• 대한임베디드공학회논문지
• /
• 제17권3호
• /
• pp.129-138
• /
• 2022

Collecting and labeling sufficient training data, which is essential to deep learning-based visual inspection, is difficult for manufacturers to perform because it is very expensive. This paper presents a steel plate surface defect detection system with industrial-grade detection performance by training a small amount of steel plate surface images consisting of labeled and non-labeled data. To overcome the problem of lack of training data, we propose two data augmentation techniques: program-based augmentation, which generates defect images in a geometric way, and generative model-based augmentation, which learns the distribution of labeled data. We also propose a 4-step semi-supervised learning using pseudo labels and consistency training with fixed-size augmentation in order to utilize unlabeled data for training. The proposed technique obtained about 99% defect detection performance for four defect types by using 100 real images including labeled and unlabeled data.

• 한국산학기술학회논문지
• /
• 제20권5호
• /
• pp.1-6
• /
• 2019

유기 염료가 도핑 된 실리카 나노입자는 바이오 라벨링, 바이오 이미징 및 바이오 센싱에 사용되고 있는 유망한 나노소재이다. 일반적으로 형광 실리카 나노입자는 수정된 스토버 방법($St{\ddot{o}}ber$ Method)으로 합성된다. 본 연구에서는 다양한 크기를 갖는 염료가 첨가되지 않은 형광 실리카 나노입자를 수정된 스토버 합성법인 졸겔 공정으로 합성하였다. 졸겔 공정 중에 기능성 물질인 APTES를 첨가제로 첨가하였다. 졸겔 공정으로 합성된 실리카 나노입자는 $400^{\circ}C$에서 2시간 동안 하소되었다. 합성된 실리카 나노입자의 표면형상과 크기를 전계방출 주사전자현미경으로 조사하였고, 합성된 실리카 나노입자의 형광 특성은 파장 365 nm의 자외선 램프를 조사하여 확인하였다. 또한 합성된 실리카 나노입자의 광발광 (PL) 특성을 형광 분석 형광법으로 조사하였다. 그 결과 합성된 실리카 나노입자는 입자의 크기와 무관하게 모두 청색 형광 특성을 갖는 것으로 확인되었다. 특히, 실리카 나노입자의 크기가 증가할수록 PL 강도는 감소하였다. 염료가 첨가되지 않은 실리카 나노입자의 청색 형광 특성은 APTES 층의 $NH_2$ 기능기와 실리카 매트릭스 뼈대 내부의 산소관련 결함과의 결합에 기인하는 것으로 추정된다.

• Transactions on Control, Automation and Systems Engineering
• /
• 제4권3호
• /
• pp.224-230
• /
• 2002

In this paper, a counting algorithm hybridized with an adaptive automatic thresholding method based on Otsu's method and the algorithm that elongates markers obtained by the well-known watershed algorithm is proposed to enhance the exactness of the microcell counting in microscopic images. The proposed counting algorithm can be stated as follows. The transformed full image captured by CCD camera set up at microscope is divided into cropped images of m$\times$n blocks with an appropriate size. The thresholding value of the cropped image is obtained by Otsu's method and the image is transformed into binary image. The microbial cell images below prespecified pixels are regarded as noise and are removed in tile binary image. The smoothing procedure is done by the area opening and the morphological filter. Watershed algorithm and the elongating marker algorithm are applied. By repeating the above stated procedure for m$\times$n blocks, the m$\times$n segmented images are obtained. A superposed image with the size of 640$\times$480 pixels as same as original image is obtained from the m$\times$n segmented block images. By labeling the superposed image, the counting result on the image of microbial cells is achieved. To prove the effectiveness of the proposed mettled in counting the microbial cell on the image, we used Acinetobacter sp., a kind of ammonia-oxidizing bacteria, and compared the proposed method with the global Otsu's method the traditional watershed algorithm based on global thresholding value and human visual method. The result counted by the proposed method shows more approximated result to the human visual counting method than the result counted by any other method.

• 대한치과보철학회지
• /
• 제29권3호
• /
• pp.55-74
• /
• 1991

The purpose of this study was to investigate the influence of early functional load around osseointegrated titanium implants. 24 titanium plasma spray coated implants (ITI HS-type) were placed into the previously extracted site in the mandible of six adult dogs. The implants were divided into three groups : the control group was the implants without abutment during the experimental period; the experimental group I was loaded by connecting the contoured abutment after 6 weeks of healing; the experimental group II was loaded after 12 weeks of healing: and the mandibular second premolar and surrounding tissues were selected for natural tooth group to compare the implanted group. All dogs were injected intravenously tetracycline, alizarin red S, and calcein for bone labeling. After the experimental period of 18 weeks, the dogs were sacrificed and longitudinal sections of the bone-implant interface were cut and observed using light microscope, scanning electron microscope, and fluorescence microscope. The results of the study were as follows: 1. Light and scanning electron microscopically, all implant surfaces were well contact with bone tissue at the cortical layer, but some areas of cancellous bone were not contact directly. 2. Fluorescence microscopically, number and size of the new secondary osteons around the implant were increased than those of the natural tooth. 3. Fluorescence microscopically, linear and concentrical fluorescence was observed at or near the surface of all implants, and the bone formation and remodeling of the implants loaded after 6 week of healing were great, and unloaded implants were worst. 4. Fluorescence microscopically, endosteal bone formation was greater than periosteal bone formation at or near the implants. 5. Fluorescence microscopically, number and size of linear and concentric fluorescence was increased at the lingual side than the buccal side of the loaded implants. The result of the study indicate the possibility of the early load to the implant via a prosthesis.

• 대한지역사회영양학회지
• /
• 제14권2호
• /
• pp.147-157
• /
• 2009

The purpose of this study was to find out the information on nutrition labeling and how many calories and nutrients the high school students consumed for 1 day from cookies. A total of 74 male and female high school students in Suwon were surveyed and 56 cookies that they consumed were examined. Background data were collected by questionnaire, cookie intake by 24-hr recall, and the calories and nutrients content in cookies and the amount of intake by nutrition information on the wrapping paper of cookie. The statistical analysis for the data was done by SPSS 12.0. Energy contents in 1 serving size of cookie were $90{\sim}315\;kcal$, average of 170 kcal. The protein contents were $0{\sim}7\;g$, fat $2{\sim}20\;g$, cholesterol $0{\sim}55\;mg$, and sodium $30{\sim}390\;mg$ in 1 serving size of cookie. Most of the cookies(80%) examined contained no trans fat at all, which is desirable. Among the types of cookies, snacks contained higher quantities of calories and sodium, the pie contained more sugar and cholesterol, and the biscuit had more trans fat. One fourth of the cookies examined belonged to 'high calorie, low nutritious food' according to the criteria proposed by The Ministry for Health, Welfare and Family Affairs. Thus the excessive intake of cookies might result in nutritional imbalance. There were large differences in calorie intake among students, from zero who did not intake any cookies at all to maximum 818 kcal/day, an average of 75 kcal/day. When the students who did not intake cookies were excluded, energy 205 kcal. fat 10 g, sodium 177 mg were consumed from the cookie for a 1 day on average.

• Journal of Endocrine Surgery
• /
• 제18권4호
• /
• pp.228-235
• /
• 2018

Purpose: The epidermal growth factor receptor (EGFR) family plays a crucial role in the growth of malignant tumors. EGFR and human EGFR 2 (HER2) protein overexpression are associated with an unfavorable prognosis and are important therapeutic targets in breast cancer. The aim of this study was to evaluate the relationship between EGFR and HER2 expression and clinicopathological factors in papillary thyroid carcinoma (PTC) at a single institution. Methods: A total of 129 consecutive patients with PTC were enrolled in this study and underwent thyroid surgery between October 2013 and February 2015. EGFR and HER2 protein expression was evaluated in the 129 primary tumors by immunohistochemistry, and the results were compared with the clinicopathological features. Results: Of the 129 PTC tumors, 20 (15.5%) were HER2 positive, and 109 (84.5%) were HER2 negative. Moreover, EGFR positivity were observed in 111 (86%) tumors. The mean age of the patients was $46.3{\pm}11.9years$ (range, 20-74 years), and the mean tumor size was $1.08{\pm}0.75cm$ (range, 0.2-3.5 cm). Tumor size, extrathyroidal extension, histological subtype, and TNM stage were not significantly associated with EGFR or HER2 expression. Meanwhile, high Ki-67 labeling index was significantly associated with EGFR expression (P=0.002), HER2 expression was significantly associated with younger age (${\leq}45years$) and cervical lymph node metastasis. Conclusion: Based on our data, it is not clear whether EGFR and HER2 expression is associated with tumor aggressiveness in PTC.

• Journal of Nutrition and Health
• /
• 제46권4호
• /
• pp.382-390
• /
• 2013

This study was designed to investigate the amount of free sugar according to each beverage category in coffee and beverage stores. The groups were categorized as 15 groups based on the kind of beverage material. The beverage groups contributing to total sugar per 100 mL were milk + syrup or powder, hot (12.9 g), ade (12.6 g), milk + syrup or powder + crushed ice (11.9 g), and espresso shot + milk + syrup + crushed ice (11.4 g). The beverage groups contributing to free sugar per 100 mL were ade (12.6 g), milk + syrup or powder + crushed ice (10.8 g), espresso shot + milk + syrup + crushed ice (10.3 g), and milk + syrup or powder, hot (9.7 g). The beverage groups contributing to total sugar (energy) per portion size were milk + syrup or powder + crushed ice 56.6 g (332.3 kcal), espresso shot + milk + syrup + crushed ice 49.3 g (333.4 kcal), milk + syrup or powder, hot 46.3 g (372.1 kcal), and milk + syrup or powder, ice 38.1 g (325.9 kcal). The beverage groups contributing to free sugar per portion size were milk + syrup or powder + crushed ice 51.2 g, espresso shot + milk + syrup + crushed ice 44.9 g, ade 37.1 g, milk + syrup or powder, hot 34.6 g, and milk + syrup or powder, ice 30.1 g. The percent of average free sugar per portion size of the WHO recommendation (free sugars <10% of total energy; <50 g/2,000 kcal) was milk + syrup or powder + crushed ice 102.4%, espresso shot + milk + syrup + crushed ice 89.8%, ade 74.1%, and milk + syrup or powder, hot 69.2%. The proportion of beverage in excess of WHO recommendation per portion size was 14.6% in espresso shot + milk + syrup + crushed ice, 22.7% in ade, and 10.9% in milk + syrup or powder, hot. Therefore, in coffee and beverage stores, menu development with reduced sugar content is needed, and nutrition information should be provided through sugar nutrition labeling.

• Journal of Nutrition and Health
• /
• 제30권6호
• /
• pp.679-689
• /
• 1997

Nutrition label (NL) on the package of processed food provides consumers with a reliable and consistent source of information . It has been considered as a useful aid for food selection and a potent educational tool for nutrition in daily life. Since current nutrition labeling regulation in Korea does not define a format for presenting nutrition information a wide variety of NL format exists in the markers created by individual manufacturers. Development of standard NL format and its registration remain to be the work for the professionals and government officials. However the acceptance and evaluation of NL by the consumers is a very important and necessary process in the development of NL formats. In this study four different formats A, B, C, D were formulated based on currently circulating labels and new U.S.NL. Subjects used for evaluation of these formats were middle -aged highly educated housewives, who and the potential users of NL. Major parameters observed through the questionnare were their nutritional knowledge of RDA, ability of IC(Information Comparison) and CA (Comprehension and Application of informed nutrient contents), as well as their preference to the different formats. The results are summarized as follows. 1) Of the 178 subjects , 89.9% of the middleaged housewives were college graduates. Their nutrition knowledge of RDA were relatively satisfactory showing over 80% correlation on the basic concepts and unit while for numerical value less than 50% correct answer. 2) IC test scores were significantly different among the formats showing the highest values for format A and B which are presented as absolute value and % RDA, respectively. Format C presented as serving size(number of products) showed the lowest score. CA scores were also significantly different, though the increased load of information did not facilitate to increase the consumers comprehension. 3) RDA knowledge test scores and the scores of IA and CA were correlated in format A and D but not in format B and C suggesting % RDA presentation would be more acceptable to the less educated group. 4) For the preference in the aspects of easiness and time-saving format A was the best one then format D supporting the result of IC and CA test. The results of the present study indicate the most useful and preferred format is the simplest format presented as absolute value without RDA, . The secondly preferred format is the new NL format of the US with much information .

• 한국산학기술학회논문지
• /
• 제22권3호
• /
• pp.446-452
• /
• 2021

우리나라 생활수준의 향상과 더불어 식품소비의 양적인 요구가 충족되면서, 세분화된 식품의 기호 성향을 충족시킬 수 있는 닭고기 소비가 증가하고 있다. 2003년 3월 축산물 품질평가원에서 고시(농림부 고시 제2003-14호)한 닭 도체 품질판정세부기준은 닭 도체 부위별 이물질 부착, 피·멍의 크기 및 중량에 따라 품질 등급을 기준을 제시하였다. 그러나 현실적으로 검사관 개개인의 주관적인 평가 기준으로 적용된 고시로 수천 마리의 닭 도체 등급판정을 유지하기가 어려운 문제점을 가지고 있다. 본 논문에서는 닭 도체 품질 세부기준에 따라 닭 도체 부위 분할하기 위해 비접촉/비파괴방식인 컴퓨터 시각 기술 알고리즘을 제안한다. 제안된 알고리즘은 실시간으로 빠르게 움직이는 닭 도체를 부위 분할하기 위하여 조명 외란에 강인하도록 보정하는 과정과 닭 도체와 배경을 구분하기 위한 EM(Expectation maximization), Erosion 및 Labeling 알고리즘, 그리고 닭 도체의 기하학적 형태를 분석하여 부위별 특징점을 찾고 점들의 위치를 계산하여 부위를 분할 할 수 있는 알고리즘을 사용하였다. 총 78마리의 닭 도체 샘플에 대하여 제안한 영상처리 알고리즘을 적용한 결과 닭 도체 부위 분할 알고리즘이 효과적임을 알 수 있었다.

• 인터넷정보학회논문지
• /
• 제24권6호
• /
• pp.91-98
• /
• 2023

오늘날 인공지능 산업이 발전함에 따라 여러 분야에 걸쳐 인공지능을 통한 자동화 및 최적화가 이루어지고 있다. 국내의 철도 분야 또한 지도 학습을 이용한 레일의 결함을 검출하는 연구들을 확인할 수 있다. 그러나 철도에는 레일만이 아닌 다른 구조물들이 존재하며 그중 선로 체결 장치는 레일을 다른 구조물에 결합시켜주는 역할을 하는 장치로 안전사고의 예방을 위해서 주기적인 점검이 필요하다. 본 논문에는 선로 체결 장치의 데이터를 이용하여 준지도 학습(semi-supervised learning)과 전이 학습(transfer learning)을 이용한 분류기를 학습시켜 선로 안전 점검에 사용되는 비용을 줄이는 방안을 제안한다. 사용된 네트워크는 Resnet50이며 imagenet으로 선행 학습된 모델이다. 레이블이 없는 데이터에서 무작위로 데이터를 선정 후 레이블을 부여한 뒤 이를 통해 모델을 학습한다. 학습된 모델의 이용하여 남은 데이터를 예측 후 예측한 데이터 중 클래스 별 확률이 가장 높은 데이터를 정해진 크기만큼 훈련용 데이터에 추가하는 방식을 채택하였다. 추가적으로 초기의 레이블된 데이터의 크기가 끼치는 영향력을 확인해보기 위한 실험을 진행하였다. 실험 결과 최대 92%의 정확도를 얻을 수 있었으며 이는 지도 학습 대비 5% 내외의 성능 차이를 가진다. 이는 제안한 방안을 통해 추가적인 레이블링 과정 없이 비교적 적은 레이블을 이용하여 분류기의 성능을 기존보다 향상시킬 수 있을 것으로 예상된다.