應用非監督式分類法於高美濕地雲林莞草生長範圍分析

Analysis of the Growth Area of Bolboschoenus planiculmis at the Gaomei Wetland Using Unsupervised Classification

楊上賢(Shang-Hsien Yang)

指導教授 ： 張憲國

國立交通大學/工學院/土木工程系所/碩士(2019年)

摘要

本研究目的為以非監督式分類法分析高美濕地雲林莞草生長範圍。本研究方法主要步驟為：(1)以Find the Best Distribution tool公開程式的貝氏訊息準則(Bayesian Information Criterion)從內定11種測試分布函數優選雲林莞草、濕沙及水體三類物體樣本的代表分布; (2)再以非監督式分類法(Unsupervised Classification)的大津演算法(Otsu method)決定出適合的最佳灰階門檻值來判斷雲林莞草與非雲林莞草兩類; (3)最後以相對誤差與重疊率兩種誤差指標判定本法決定的雲林莞草範圍與徒步實測範圍及面積，驗證本研究方法的正確性。
若忽略系統誤差，由結果發現本研究在6月至8月的雲林莞草範圍及面積與步測範圍及面積的重疊率有80%之上的一致性，但是在其他月份的辨識度能力，本文與楊(2018)接近，而本方法比楊(2018)有較低的相對誤差，但楊(2018)有較本法高的重疊率。

關鍵字

雲林莞草 ； 高美濕地 ； 非監督式分類法 ； 大津演算法

並列摘要

The purpose of this study is to investigate the growth area of Bolboschoenus planiculmis at the Gaomei wetland using unsupervised classification. The method includes four steps. Those are (1) to choose the best three fitting distribution of three kinds of representative samples, which are Bolboschoenus planiculmis, wet sand and water, from defaulted 11 kinds of distribution function through the Bayesian Information Criterion in the open code, Find the Best Distribution tool;(2) to use the Otsu method to determinate appropriate threshold and then to distinguish the part of Bolboschoenus planiculmis from the parts of other objects; (3) to evaluate the accuracy of the proposed method by comparing the detected the range and growth area of Bolboschoenus planiculmis with in-site foot-measurement ones depending on two criteria, relative errorand relative duplication area.
Neglecting the systematic error, the determined area of Bolboschoenus planiculmis for June to August by the present method has more than 80% relative duplication with that by foot-measurement. the results showed fair agreement. However, for other months the present and Young’s (2018) results have similar detecting capacity. This present method has lower relative errors than Yang’s (2018), but Yang’s method (2018) has higher relative duplication areas than this method.

並列關鍵字

Bolboschoenus planiculmis ； Gaomei Wetland ； Unsupervised Classification ； Otsu method

參考文獻

1. Akaike, H. (1998) Information theory and an extension of the maximum likelihood principle. Selected papers of hirotugu akaike (pp. 199-213). Springer, New York, NY.

2. Belgiu, M., and Drǎguţ, L. (2014) Comparing supervised and unsupervised multiresolution segmentation approaches for extracting buildings from very high resolution imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 96, 67-75.

3. Cavanaugh, J. E. (2009) 171: 290 model selection, lecture VI: the Bayesian information criterion. University Lecture.

4. Govender, M., Chetty, K., Naiken, V., and Bulcock, H. (2008) A comparison of satellite hyperspectral and multispectral remote sensing imagery for improved classification and mapping of vegetation. Water SA, 34(2), 147-154.

5. Hurvich, C. M., and Tsai, C. L. (1989) Regression and time series model selection in small samples. Biometrika, 76(2), 297-307.