• 방송공학회논문지
• /
• 제26권6호
• /
• pp.778-789
• /
• 2021

최근 다양한 분야에서 뛰어난 성능을 나타내는 Convolutional Neural Network(CNN)모델을 모바일 기기에서 사용하기 위한 다양한 연구가 진행되고 있다. 기존의 CNN 모델은 모바일 장비에서 사용하기에는 가중치의 크기가 크고 연산복잡도가 높다는 문제점이 있다. 이를 해결하기 위해 가중치의 표현 비트를 낮추는 가중치 양자화를 포함한 여러 경량화 방법들이 등장하였다. 많은 방법들이 다양한 모델에서 적은 정확도 손실과 높은 압축률을 나타냈지만, 대부분의 압축 모델들은 정확도 손실을 복구하기 위한 재학습 과정을 포함시켰다. 재학습 과정은 압축된 모델의 정확도 손실을 최소화하지만 많은 시간과 데이터를 필요로 하는 작업이다. Weight Quantization이후 각 층의 가중치는 정수형 행렬로 나타나는데 이는 이미지의 형태와 유사하다. 본 논문에서는 Weight Quantization이후 각 층의 정수 가중치 행렬을 이미지의 형태로 비디오 코덱을 사용하여 압축하는 방법을 제안한다. 제안하는 방법의 성능을 검증하기 위해 ImageNet과 Places365 데이터 셋으로 학습된 VGG16, Resnet50, Resnet18모델에 실험을 진행하였다. 그 결과 다양한 모델에서 2%이하의 정확도 손실과 높은 압축 효율을 달성했다. 또한, 재학습 과정을 제외한 압축방법인 No Fine-tuning Pruning(NFP)와 ThiNet과의 성능비교 결과 2배 이상의 압축효율이 있음을 검증했다.

• 한국문화재보존과학회:학술대회논문집
• /
• 한국문화재보존과학회 2004년도 제20회 발표논문집
• /
• pp.115-130
• /
• 2004

Plant-based cultural assets using straw and grass as household goods of our people's have been used as indispensable tools for practical living for a very long time. However, only a limited number of artifacts were unearthed so far due to tile fragility of the material. For this reason, research on plant-based cultural properties had close to no progress, and the appropriateness of the PEG method, high-grade alcohol method, alcohol--ether-resin method, and Paraloid B-72 used in preserving plant-based cultural properties has not been sufficiently investigated. Therefore, this study examined the weight change rate by applying the methods of Primal MC-76 and vacuum freeze-drying used mostly as a earth-layer hardening material among PEG and acrylic resin, which are applied widely for preservation of waterlogged archaeological wood, as a means to preserve plant-based cultural properties along with the examination of the subject material, and an experiment was also performed on moisture absorption. The findings as a result were, first, the plant-based material being studied was found to be Typha (Typha orientalis Presl). Secondly, the weight change experiment applying $PEG\#400$ and $PEG\#4000$ confirmed a steady increase of weight if PEG -2Step is used for treatment. Third, in preserving all subject materials with soil, treatment with $PEG\#4000$, Primal MC-76, and vacuum freeze-drying showed that tile vacuum freeze-drying method resulted in the largest or $20\%$ reduction in weight, while Primal MC-76 resulted in $18\%$ and $PEG\#4000$ in $8\%$ of weight reduction. It was concluded that, considering the stability of soil measurement, this came to be because resin permeation was carried out along with tile drying process. Fourth, the weight changes were found to be around $10\%$ in various humidity conditions after the preservation treatment. The greatest weight change rate was seen in the case of $PEG\#4000$, particularly having chemicals gush out in a high humidity (RH $84\%$ or higher) environment. In the case of Primal MC-76 and vacuum freeze-drying methods, $6\~8\%$ weight changes were detected, and the lowest weight change was found in the case of the vacuum freeze-drying method. Fifth, as for color changes after treatment, blackening occurred most strongly with $PEG\#4000$, while Primal MC-76 and vacuum freeze-drying manifested colors closest to dry straw or grass. However, the texture of straw was not very evident in the case of Primal MC-76, due to a glossy surface, but vacuum freeze-drying was found to offer tile best result in terms of texture. Putting together the results of the above experiments, vacuum freeze-drying presented after being treated with PEG2-Step the most stabilized changes in weight, while it offered the smallest change in color as well.

• Asian-Australasian Journal of Animal Sciences
• /
• 제32권6호
• /
• pp.865-873
• /
• 2019

Objective: The study was designed to investigate the effects of restraint method, restraint duration, and body weight on stress-linked hormones (corticosterone, adrenaline, and noradrenaline), blood biochemical (namely glucose and lactate), and the meat quality in broiler chickens. Methods: A total of 120 male broiler chickens (Cobb 500) were assigned to a $2{\times}3{\times}2$ factorial arrangement in a completely randomized design using two restraint methods (shackle and cone), three durations of restraint (10, 30, and 60 s), and two categories of live body weight ($1.8{\pm}0.1kg$ as lightweight and $2.8{\pm}0.1kg$ as heavyweight). Results: Irrespective of the duration of restraint and body weight, the coned chickens were found to have lower plasma corticosterone (p<0.01), lactate (p<0.001), lower meat drip loss (p<0.01), cooking loss (p<0.05), and higher blood loss (p<0.05) compared with their shackled counterparts. The duration of restraint had significant effects on the meat initial pH (p<0.05), ultimate pH (p<0.05), and yellowness (p<0.01). The lightweight broilers exhibited higher (p<0.001) blood loss and lower (p<0.05) cooking loss compared to the heavyweight broilers, regardless of the restraint method used and the duration of restraint. However, the interaction between the restraint method, duration of restraint, and body weight contributed to differences in pre-slaughter stress and meat quality. Therefore, the interaction between the restraint method and the duration of restraint affected the meat shear force, lightness ($L^*$) and redness ($a^*$). Conclusion: The duration of restraint and body weight undoubtedly affect stress responses and meat quality of broiler chickens. Regardless of the duration of restraint and body weight, the cone restraint resulted in notably lower stress, lower meat water loss, and higher blood loss compared to shackling. Overall, the findings of this study showed that restraint method, duration of restraint, and body weight may affect the stress response and meat quality parameters in broilers and should be considered independently or interactively in future studies.

• 대한한방내과학회지
• /
• 제13권1호
• /
• pp.61-69
• /
• 1992

In Nan-Gyung, showed that could know the lung condition taking pulse with the weight of three beans, the heart condition taking pulse with the weight of six beans, the spleen condition taking pulse with the weight of nine beans, the liver condition taking pulse with the weight of twelve beans, the kidney condition pressing to bone(骨). This theory is first suggested in Nan-Gyung(難經). In those case, the weight of three, six, nine, twelve beans and pressing to bone don't mean not the real weight but the relative weight(輕重) of taking pulse(按脈). In other words, those represent Boo Jung Chin(浮中沈), which are the conception of the upper, the meddle, the lower part(上中下). So, we could take pulse of the heart and the lung condition in Boo(浮), the spleen condition in Jung(中), and the liver and the kidney condition in Chim(沈). The heart and the lung pulse showed in the Boo(float level) must be seen with Boo-Mack(부맥 : float pulse), the liver and the kidney pulse showed in Chim (sinklevel) must be seen also with Chin-Mack(沈脈 : sink pulse). The result of the method of taking pulse of viscera with relative weight focused on the as pect of mornal pulse(平脈) and disease pulse(病脈) of five viscera in Mack-Gyung publeshed later than Nan-Gyung and special works which made a comprehensive survey the result is as follow. 1. In normal pulse of five viscera, the heart and the lunk pulse were shown with Boo-Mack(浮脈:float pulse) as the central figure, the liver's and the kidney's pulse were shown centering around Chim-Mack(沈脈: sink pulse) and the spleen's pulse was shown with Wan-Mack(緩脈) which is vital force of stomach(胃氣) and seen in only middle part. 2. In disease pulse of five viscera, frequently, the heart and the lung pulse was shown as Chim-Mack(sink pulse), the liver and the kidney pulse was seen as Boo-Mack (float pulse). 3. In the case of normal pulse. the method of taking pulse with relative weight in Nan-Gyung agree with the normal pulse of five viscera in Mack-Gyung. But in the case of disease pulse, they didn't correspond with the other. 4. So the method of taking pulse with relative weight in Nan-Gyung is not the exam pulse which ca be used in the clinical diagnosis but one of the feeling pulse way to bring in the conception of location of the visceras. 5. From now on, the method of taking pulse rdlated to relative weight need to be looked into minutely compared with later physician's theory than Mack-Gyung.

• 한국식품조리과학회지
• /
• 제7권3호
• /
• pp.69-76
• /
• 1991

6종류의 식용 버섯에 함유되어 있는 dietary fiber를 Southgate방법, NDF방법, FRI방법으로 측정한 결과는 다음과 같았다. 식용 버섯 중에 함유되어 있는 dietary fiber를 Southgate방법, NDF방법, FRI방법으로 측정한 결과는 각각 $20.08\pm1.45g$/100g dry weight, $20.24\pm1.85g$/100g dry weight, $27.5\pm2.7g$/100g dry weight이었고, 세가지 실험 방법간의 측정치 차이를 보면, 대체로 Southgate방법과 NDF방법과는 유의적인 차이가 없었던 반면, FRI방법과는 유의적인 차이가 크게 나타났다. 버섯중의 dietary fiber성분 비율은 Southgate방법에 의 한 water-soluble fiber성분이 1.7~3.1%, hemicellulose가 47.0~66.6%, cellulose가 28.4~57 7%, lignin 0.9~3.3%였고, NDF방법에 의해 측정한 hemicellulose는 61.8-79.1%, cellulose는 5.4~32.7%, lignin은 4.5~15.5% 였다. 일반적으로 버섯류의 dietary fiber성분은 hemicellulose가 가장 많았고 그 다음으로 cellulose가 많이 함유되어 있었다. Dietary fiber와 crude fiber의 함량비 (DF/CF)는 2~4배 정도로 dietary fiber가 높았다.

• 복식문화연구
• /
• 제11권3호
• /
• pp.441-446
• /
• 2003

Water-insoluble chitosan with molecular weight of 2,000,000, 500,000, 80,000, and 40,000 and more than 90% of degree of deacetylation were used to test antibacterial activity of chitosan against a pathogenic bacteria, methicillin resistant Staphylococcus aureus(MRSA), which is being issued in the world. As experimental method, Agar plate Smear Method and Agar plate Contact Method were used. The moleculur weight of chitosan didn't exert significant influences on its antibacterial activity against MRSA but chitosan having molecular weight 40,000, 80,000 and 150,000 showed the excellent antibacterial activity. The antibacterial efHciency was excellent in applying it after chitosan was dissolved in acetic acid solution, while the antibacterial efficiency was not expressed nearly in case of applying after chitosan was dissolved in neutral water. Therefore, it is considered that chitosan can show the antibacterial efficiency only if a positive ion status of -NH₃/sup +/ is maintained. MIC of chitosan/acetic acid solution and cotton fabrics finished with chitosan/acetic acid solution showed in concentration of 0.05%.

• 한국신재생에너지학회:학술대회논문집
• /
• 한국신재생에너지학회 2007년도 추계학술대회 논문집
• /
• pp.152-155
• /
• 2007

Fuel cell electric vehicles (FCEVs) using hydrogen gas are zero emission vehicles, thus emission measurement for combustion vehicles is not applicable. The hydrogen gas consumption for fuel economy will be measured by the stabilized pressure/temperature method, mass flow method and electrical current method, etc. In this research, weight method with a newly manufactured test equipment is applied to measure the hydrogen consumption because above 3-methods have a deviation. The hydrogen consumption is directly calculated by the weight differences of the external hydrogen tank before and after the chassis dynamometer test. Ultimately the fuel economy for FCEVs is obtained with a deviation less than 1% in all chassis dynamometer tests.

• 소음진동
• /
• 제10권1호
• /
• pp.97-106
• /
• 2000

The dynamic behavior of crankshaft-bearing system in scroll compressor has been investigated using the combined methodologies of finite elements and transfer matrices. The finite element formulation is proposed including the field element for a shaft section and the point element at balancer weight locations, bearing locations, etc., whereas the conventional method is used with the elements. The Houbolt method is used to consider the time march for the integration of the system equations. The linear stiffness and damping coefficients are calculated for a finite cylindrical fluid-film bearing by solving the Reynolds equation, using finite difference method. The orbital response of crankshaft supported on the linear bearing model is obtained, considering balancer weights of motor rotor. And, the steady state displacement of crankshaft are compared with a variation in balancer weight. The loci of crankshaft at bearing locations are composed of the synchronous whirl component and the non-synchronous whirl component.

• 한국컴퓨터정보학회논문지
• /
• 제23권7호
• /
• pp.81-90
• /
• 2018

In this paper, reducing lense Vignetting effect and adaptive learning rate method are proposed to complement Scribner's neural network for nuc algorithm which is the effective algorithm in statistic SBNUC algorithm. Proposed reducing vignetting effect method is updated weight and bias each differently using different cost function. Proposed adaptive learning rate for updating weight and bias is using sobel edge detection method, which has good result for boundary condition of image. The ordinary statistic SBNUC algorithm has problem to compensate lense vignetting effect, because statistic algorithm is updated weight and bias by using gradient descent method, so it should not be effective for global weight problem same like, lense vignetting effect. We employ the proposed methods to Scribner's neural network method(NNM) and Torres's reducing ghosting correction for neural network nuc algorithm(improved NNM), and apply it to real-infrared detector image stream. The result of proposed algorithm shows that it has 10dB higher PSNR and 1.5 times faster convergence speed then the improved NNM Algorithm.

• 산업공학
• /
• 제16권spc호
• /
• pp.144-148
• /
• 2003

A group of digital human models with various sizes which properly represents a population under consideration is needed in the design process of an ergonomic product in virtual environment. The present study proposes a two-step method which produces a representative group of human models in terms of stature and weight. The proposed method first generates a designated number of pairs of stature and weight within an accommodation range from the bivariate normal distribution of stature and weight of the target population. Then, from each pair of stature and weight, the method determines the sizes of body segments by using 'hierarchical' regression models and corresponding prediction distributions of individual values. The suggested method was applied to the 1988 US Army anthropometric survey data and implemented to a web-based system which generates a representative group of human models for the following parameters: nationality, gender, accommodation percentage, and number of human models.