http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
정대현(Daehyeon Jeong),김영호(YoungHo Kim),김창현(ChangHyeon Kim),이후상(HooSang Lee),유홍제(Hongje Yu),정원호(Wonho Jung),오현석(Hyunseok Oh),류제하(Jeha Ryu) 제어로봇시스템학회 2018 제어·로봇·시스템학회 논문지 Vol.24 No.9
Damage in fish farming nets can lead to serious losses and/or adverse environmental impact. Nonetheless, detecting such damage is challenging. Human experts could inspect the nets, but this process is costly and time-consuming. Alternatively, remotely operated underwater vehicles (ROV) can be used to inspect the fishnets. By using advanced deep-learning techniques for autonomous navigation and object detection, fishnets can be inspected efficiently while minimizing human intervention. In this paper, a deep convolutional neural networks (CNN) is employed to classify images of torn and normal fishnets. Training deep CNN models requires numerous image data, whereas a limited amount of fishnet images are available. To resolve the dearth of available data, data-augmentation techniques are adopted to generate images of torn and normal fishnets. The trained CNN model shows high accuracy for classifying the given augmented test dataset.
구조물 건전성 진단에서 데이터 부족 문제 극복을 위한 심층 생성 모델의 활용
정원호(Wonho Jung),정대현(Daehyeon Jeong),김영호(Youngho Kim),김창현(Changhyeon Kim),이후상(Hoosang Lee),유홍제(Hongje Yu),류제하(Jeha Ryu),오현석(Hyunseok Oh) 대한기계학회 2019 大韓機械學會論文集A Vol.43 No.3
딥러닝 알고리즘 훈련을 위해서 충분한 양의 데이터 확보가 필수적이다. 그러나, 공학시스템에서 데이터 취득은 매우 어렵거나, 상황에 따라 불가능한 경우가 존재한다. 이러한 데이터 부족 문제는 딥러닝 알고리즘 개발에 큰 걸림돌이 되고 있다. 본 논문은 구조물 건전성 진단을 위한 딥러닝 알고리즘 개발에서 발생하는 데이터 부족 문제 해결을 시도하였다. 깊은 생성 모델을 구축하고 딥러닝 학습을 위한 훈련 데이터를 생성하는 방법을 제안한다. 제안된 방법의 성능을 검증하기 위해 수상 양식장 어망 데이터를 바탕으로 사례 연구를 진행하였다. 본 연구는 제안된 심층 생성 모델을 통해 데이터를 직접 만들어 냄으로써 구조물 건전성 진단에서 발생되는 데이터 부족 문제 해결에 기여할 것으로 기대된다. A sufficient amount of data are required for training deep learning algorithms. However, in engineered systems, data acquisition is difficult or sometimes not feasible. A dearth of data is one of the major challenges for the development of deep learning algorithms. This paper proposes a deep generative model to generate pseudo data that emulate real data. To verify the performance of the proposed model, a case study is conducted using aquaculture fishnet image data. We demonstrate that the insufficient data problem in structural health monitoring can be relieved by generating data through the proposed deep generative model. The reliability of engineered systems can be improved by incorporating the deep learning algorithms developed with real data as well as generated data.
오미혜(Meehye Oh),윤여성(Yeoseong Yoon),오광철(Kwangchul Oh),김덕진(Deokjin Kim),최성규(Seongkyu Choi),이후상(Hoosang Lee),신재섭(Jaesup Shin) 한국자동차공학회 2007 한국자동차공학회 춘 추계 학술대회 논문집 Vol.- No.-
This study is estimate the application of polycarbonate polymer glazing application for automotive. The polymer glazing made out of poly carbonate by compressed process. The manufacturing system is useful to the trend of automotive industry. The comfortable and safety test is very important matter for car application of polymer glazing. The one test is NVH test and the others is fatigue test. The NVH tests are noise and vibration test at accelerated mode and model road. Each test is make comparison between glass and polymer. The noise property is shown about 70~75㏈. So, there is no difference glass and polycarbonate. The vibration property compared with polymer, glass is batter about 50%. But there is insensitive, no difference between glass and polymer. And fatigue test is very superior in polycarbonate glazing. It is shown 80㎜ of displacement and the restoration is reach stage of perfection. We propose the test guideline about car application of polymer glazing.