¼ö·Ï»çÇ× |
Çѱ¹°æ°üÇÐȸÁö , Vol.16 No.1(2024-06) |
ÁÖÁ¦¾î |
ÀÚ¿¬À¯»ê; õ¿¬±â³ä¹°; ¹æÀç; ¾ÈÀü°ü¸®; ±â°èÇнÀ ; Natural Heritage; Natural Monument; Disaster Prevention; Safety Management; Machine Learning |
¿ä¾à1 |
º» ¿¬±¸´Â Ç×°ø Ãʺб¤ ¿µ»ó¿¡¼ ´ë¸éÀû ÀÚ¿¬À¯»êÀÇ ½Ä»ý °æ°ü°ú ½Ä¹° »ýÀ°¿¡ ¿µÇâÀ» ÁÖ´Â ¾Æ±î½Ã³ª¹«(Robinia pseudoacacia L.), Ħ(Pueraria lobata (Willd.) Ohwi), ȯ»ïµ¢±¼(Humulus japonicus Siebold & Zucc.)ÀÇ Å½Áö¿Í ºÐ·ù¿¡ »ç¿ëµÉ ºÐ±¤¶óÀ̺귯¸® ±¸ÃàÀ» ¸ñÀûÀ¸·Î ÇÏ¿´À¸¸ç, ±× °á°ú´Â´ÙÀ½°ú °°´Ù. ù°, ºÐ±¤¶óÀ̺귯¸® ±¸Ãà ½Ã ÃøÁ¤ ´ë»óÀÇ ÇüÅÂ, ÃøÁ¤ ¹è°æÆÇÀÇ ÀçÁú°ú »ö»ó, ÃøÁ¤ °Å¸® µîÀ»°í·ÁÇÑ Ãëµæ±â¹ýÀ» Á¦½ÃÇÏ¿´´Ù. Ãëµæ ½Ã Ç×°ø Ãʺб¤ ¿µ»ó¿¡¼ º¸ÀÌ´Â ½Ä¹°ÀÇ Çü»ó°ú À¯»çÇÏ°ÔÀÙÀÇ ¾Õ¸é°ú µÞ¸é ±×¸®°í °¡Áö¸¦ È¥ÇÕÇÏ¿© °ËÀº»ö õÀ» ¹è°æÀ¸·Î ¹èÄ¡ÇÏ°í, ¹Ý»ç°ª º¸Á¤À» À§ÇÑÈÀÌÆ®º¸µå ¹× ½Ä¹°°ú Àåºñ »çÀÌÀÇ ÃÔ¿µ°Å¸®´Â Áö»óºÐ±¤°èÀÇ Á¦¿ø°ú ¿ÜºÎ ȯ°æÀ» °í·ÁÇÏ¿© 10§¯ ·Î ¼³Á¤ÇÏ´Â °ÍÀÌ È¿°úÀûÀ̾ú´Ù. µÑ°, ÃëµæµÈ 3Á¾ÀÇ ºÐ±¤¶óÀ̺귯¸®ÀÇ ºÐ»êºÐ¼®°ú ±â°èÇнÀÀ»ÅëÇØ ºñ±³½Ä¹° 4Á¾ÀÇ ºÐ±¤¶óÀ̺귯¸®¿Í ºÐº°·ÂÀÌ ÀÖÀ½À» ¹àÇû´Ù. ¾Æ±î½Ã³ª¹«¿Í ĦÀº RF(Random Forest)¿Í SVM(Support Vector Machine)ÀÇ ºÐ¼® °á°ú, 8¡12%·Î ³·Àº ¿À·ùÀ²¿¡¼ ºÐº°·ÂÀÌ ³ªÅ¸³ª ÇâÈÄ ºÐ±¤¶óÀ̺귯¸®·Î È°¿ë °¡´É¼ºÀÌ ½Ã»çµÇ¾ú´Ù. ƯÈ÷, ĦÀº ¸ðµç ÆÄÀå´ë¿¡¼ ºñ±³½Ä¹°°ú°¡Àå À¯ÀÇÇÑ Â÷À̸¦ º¸¿´´Ù. ¹Ý¸é¿¡ ȯ»ïµ¢±¼Àº ƯÁ¤ ÆÄÀå´ë¸¦ Á¦¿ÜÇÏ°í, RF ºÐ·ù¿¡¼ °¡Àå ³ôÀº¿À·ùÀ²À» º¸¿© ÀÏ°üµÈ °æÇâÀÌ ¹ß°ßµÇÁö ¾Ê¾Ò´Ù. ¿©Å¸ ½Ä¹°°ú ±¸º°µÇ´Â ȯ»ïµ¢±¼ÀÇ ºÐ±¤°ªÀ» µµÃâÇϱâ À§Çؼ´Â 2°³ ÀÌ»ó ÆÄÀåÀÇ ¹êµå ¿¬»êÀ» ÅëÇÑ Áö¼öÀÇ °³¹ßÀÌ ÇÊ¿äÇÒ °ÍÀ¸·Î º¸ÀδÙ. º» ¿¬±¸´Â À§ÇؽĹ° 3Á¾¿¡ ÇÑÁ¤ÇÏ¿© ÁøÇàµÈ ÇÑ°è°¡ ÀÖ´Ù. µû¶ó¼ ÇâÈÄ ½Ä¹°ÀÇ »ýÀ° ½Ã±â¿¡ ¸ÂÃç ´Ù¾çÇÑÁ¾·ùÀÇ ºÐ±¤¶óÀ̺귯¸®¸¦ ±¸ÃàÇÏ°í, Ç×°ø Ãʺб¤ ¿µ»ó¿¡¼ ´ë¸éÀû ÀÚ¿¬À¯»ê¿¡¼ »ýÀ°È¯°æÀÇ ¸ð´ÏÅ͸µ°ú À§ÇؽĹ°ÀÇ ÇÇÇØ Å½ÁöÇϱâ À§ÇÑ °íµµÈ ¿¬±¸°¡ ¼ö¹ÝµÇ¾î¾ß ÇÑ´Ù. |
¿ä¾à2 |
The purpose was to build hyper-spectral libraries to be used for aerial hyper-spectral images detection and classification of Robinia pseudoacacia L., Pueraria lobata (Willd.) Ohwi, Humulus japonicus Siebold & Zucc.., which influence the vegetation landscape and plant growth of large-area natural heritage sites, and the results are as follows. First, when building hyper-spectral libraries, an acquisition technique was presented that takes into account the shape of the measurement object, the material and color of the measurement background, and the measurement distance. In case of acquisition, the front and back sides of the leaves and branches, which are similar to the crown of the plant seen in the aerial hyper-spectral image, are mixed and placed against a black fabric background. In addition, the shooting distance, among a whiteboard for reflection value correction is used, the plant and the equipment, is 10cm was effective, considering equipment specification and external environment condition. Second, through variance analysis and machine learning of the three acquired spectral libraries, it was revealed that they had discriminatory power compared to the spectral libraries of the four comparative plants. As a result of analysis of Robinia pseudoacacia L., Pueraria lobata (Willd.) Ohwi, using RF(Random Forest) and SVM (Support Vector Machine), discrimination was found with a low error rate of 8-12%, suggesting the possibility of using them as a spectral library in the future. In particular, Pueraria lobata (Willd.) Ohwi showed the most significant difference from comparison plants in all wavelength bands. On the other hand, Humulus japonicus Siebold & Zucc.. showed the highest error rate in RF classification, excluding specific wavelength bands, and no consistent trend was found. In order to derive the spectral value of the other vine, which distinguishes it from other plants, it seems necessary to develop an index through band calculation of two or more wavelength. This study has limitations in that it was limited to three species of hazardous plants. Therefore, in the future, various types of spectral libraries should be built according to the growth period of plants, and advanced research should be conducted to monitor the growth environment and detect damage from harmful plants in large-area natural heritage sites using aerial hyper-spectral images. |