MIX and CHMM +DD respectively improve 1.4%, 2.36% and 2.78% over 85.19% of CHMM. And those using vector compressed by K-L expansion are less than MCEP + ${\bigtriangleup}$MCEP but those using K-L + MCEP, K-L + ${\bigtriangleup}$MCEP are almost same. Neural networks reflect more the speech dynamic variety than K-L expansion because they use the sigmoid function for the non-linear transform. Recognition rates using vector compressed by neural networks are higher than those using of K-L expansion and other methods.

• Journal of Biomedical Engineering Research
• /
• v.20 no.3
• /
• pp.307-314
• /
• 1999

This paper describes an informatlon compression of electrically evoked myoelectric signal, M-wave. This wave shows a direct response m lato-response of nerve conductlQn study and has a characteristic with finite time support. M-wave is a useful factor for investing neurodi~ease and is often desirable to have a compact description of its shape and time evolution. The aim of this paper is to show that the AR modeling IS a effective method for compressing an information of M-wave. First, AR model parameters of real M-wave are estimated. And then. they are verified by approximatmg a M-wave using estimated AR parameters and by comparing to other melhod, Hermite tlansform[4]. To concretely evaluate the proposed method, the NMSE(normalized mean square error) of approximation curves are compared. As a result, AR modeling is effective for M-wave assessment because of its capability for the information compression.

• Journal of the Korean Regional Science Association
• /
• v.37 no.2
• /
• pp.49-61
• /
• 2021

Long-distance commuting is a problem as people living in Gyeonggi-Incheon contitue to commute to Seoul in the Seoul metropolitan area. To solve this problem, policies in the region are aiming for a self-sufficient zone formation plan and a compact city. However, urban problems caused by such long-distance commuting continue. This appears to be due to excessive density and Seoul-dependent networks. However, existing studies have focused on individual cities despite the importance of inter-city interactions, and had limitations in not considering the characteristics of the Seoul-dependent networks. Therefore, the purpose of this study is to empirically analyze the effect of the compactness on commuter travels by comprehensively considering the interactions between cities within the region using multiple regression. As a result of the analysis, it was found that that commuting efficiency increases when a network of more than a certain size is formed, and the results imply that policies should focus on fostering network centers in Incheon and Gyeonggi regions, which are outside the metropolitan area, and consider to expand the transportation networks at the regional level.

• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.200-207
• /
• 2020

Recently, deep neural networks such as CNN are showing excellent performance in various fields such as image classification, object recognition, visual quality enhancement, etc. However, as the model size and computational complexity of deep learning models for most applications increases, it is hard to apply neural networks to IoT and mobile environments. Therefore, neural network compression algorithms for reducing the model size while keeping the performance have been being studied. In this paper, we apply few compression methods to CNN models and evaluate their performances in the embedded environment. For evaluate the performance, the classification performance and inference time of the original CNN models and the compressed CNN models on the image inputted by the camera are evaluated in the embedded board equipped with QCS605, which is a customized AI chip. In this paper, a few CNN models of MobileNetV2, ResNet50, and VGG-16 are compressed by applying the methods of pruning and matrix decomposition. The experimental results show that the compressed models give not only the model size reduction of 1.3~11.2 times at a classification performance loss of less than 2% compared to the original model, but also the inference time reduction of 1.2~2.21 times, and the memory reduction of 1.2~3.8 times in the embedded board.