• 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 : 17 / 17

ÇѱÛÁ¦¸ñ(Korean Title) °áÃø°ªÀ» Æ÷ÇÔÇÑ ¼¾¼­ ½ºÆ®¸²¿¡ ´ëÇÑ ¾îÅÙ¼Ç ¸ÞÄ¿´ÏÁò ¹× ÇÕ¼º°ö ½Å°æ¸Á ±â¹ÝÀÇ ÆÐÅÏ ºÐ·ù ±â¹ý
¿µ¹®Á¦¸ñ(English Title) Pattern Classification based on Attention Mechanism and CNN for Sensor Stream Data including Missing Values
ÀúÀÚ(Author) ÀÌÀºÁø   Eunjin Lee   ¿À¼Ò¿¬   Soyeon Oh   À̹μö   Minsoo Lee  
¿ø¹®¼ö·Ïó(Citation) VOL 36 NO. 02 PP. 0056 ~ 0068 (2020. 08)
Çѱ۳»¿ë
(Korean Abstract)
´Ù¾çÇÑ ¼¾¼­·ÎºÎÅÍ ¼öÁýµÈ IoT ½ºÆ®¸² µ¥ÀÌÅÍ ºÐ¼®Àº ´ëÇ¥ÀûÀÎ ºñ¼±Çü ºÐ¼® ¹®Á¦·Î, ÃÖ±Ù ÀÌ·¯ÇÑ ¹®Á¦µéÀÇ ÇØ°á¿¡ ÇÕ¼º°ö ½Å°æ¸Á(Convolutional Neural Network, CNN)À» ºñ·ÔÇÑ µö·¯´× ±â¹ýµéÀ» ´Ù¹æ¸éÀ¸·Î Àû¿ëÇÏ°í ÀÖ´Ù. ¶ÇÇÑ, IoT ¼¾¼­ ½ºÆ®¸² µ¥ÀÌÅÍ´Â ±× ¼öÁý °úÁ¤¿¡¼­, ¼¾¼­¿Í ¼­¹ö °£ÀÇ Åë½Å Àå¾Ö ¶Ç´Â ¼¾¼­ÀÇ Çϵå¿þ¾îÀû °áÇÔ µîÀ¸·Î ÀÎÇÑ °áÃø°ª Áï, ¼Õ½Ç µ¥ÀÌÅ͸¦ Æ÷ÇÔÇÏ´Â °æ¿ì°¡ ¸¹À¸¸ç, ÀÌ·¯ÇÑ ¼Õ½Ç µ¥ÀÌÅÍ´Â ºÐ¼®ÀÇ Á¤È®µµ¸¦ °¨¼Ò½ÃŲ´Ù. ÇÑÆí, ´Ù¾çÇÑ ¼¾¼­ ½ºÆ®¸² µ¥ÀÌÅÍ Áß, ·çÇÁ ¼¾¼­¸¦ ÅëÇØ ¼öÁýµÈ ±³Åë·® µ¥ÀÌÅÍ ºÐ¼®Àº µµ½Ã °èȹ, ±³Åë °øÇÐ, ´Ù¾çÇÑ ±³Åë ¹× À§Ä¡ ±â¹Ý ¼­ºñ½ºÀÇ ±¸Çö µî¿¡ È°¿ëµÈ´Ù. ±×·¯³ª ·çÇÁ ¼¾¼­¸¦ ÅëÇÑ ±³Åë·® µ¥ÀÌÅÍ ¼öÁý °úÁ¤¿¡¼­ °áÃø°ªÀÌ ¹ß»ýÇÏ´Â °æ¿ì°¡ ¸¹´Ù. º» ³í¹®¿¡¼­´Â ÀÌ·¸°Ô °áÃø°ªÀÌ Æ÷ÇÔµÈ ¼¾¼­ ½ºÆ®¸² µ¥ÀÌÅÍÀÇ ÆÐÅÏ ºÐ·ù Á¤È®µµ¸¦ ³ôÀ̱â À§ÇÑ ±â¹ýÀ» Á¦¾ÈÇÑ´Ù. Á¦¾ÈÇÏ´Â ±â¹ýÀº ÇÕ¼º°ö ½Å°æ¸Á ±â¹ÝÀÇ ÆÐÅÏ ºÐ·ù ¸ðµ¨¿¡ ¾îÅÙ¼Ç ¸ÞÄ¿´ÏÁò(Attention Mechanism)À» µµÀÔÇÏ¿© ºñ¼Õ½Ç µ¥ÀÌÅÍ¿¡ ´ëÇÑ °¡ÁßÄ¡¸¦ ºÎ¿©ÇÔÀ¸·Î½á °áÃø°ªÀ¸·Î ÀÎÇÑ Á¤È®µµÀÇ ¼Õ½ÇÀ» º¸¿ÏÇÑ´Ù. º» ³í¹®¿¡¼­´Â °áÃø°ªÀÇ ¹ß»ýÀÌ ÀæÀº ·çÇÁ ¼¾¼­ ±â¹ÝÀÇ ±³Åë·® µ¥ÀÌÅ͸¦ ´ë»óÀ¸·Î Á¦¾ÈÇÏ´Â ÆÐÅÏ ºÐ·ù ±â¹ýÀ» Àû¿ëÇÏ¿´°í, Á¦¾ÈÇÏ´Â ±â¹ýÀÌ °áÃø°ªÀ» Æ÷ÇÔÇÑ ¼¾¼­ ½ºÆ®¸² µ¥ÀÌÅÍ¿¡ ´ëÇÑ ÆÐÅÏ ºÐ·ù Á¤È®µµ¸¦ Çâ»ó½Ãų ¼ö ÀÖÀ½À» ½ÇÇèÀ» ÅëÇØ È®ÀÎÇÏ¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
Analysis for IoT stream data collected from various sensors is a typical non-linear analysis problem, and recently, deep learning techniques including convolutional neural networks have been applied to these problems in various ways. In addition, the IoT sensor stream data often includes missing data, that is, loss data due to a communication failure between the sensor and the server or a hardware defect of the sensor during the collection process, and such loss data reduces the accuracy of analysis. Meanwhile, among the various sensor stream data, the analysis of traffic volume data collected through the loop coil sensor is used for urban planning, traffic engineering, and implementation of various traffic and location-based services. However, during the process of collecting traffic data through the loop coil sensor, missing values are often generated. In this paper, we propose a method to increase the accuracy of pattern classification of sensor stream data containing missing values. The proposed method compensates for the loss of accuracy due to missing values assigning weights to non-loss data by applying attention mechanism to the pattern classification model based on the convolutional neural network. In this paper, the proposed pattern classification method is applied to traffic volume data measured by loop coil sensors that frequently generate missing values, and it was confirmed through experiments that the proposed method can improve the accuracy of pattern classification for sensor stream data including missing values.
Å°¿öµå(Keyword) IoT   ½ºÆ®¸² µ¥ÀÌÅÍ   µö·¯´×   ÇÕ¼º°ö ½Å°æ¸Á   ¾îÅÙ¼Ç ¸ÞÄ¿´ÏÁò   Stream Data   Deep Learning   CNN   Attention Mechanism  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå