基因演算法進行非監督式分類應用於山崩潛勢分析

CNU IR > College of Sustainable Environment and Recreation > Dept. of Applied Geoinformatics > Dissertations and Theses > Item 310902800/31956

Please use this identifier to cite or link to this item: https://ir.cnu.edu.tw/handle/310902800/31956

Title:	基因演算法進行非監督式分類應用於山崩潛勢分析 Using Genetic Algorithms of Unsupervised Classification for the Landslide Potential Analyzing
Authors:	沈宸緯
Contributors:	應用空間資訊系徐博賢
Keywords:	基因演算法 Genetic Algorithms
Date:	2018
Issue Date:	2019-02-27 16:49:51 (UTC+8)
Abstract:	台灣位於環太平洋地震帶及颱風頻仍生成的西北太平洋地區，以及熱帶與亞熱帶之間。讓台灣長期承受地震與颱風等影響，導致土質鬆動，嚴重甚至會引發山崩、土石流等重大災情的發生，危害人民的生命財產安全。在遙測分析中對地物的分類方法上，雖然已有監督式分類方法計算地物光譜值加以區分，但是礙於監督式分類須配合各地物的光譜灰階平均值，而地物的光譜灰階平均值需數年時間之收集。本研究利用基因演算法進行非監督式分類取代傳統的監督式分類，以節省監督式分類冗長的地物光譜值資料收集時間，結合七種潛勢因子崩塌潛勢圖，並將該潛勢圖畫分成五種等級，低危險、中低危險、中度危險、中高危險、高危險。實驗成果顯示出本研究的基因演算法DBFCMI指標與另外兩種非監督式分類ISODATA與基因演算法DBI指標在整體精度方面提升了0.7%~1%，與2006年該區實際崩塌比對，高崩塌危險地區誤判比例為0.02%。藉由本研究證實基因演算法DBFCMI指標進行的非監督式分類不僅可以節省地物的光譜灰階平均值收集時間，而不降低崩塌潛勢圖的精度。 Taiwan’s geographical location is in the fault zone between the tropics and the subtropics, which has long subjected Taiwan to the effects of earthquakes and typhoons, resulting in loose soils and in serious cases, landslides, earth flows, and other major disasters that jeopardize the lives and property of the people. In terms of the classification of topographic features in remote sensing analysis, despite the availability of the supervised classification techniques for calculating and distinguishing the spectral values of topographic features, supervised classification is hindered by the fact that it requires the conjunctive use of the average of spectral values of topographic features, which may take years to collect. This study adopted the genetic algorithm to carry out non-supervised classification in place of the conventional supervised classification in order to save the time-consuming spectral value data collection of topographic features for supervised classification. Seven potential factors were conjunctively used to create a landslide potential map, which was divided into five classes: low risk, medium-low risk, medium risk, medium-high risk, and high risk. Experimental results show that the research method in this study and two other non-supervised classification methods improved overall accuracy by 0.7%~1%. Compared to the actual landslides in the area in 2006, the ratio of misjudgment of areas highly prone to landslides was 0.02%. Through the non-supervised classification carried out using the empirical genetic algorithm in this study, this method not only saves the average spectral value collection time of topographic features but also does not reduce the accuracy of the landslide potential maps.
Relation:	電子全文公開日期：2018-09-05，學年度：106， 41頁
Appears in Collections:	[Dept. of Applied Geoinformatics] Dissertations and Theses

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	1516	View/Open

Loading...