• Proceedings of the Korea Information Processing Society Conference
• /
• 2005.05a
• /
• pp.1713-1716
• /
• 2005

최근 3 차원 그래픽 영상의 복잡도가 점점 증가함에 따라, 가시성 선별에 관련된 연구는 3 차원 렌더링 프로세서 설계에 있어서 중요한 핵심 연구 중 하나가 되었다. 본 논문에서는 기존의 픽셀 캐쉬의 정보를 이용하여 가시성 선별을 수행하는 새로운 래스터라이제이션 파이프라인을 제안하고 있다. 제안 구조에서는 가시성 정보를 관리하기 위해서 계층적 z-버퍼 (HZB)와 같이 규모가 큰 별도의 하드웨어를 추가하지 않고, 픽셀 캐쉬에 저장되어 있는 데이터를 참조하여 주사 변환 과정에서 가시성 선별을 수행하고 있다. 캐쉬에서 접근 참조 실패된 프리미티브에 대해서는 픽셀 래스터라이제이션 파이프라인의 z-테스트 과정에서 은면 제거를 수행하도록 하였고, 선 인출 기법을 적용하여 픽셀 캐쉬의 접근 실패에 따른 손실을 줄여주었다. 실험 결과, 제안 구조는 일반 픽셀 파이프라인 구조에 비해 약 32%, HZB 구조에 비해 약 7%의 성능 향상을 보이고 있다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.11a
• /
• pp.640-642
• /
• 2021

비전 기반의 재활용-PET 선별공정에서, PET 외 물체와의 식별 성능은 물론 PET 용기 내 포함된 이물질 및 라벨, 뚜껑의 존재 여부, 색상에 대한 검출 성능은 재활용 소재 품질에 중요한 영향을 미친다. 본 연구에서는 비전 인공지능 기반의 재활용-PET 자동 선별 시스템을 제안하고, 인공지능 모델의 제작에서 감독학습의 학습 효과를 최적화하기 위한 데이터 레이블링 기법을 제안한다. 재활용대상 PET 와 이물질 파트가 포함된 용기의 컨베이어벨트 선별공정 혼입을 재현한 실험을 통해서, 재활용 소재화 물량과 순도를 최대화하기 위한 인공지능 모델 생성 방법에 대해 고찰한다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.12
• /
• pp.2305-2311
• /
• 2017

In recent years, there have been tries to open public data from various government agencies along with publicization of public information for the public interest. In other words, various kinds of electronic data generated and collected by the public institutions as a result of their work are opened in the public portal sites. However, users who use it are limited in their use of big data due to lack of understanding of data format, lack of data processing knowledge, difficulty in accessing and managing data, and lack of visualization data to understand collected and stored data. Therefore, in this study, we propose a big data collection, storing and visualization platform that can collect big data provided by various public sites using data set URL and API regardless of data format, re-process collected data using XML structure.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.1
• /
• pp.13-19
• /
• 2019

The accuracy of machine learning is greatly affected by amount of learning data and quality of data. Collecting existing Web-based learning data has danger that data unrelated to actual learning can be collected, and it is impossible to secure data transparency. In this paper, we propose a method for collecting data directly in parallel by blocks in a block - chain structure, and comparing the data collected by each block with data in other blocks to select only good data. In the proposed system, each block shares data with each other through a chain of blocks, utilizes the All-reduce structure of Parallel-SGD to select only good quality data through comparison with other block data to construct a learning data set. Also, in order to verify the performance of the proposed architecture, we verify that the original image is only good data among the modulated images using the existing benchmark data set.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2007.11a
• /
• pp.59-62
• /
• 2007

본 논문에서는 불균형 및 Under-sampling된 바이오 데이터에 대하여 클래스 구분력이 없는 특징의 소거를 통해 이후 이어질 FLDA 둥 다양한 방법론올 적용할 수 있는 방법을 제안하고자 한다. 제안하는 알고리즘은 평균과 분산을 통해 클래스의 형태를 결정하는 기존 방법론의 문제점을 회피할 수 있는 방법을 제공하며, 클래스 구분력에 중점을 두어 특정을 선별하였을 경우 선별된 특정들의 상관 계수가 높은 문제를 극복할 수 있도록 한다. 이에 따라 알고리즘이 선택한 특정집합은 서로의 특징에 대해 상관계수가 낮으며, 클래스의 구분력이 높은 특정을 갖게 된다.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.4
• /
• pp.525-531
• /
• 2009

Data mining is the process of extracting hidden patterns from collected data. At this time, for collected data which take important role as the basic information for prediction and recommendation, the process to discriminate incorrect data in order to enhance the performance of analysis result, is needed. The existing methods to discriminate unexpected data from collected data, mainly relies on methods which are based on statistics or simple distance between data. However, for these methods, the problematic point that even meaningful data could be excluded from analysis due that the environment and characteristic of the relevant data are not considered, exists. This study proposes a method to endow human heuristic knowledge with weight value through the comparison between collected data and human heuristic knowledge, and to use the value for creating a decision tree. The data discrimination by the method proposed is more credible as human knowledge is reflected in the created tree. The validity of the proposed method is verified through an experiment.

• Proceedings of the Korean Society of Surveying, Geodesy, Photogrammetry, and Cartography Conference
• /
• 2004.04a
• /
• pp.97-102
• /
• 2004

현재 4차원 정밀측정이 가능한 GPS는 데이터 처리와 사용기기 면에서 많은 발전을 이루고 있으며, 정밀한 데이터 결과를 제공하고 있다. 처리 방식에 따른 기선거리별 한계치에 대한 재검증을 실시하고자 하였으며, 측지용 GPS 수신기를 이용한 기선측정에 있어, L$_1$주파수 수신 GPS 시스템의 유효측정거리에 대하여 논하였다. 또한 일반적인 기선처리방식 이외의 다양한 처리기법들을 적용하여 각 단ㆍ중ㆍ장기선에 대해 가장 알맞은 처리방식을 도출하려고 하였으며 결과를 도출하는 과정에서 GPS 상대거리 관측데이터에 대해 양호한 데이터를 선별할 수 있는 기준 안을 마련하고자 하였다.

• Resources Recycling
• /
• v.30 no.4
• /
• pp.3-10
• /
• 2021

The field of recycling for waste electronic components, which is the typical example of an urban mine, requires the development of useful sorting techniques. In this study, a sorter based on image identification by deep learning was developed to select electronic components into four groups. They were recovered from waste printed circuit boards and should be separated to depend on the difference after treatment. The sorter consists of a workstation with GPU, camera, belt conveyor, air compressor. A small piece (less than 3.5 cm) of electronic components on the belt conveyor (belt speed: 6 cm/s) was taken and learned as teaching data. The accuracy of the image identification was 96% as kinds and 99% as groups. The optimum condition of sorting was determined by evaluating accuracies of image identification and recovery rates by blowdown when changing the operating condition such as belt speed and blowdown time of compressed air. Under the optimum condition, the accuracy of image classification in groups was 98.7%. The sorting rate was more than 70%.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.7
• /
• pp.1-7
• /
• 2021

In this paper, we propose a feature selection technique for multi-label classification. Many existing feature selection techniques have selected features by calculating the relation between features and labels such as a mutual information scale. However, since the mutual information measure requires a joint probability, it is difficult to calculate the joint probability from an actual premise feature set. Therefore, it has the disadvantage that only a few features can be calculated and only local optimization is possible. Away from this regional optimization problem, we propose a feature selection technique that constructs a low-rank space in the entire given feature space and selects features with sparsity. To this end, we designed a regression-based objective function using Nuclear norm, and proposed an algorithm of gradient descent method to solve the optimization problem of this objective function. Based on the results of multi-label classification experiments on four data and three multi-label classification performance, the proposed methodology showed better performance than the existing feature selection technique. In addition, it was showed by experimental results that the performance change is insensitive even to the parameter value change of the proposed objective function.

• 한국노년학
• /
• v.40 no.2
• /
• pp.325-340
• /
• 2020

Early diagnosis and intervention of dementia is critical to minimize future risk and cost for patients and their families. The purpose of this study was to develop and validate Figure-Copy Test(FCT), as a new dementia screening test, that can measure neurological damage and cognitive impairment, and then to examine whether the grading precesses for screening can be automated through machine learning procedure by using FCT imag es. For this end, FCT, Korean version of MMSE for Dementia Screening (MMSE-DS) and Clock Drawing Test were administrated to a total of 270 participants from normal and damaged elderly groups. Results demonstrated that FCT scores showed high internal constancy and significant correlation coefficients with the other two test scores. Discriminant analyses showed that the accuracy of classification for the normal and damag ed g roups using FCT were 90.8% and 77.1%, respectively, and these were relatively higher than the other two tests. Importantly, we identified that the participants whose MMSE-DS scores were higher than the cutoff but showed lower scores in FCT were successfully screened out through clinical diagnosis. Finally, machine learning using the FCT image data showed an accuracy of 73.70%. In conclusion, our results suggest that FCT, a newly developed drawing test, can be easily implemented for efficient dementia screening.