• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ÇÐȸÁö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ÇÐȸÁö > µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) ´ë±Ô¸ð Çؾç°üÃø µ¥ÀÌÅÍ¿¡¼­ AutoEncoder¸¦ È°¿ëÇÑ °ú°Å µ¥ÀÌÅÍÀÇ ºü¸¥ °Ë»ö
¿µ¹®Á¦¸ñ(English Title) Fast Retrieval of Past Similar data using An AutoEncoder in Large Oceanographic Observation Data
ÀúÀÚ(Author) Á¤¿øÁØ   ±Ç¿ÀÈì   ¼ÛÇÏÁÖ   Won-Joon Jeong   Oh-Heum Kwon   Ha-Joo Song  
¿ø¹®¼ö·Ïó(Citation) VOL 38 NO. 03 PP. 0096 ~ 0106 (2022. 12)
Çѱ۳»¿ë
(Korean Abstract)
Çؾç°üÃø µ¥ÀÌÅÍ´Â °ú°Å ¼ö½Ê ³â°£ÀÇ µ¥ÀÌÅ͸¦ º¸°üÇϱ⠶§¹®¿¡ ±× Å©±â´Â ¼ö½Ê ±â°¡¹ÙÀÌÆ® ÀÌ»ó¿¡ À̸£´Â °æ¿ì°¡ ÈçÇÏ´Ù. µû¶ó¼­ ÀÏÁ¤ ½ÃÁ¡ÀÇ °üÃø µ¥ÀÌÅÍ¿Í À¯»çÇÑ °ú°Å °üÃøÀڷḦ ã±â À§Çؼ­ µ¥ÀÌÅ͸¦ Çϳª¾¿ ºñ±³ÇÏ¿© °Ë»öÇÏ´Â °ÍÀº ¼ö ½ÊºÐ ÀÌ»óÀÇ ½Ã°£ÀÌ ¼Ò¿äµÈ´Ù. º» ³í¹®Àº µö·¯´× ¾Ë°í¸®Áò Áß ÇϳªÀÎ ¿ÀÅäÀÎÄÚ´õ¸¦ ÀÌ¿ëÇÑ Â÷¿øÃà¼Ò ±â¹ýÀ» »ç¿ëÇØ °Ë»ö ¼Ò¿ä½Ã°£À» ´ÜÃà½Ãų ¼ö ÀÖ´Â ±â¹ýÀ» Á¦½ÃÇÑ´Ù. Á¦¾È±â¹ýÀº °¢°¢ÀÇ °üÃø µ¥ÀÌÅÍ¿¡ ´ëÇØ ¿ø µ¥ÀÌÅ͸¦ ¾ÐÃàÇÑ µÚ ¿ÀÅäÀÎÄÚ´õ¸¦ ÀÌ¿ëÇÏ¿© Ư¡ º¤Å͸¦ ÃßÃâÇÏ°í Ư¡ º¤ÅÍ ÁýÇÕÀ» ±¸ÃàÇÑ´Ù. °Ë»ö½Ã¿¡´Â °Ë»ö ±âÁØ µ¥ÀÌÅÍ¿¡ ´ëÇÑ Æ¯Â¡ º¤ÅÍ¿Í ÀúÀåµÈ Ư¡ º¤ÅÍ ÁýÇÕ °£ÀÇ À¯»çµµ ºñ±³¸¦ ÅëÇØ È帱ºÀ» ÃßÃâÇÑ´Ù. ±×¸®°í È帱º°ú ±âÁØ µ¥ÀÌÅ͸¦ ¿ø µ¥ÀÌÅÍ ¼öÁØ¿¡¼­ ºñ±³ÇÏ¿© °¡Àå À¯»çÇÑ °ÍÀ» ã´Â´Ù. netCDF Æ÷¸ËÀ¸·Î ÀúÀåµÈ À§¼º °üÃø µ¥ÀÌÅ͸¦ »ç¿ëÇÏ¿© ½ÇÇèÀ» ¼öÇàÇÏ¿´´Ù. Á¦¾È ±â¹ýÀº Ãß°¡ÀûÀÎ ÀúÀå°ø°£À» Àû°Ô Â÷ÁöÇϸ鼭µµ °Ë»ö¼Óµµ Ãø¸é¿¡¼± ¾à 4 ¹è Á¤µµÀÇ ¼º´É Çâ»óÀ» ´Þ¼ºÇÏ¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
Oceanographic observation data often reaches more than tens of gigabytes in size because it holds data from the past decades. Therefore, it takes several tens of minutes or more to compare and search data one by one to find historical observations similar to observation data at a certain point in time. This paper presents a technique that can shorten the search time by using a dimension reduction technique using an autoencoder, one of the deep learning algorithms. The proposed technique compresses the original data for each observation data and then uses an autoencoder to extract feature vectors and construct an array of feature vectors. At the time of search, a candidate group is extracted by comparing the similarity between the feature vector of the search data and the stored feature vector array. The candidate group and the search data are compared at the original scale to find the most similar one. We conducted performance tests using satellite observation data stored in netCDF format. The proposed method showed performance improvement of around four times in terms of search speed while occupying less additional storage space.
Å°¿öµå(Keyword) netCDF   ¿ÀÅäÀÎÄÚ´õ   À¯»çµµ °Ë»ö   AutoEncoder   Similarity Search  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå