• 한국군사과학기술학회지
• /
• 제24권3호
• /
• pp.281-292
• /
• 2021

With the development of deep learning technology, researchers and technicians keep attempting to apply deep learning in various industrial and academic fields, including the defense. Most of these attempts assume that the data are balanced. In reality, since lots of the data are imbalanced, the classifier is not properly built and the model's performance can be low. Therefore, this study proposes cost-sensitive learning as a solution to the imbalance data problem of image classification in the defense field. In the proposed model, cost-sensitive learning is a method of giving a high weight on the cost function of a minority class. The results of cost-sensitive based model shows the test F1-score is higher when cost-sensitive learning is applied than general learning's through 160 experiments using submarine/non-submarine dataset and warship/non-warship dataset. Furthermore, statistical tests are conducted and the results are shown significantly.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제14권9호
• /
• pp.3797-3822
• /
• 2020

The inspection of cracks on the surface of tunnel linings is a common method of evaluate the condition of the tunnel. In particular, determining the thickness and shape of a crack is important because it indicates the external forces applied to the tunnel and the current condition of the concrete structure. Recently, several automatic crack detection methods have been proposed to identify cracks using captured tunnel lining images. These methods apply an image-segmentation mechanism with well-annotated datasets. However, generating the ground truths requires many resources, and the small proportion of cracks in the images cause a class-imbalance problem. A weakly annotated dataset is generated to reduce resource consumption and avoid the class-imbalance problem. However, the use of the dataset results in a large number of false positives and requires post-processing for accurate crack detection. To overcome these issues, we propose a crack detection method using a ternary classifier. The proposed method significantly reduces the false positive rate, and the performance (as measured by the F1 score) is improved by 0.33 compared to previous methods. These results demonstrate the effectiveness of the proposed method.

• 천문학회보
• /
• 제37권1호
• /
• pp.79.1-79.1
• /
• 2012

We present here $UBVI$ and H${\alpha}$ photometric results of stellar sources in the bubble nebula NGC 7635. The early type members are selected from the photometric membership criteria. H${\alpha}$ photometry allows us to detect 11 pre-main sequence candidates with H${\alpha}$emission. In addition, we performed PSF photometry for the Spitzer IRAC and MIPS 24${\mu}m$ images from archive (program ID 20726, PI: J. Hester) in order to search for the young stellar objects (YSOs). Total 19 sources are classified as YSOs (7 class I, 11 class II, and 1 transitional disk candidates) in the color-color diagrams according to the classification scheme of Gutermuth et al.. Among them, 7 YSOs have counterparts in optical photometric data. These stars can be divided into two groups at given color indices. It implies that there occurred the star formation events more than twice. We would like to discuss the star formation history in the bubble nebula using the results from SED fitter (Robitaille et al.), color composite image from IRAC bands, and spatial distribution of early type stars and YSOs.

• 보존과학회지
• /
• 제36권2호
• /
• pp.73-81
• /
• 2020

초분광 이미지 분석은 석조문화재 손상지도 제작 시 화학적·생물학적 오염의 정량적인 면적산출이 어려운 단점을 보완하기 위해 제안되었다. 이 연구에서는 다양한 표면오염물이 나타나는 경주굴불사지 석조사면불상을 대상으로 초분광 이미지 분석을 수행하였다. 이때 화학적·생물학적 오염은 색상과 형태에 따라 10가지 범주로 구분하였고 범주의 참조 이미지 제작법을 제시하였다. 또한 오염물을 분류하기 위해 SAM 알고리듬을 사용하고 관심영역을 사용한 분류방법(Method A)과 영상에서 추출한 분광정보를 라이브러리로 구축하여 분류하는 방법(Method B)을 적용하였다. 분류된 이미지를 참조 이미지와 비교한 결과, 정확도는 분류 방법에 따라 52.07%와 63.61%, Kappa 지수는 0.43과 0.55로 산출되었고, 분류시 오분류 화소는 동일한 계열의 오염에 분포하는 경향을 보인다.

• 한국정보통신학회:학술대회논문집
• /
• 한국정보통신학회 2021년도 춘계학술대회
• /
• pp.614-616
• /
• 2021

세계는 안구 질병 치료, 시력 회복 서비스, 훈련된 안과 전문의의 부족 등 안과 측면에서 어려움에 직면해 있다. 안구 병리를 조기에 발견하고 진단하면 시각 장애를 예방할 수 있다. 하지만 기존의 망막 영상 공개 데이터 세트는 임상에서 발견되는 다양한 질병으로 구성되어 있지 않기 때문에 다양한 안구 질환을 분류하는 방법을 개발하기가 어렵다. 본 연구는 2021 ISBI challenge에서 공개된 데이터 세트인 Retinal Fundus Multi-disease Image Dataset (RFMiD) 을 이용하여 안구 질환을 분류하는 방법을 제안한다. 본 연구의 목표는 망막 이미지를 정상, 비정상 범주로 선별하기 위한 강력하고 일반화 가능한 모델을 개발하는 것이다. 제안된 모델의 성능은 수신자 조작 특성 곡선 아래 면적 점수로 비공개 테스트 데이터 세트에 대해 0.9782의 값을 보여준다.

• 한국과학교육학회지
• /
• 제29권3호
• /
• pp.329-347
• /
• 2009

Science curriculums for elementary schools were, traditionally, developed to be balanced in content and contain equal proportions of the four branches of science: physics, chemistry, biology, and earth science. To develop a successful science curriculum, we asked some questions about how elementary students recognize these branches and about what they think of the domains of science in the science curriculum. Our study was designed to investigate how elementary students classify the domains of science in the curriculum. Previous research (Lee et al., 2001) seemed not to be successful, because verbal expressions in that research might be inappropriate for elementary students who were unaccustomed to the technical language of science. For this reason, instead of using only words, we developed image card instruments, made of picture duplicates of the introductory covers of each unit in the 3$^{rd}$, 4$^{th}$, and 5$^{th}$ grades' science textbooks. We asked students to classify these cards into their own categories and record the reasons for classifying them. The ratio and distribution of the units was then analyzed to identify their view of the science domains. 30% of the 4$^{th}$ grade students created the following categories: 'nature,' 'observation,' 'seasons,' 'living things,' 'sounds,' 'separating,' and 'the things necessary for everyday life'. In the case of the 5$^{th}$ grade, over 30% created the categories of 'living things,' 'weight,' and 'water.' Over 30% of the 6$^{th}$ grade created the categories of 'nature,' 'light,' 'water,' 'living things,' 'solution,' 'fire,' 'properties of an object,' and 'experiment.' Upon scrutinizing the above results, we discovered that the science domains selected by students into three types of domains: academic contents and concepts; activities related to a science class; and lessons and experiences in students ' lives. The last category was a new, complex kind of domain. We concluded that students did not utilize the four branches of science when constructing their own domains of science. Instead, they created many alternative domains, which reflected students' thoughts of and their experiences. The educational needs of elementary students suggest that when organizing science curriculum as 25 % allocation of the four science branches, newly-created domains should be considered.

• 대한원격탐사학회지
• /
• 제27권6호
• /
• pp.637-651
• /
• 2011

This study presents a rule-based urban image classification method for time series analysis of changes in the vicinity of Asan-si and Cheonan-si in Chungcheongnam-do, using Landsat satellite images (1991-2006). The area has been highly developed through the relocation of industrial facilities, land development, construction of a high-speed railroad, and an extension of the subway. To determine the yearly changing pattern of the urban area, eleven classes were made depending on the trend of development. An algorithm was generalized for the rules to be applied as an unsupervised classification, without the need of training area. The analysis results show that the urban zone of the research area has increased by about 1.53 times, and each correlation graph confirmed the distribution of the Built Up Index (BUI) values for each class. To evaluate the rule-based classification, coverage and accuracy were assessed. When Optimal allowable factor=0.36, the coverage of the rule was 98.4%, and for the test using ground data from 1991 to 2006, overall accuracy was 99.49%. It was confirmed that the method suggested to determine the maximum allowable factor correlates to the accuracy test results using ground data. Among the multiple images, available data was used as best as possible and classification accuracy could be improved since optimal classification to suit objectives was possible. The rule-based urban image classification method is expected to be applied to time series image analyses such as thematic mapping for urban development, urban development, and monitoring of environmental changes.

• 통합자연과학논문집
• /
• 제9권2호
• /
• pp.155-160
• /
• 2016

Applying image processing techniques to education, the face of the learner is photographed, and expression and movement are detected from video, and the system which estimates degree of concentration of the learner is developed. For one learner, the measuring system is designed in terms of estimating a degree of concentration from direction of line of learner's sight and condition of the eye. In case of multiple learners, it must need to measure each concentration level of all learners in the classroom. But it is inefficient because one camera per each learner is required. In this paper, position in the face region is estimated from video which photographs the learner in the class by the difference between frames within the motion direction. And the system which detects the face direction by the face part detection by template matching is proposed. From the result of the difference between frames in the first image of the video, frontal face detection by Viola-Jones method is performed. Also the direction of the motion which arose in the face region is estimated with the migration length and the face region is tracked. Then the face parts are detected to tracking. Finally, the direction of the face is estimated from the result of face tracking and face parts detection.

• 보존과학회지
• /
• 제38권1호
• /
• pp.1-13
• /
• 2022

Colour consistency is an important consideration when selecting pigments used on works of art. In this study, we analyse the colour difference between two sets of synthetic blue glass pigments acquired at least 8 years apart from the same manufacturer in Japan. The old pigment set (unused, dry powder with four different grain sizes) appears faded compared to the new set. These pigments are made available for artistic use, commonly in Nihonga or Japanese paintings. Raman spectroscopy and SEM-EDS results characterize these pigments as cobalt aluminate spinels dissolved in leaded glaze, a special class of complex coloured inorganic pigments that is not well-understood in the field of conservation. Colour difference between the old and new pigments with four different grain sizes were quantified by analysing photomicrographs with image analysis software. Blue pigments with coarse and extremely fine grains showed significant colour change compared to pigments with medium and fine grain sizes. The high occurrence of crystallites in the finer grains give a final colour that is bluer and lighter. Possible causes for the colour difference including manufacturing methods and storage environment are discussed.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권8호
• /
• pp.4300-4314
• /
• 2019

With the development of video-related applications, media content has increased dramatically through applications. There is a substantial amount of near-duplicate videos (NDVs) among Internet videos, thus NDVR is important for eliminating near-duplicates from web video searches. This paper proposes a novel NDVR system that supports large-scale retrieval and contributes to the efficient and accurate retrieval performance. For this, we extracted keyframes from each video at regular intervals and then extracted both commonly used features (LBP and HSV) and new image features from each keyframe. A recent study introduced a new image feature that can provide more robust information than existing features even if there are geometric changes to and complex editing of images. We convert a vector set that consists of the extracted features to binary code through a set of hash functions so that the similarity comparison can be more efficient as similar videos are more likely to map into the same buckets. Lastly, we calculate similarity to search for NDVs; we examine the effectiveness of the NDVR system and compare this against previous NDVR systems using the public video collections CC_WEB_VIDEO. The proposed NDVR system's performance is very promising compared to previous NDVR systems.