µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
Àû´ëÀû ¿ÀÅäÀÎÄÚ´õ ±â¹Ý ÈÀç À§Çè °Ç¹° °ËÃâ ±â¹ý |
¿µ¹®Á¦¸ñ(English Title) |
Adversarial Autoencoder-based Fire-risk Building Detection Technique |
ÀúÀÚ(Author) |
À̱æÀç
Gil-Jae Lee
±èÇÑÁØ
Han-Joon Kim
½Å½Â¿±
Seung-Yeop Shin
±èÇÑÁØ
Han-Joon Kim
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 35 NO. 03 PP. 0087 ~ 0097 (2019. 12) |
Çѱ۳»¿ë (Korean Abstract) |
°Ç¹°ÀÇ ÈÀ縦 ¿¹¹æÇϱâ À§Çؼ´Â °Ç¹°¿¡ ´ëÇÑ ¼Ò¹æ Á¡°ËÀÌ ÇÊ¿äÇÏ´Ù. ÀÌ ¶§, ÈÀç À§ÇèÀÌ ³ôÀº °Ç¹°À» ½Äº°ÇÒ ¼ö ÀÖ´Ù¸é ¼Ò¹æ ÇàÁ¤·ÂÀ» È¿À²ÀûÀ¸·Î ¹èºÐÇÒ ¼ö ÀÖ´Ù. º» ³í¹®¿¡¼´Â ¿ÀÅäÀÎÄÚ´õ ±â¹ÝÀÇ ÀÌ»óÄ¡ °ËÃâ ±â¹ýÀ» ÀÌ¿ëÇÏ¿© ÈÀç À§Çè °Ç¹°À» °ËÃâÇÑ´Ù. ¿ÀÅäÀÎÄÚ´õ ±â¹ÝÀÇ ÀÌ»óÄ¡ °ËÃâÀº Á¤»ó µ¥ÀÌÅÍ¿¡ ´ëÇÑ º¹¿ø ¿À·ù´Â ³·°í ÀÌ»ó µ¥ÀÌÅÍ¿¡ ´ëÇÑ º¹¿ø ¿À·ù´Â ³ô´Ù´Â ¿ø¸®¿¡ ±Ù°ÅÇÑ´Ù. º» ³í¹®Àº Àû´ëÀû ¿ÀÅäÀÎÄÚ´õ¿¡ Latent GANÀ» Ãß°¡ÇÏ¿© ÀÌ»ó µ¥ÀÌÅÍ¿¡ ´ëÇÑ º¹¿ø ¿À·ù°¡ ³ôµµ·Ï ¼³°èµÈ ¿ÀÅäÀÎÄÚ´õ ¸ðµ¨À» Á¦¾ÈÇÑ´Ù. ½ÇÇèÀ» ÅëÇØ º» ³í¹®¿¡¼ Á¦¾ÈÇÏ´Â ¸ðµ¨ÀÌ ÈÀç À§Çè °Ç¹°À» °ËÃâÇϴµ¥ ÀÖ¾î ¿ì¼öÇÑ ¼º´ÉÀ» º¸ÀÓÀ» Áõ¸íÇÏ¿´´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
In order to prevent a fire in the building, fire safety checks on the building are necessary. If the fire-risk buildings can be detected, the fire authority can efficiently perform a fire service. In this paper, an autoencoder-based outlier detection technique is used to detect fire-risk buildings. Autoencoder-based outlier detection is based on the principle that the reconstruction error for normal data is low and the reconstruction error for abnormal data is high. In this paper, we propose an autoencoder model designed for high reconstruction error for abnormal data by adding Latent GAN to adversarial autoencoder. The experiments show that ¡®Autoencoder with latent GAN¡¯is excellent at detecting fire-risk buildings.
|
Å°¿öµå(Keyword) |
µ¥ÀÌÅÍ Áõ½Ä
ÅؽºÆ®µ¥ÀÌÅÍ
½Ã¸Çƽ ÅÙ¼°ø°£¸ðµ¨
±â°èÇнÀ. ÇнÀµ¥ÀÌÅÍ
Data Augmentation
Text Data
Training Data
Semantic Tensor Space Model
Machine Learning
ÀÌ»óÄ¡ °ËÃâ
¿ÀÅäÀÎÄÚ´õ
Àû´ëÀû »ý¼º¸Á
Anomaly detection
autoencoder
generative adversarial networks
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|