| 학회 |
한국화학공학회 |
| 학술대회 |
2019년 가을 (10/23 ~ 10/25, 대전컨벤션센터) |
| 권호 |
25권 2호, p.1693 |
| 발표분야 |
열역학 (Thermodynamics) |
| 제목 |
Extracting Hidden Signals in Chemical Sensors using Deep Learning Based Anomaly Detection |
| 초록 |
Deep learning have shown remarkable performances to extract features automatically in various fields. Especially, auto-encoder have been applied to extract patterns of normal states which is used to classify the normal from the anomaly in anomaly detection. Gas sensing is a kind of anomaly detection in which anomalous gas is detected from normal baseline. Limit of detection (LOD) is used to determine what is normal or abnormal in chemical and biological sensors, but this is only based on amplitude. In this work, we apply deep learning-based anomaly detection to metal film-based gas sensing to enhance selectivity by monitoring not only amplitude also overall patterns of signals. With this approach, we show that deep neural network can find the hidden signal under LOD, which lead to enhanced sensing for H2. Our approaches are end-to-end (no need for feature engineering) and easy-and-fast (need simple architecture and only few minutes for training), which can take a crucial role for next-generation chemical and biological sensing. |
| 저자 |
이유한1, 조수연2, 이상원1, 강호형1, 김재훈3, 최정훈1, 류 진1, 주희은1, 정희태1, 김지한1
|
| 소속 |
1한국과학기술원, 2Massachusetts Institute of Technology, 3삼성SDS |
| 키워드 |
분자모델링 및 전산모사 |
| E-Mail |
|
| VOD |
VOD 보기 |
| 원문파일 |
초록 보기 |
