ÇѱÛÁ¦¸ñ(Korean Title) |
ÀǹÌÀû ºÐÇÒ ½ÉÃþÇнÀÀ» À§ÇÑ µ¥ÀÌÅͼ »ý¼º ¹æ¹ý |
¿µ¹®Á¦¸ñ(English Title) |
A Method of Generating Dataset for Sematic Segmentation Based on Deep Learning |
ÀúÀÚ(Author) |
ÀÌ»óÇù
¹ÚÀå½Ä
Sang-hyeop Lee
Jangsik Park
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 23 NO. 02 PP. 0088 ~ 0090 (2019. 10) |
Çѱ۳»¿ë (Korean Abstract) |
º» ³í¹®¿¡¼´Â ¿µ»ó ±â¹ÝÀÇ °¡½º´©Ãâ °¨Áö¸¦ À§ÇÑ ÇнÀ µ¥ÀÌÅͼ¿¡ È¿À²ÀûÀÎ GT(Ground Truth)¸¦ »ý¼ºÇϱâ À§ÇÑ ½Ã½ºÅÛÀ» Á¦¾ÈÇÑ´Ù. K-Æò±Õ ±ºÁýÈ(K-Means Clustering)¸¦ ÀÌ¿ëÇÏ¿© °¡½º´©Ãâ ¿µ¿ª°ú ÀÌ¿ÜÀÇ ¿µ¿ªÀ» ºÐÇÒ ¹× Ŭ·¯½ºÅ͸µÀ» ÇÑ´Ù. °¡½º´©Ãâ ¿µ¿ª ³»¿¡ Á¦°Å µÇÁö ¸øÇÑ ºÎºÐÀ» ÀÓ°è°ª(Threshold) ºÐÇÒ°ú Â÷¿µ»óÀ» »ç¿ëÇÏ¿© °¡½º´©Ãâ ¿µ¿ªÀ» °ËÃâÇÑ´Ù. Â÷¿µ»óÀ¸·Î ¹ß»ýÇÑ ÀâÀ½À» Á¦°ÅÇϱâ À§ÇØ ¸ðÆú·ÎÁö ¿¸² ¿¬»ê(Morphology Opening Operations)À» »ç¿ëÇÑ´Ù. ±âÁ¸ÀÇ ¼öÀÛ¾÷º¸´Ù º» ³í¹®¿¡¼ Á¦¾ÈÇÏ´Â ÀÚµ¿À¸·Î »ý¼ºµÈ GT°¡ È¿À²ÀûÀÓÀ» °¢ ´Ü°èº° ºñ±³¸¦ ÅëÇØ ¿ì¼öÇÔÀ» º¸ÀδÙ. |
¿µ¹®³»¿ë (English Abstract) |
In this paper, we propose a system for generating an efficient GT(Ground Truth) in the training dataset for image-based gas leak detection. K-Means Clustering is used to segment and cluster gas leaks and other regions. The gas leakage region is detected using a threshold and a difference image of the portion that cannot be removed in the gas leakage region. Morphology Opening Operations are used to remove the noise caused by the difference image. It shows that the automatically generated GT proposed in this paper is more efficient than the existing manual work by comparing each step. |
Å°¿öµå(Keyword) |
Semantic segmentation
Deep learning
Ground truth
K-means algorithm
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|