以多重解析度分析研探數值地形模型之特徵萃取

A Study of Feature Extraction in Multi-Resolution Analysis for DTM

賴巨峰(Jyu- Fong Lai)

指導教授 ： 張哲豪

國立臺北科技大學/工程學院/土木與防災研究所/碩士(2010年)

https://doi.org/10.6841/NTUT.2010.00006

摘要

數值地形模型(Digital Terrain Model， DTM)係以數字形式儲存描述地表特徵及空間分佈的數值集合。近年來，獲得高解析度DTM技術已逐漸成熟，使得三維空間之立體景觀模擬可栩栩如生地展現在使用者眼前。然於實務應用層面，當展示大範圍與低複雜度之數值地形時，高解析度與高精度DTM資料不僅增加儲存空間，更降低其處理效率。因此，針對觀看距離，選擇適當解析度之數值地形模型予以呈現，即為多重解析度分析(Multi-Resolution Analysis, MRA)之概念，亦為近年來相當重要之研究議題。
本研究應用希伯特黃轉換（Hilbert-Huang Transform, HHT)中之經驗模態分解法(Empirical Mode Decomposition, EMD)，可將DTM分解成數個IMF分量。本研究提出兩種特徵萃取方法，從IMF分量得到DTM特徵位置，即可作為多重解析度分析之依據。特徵萃取法之一稱為低差異法：以IMF獲得主趨勢面，與原始高程相減後，差異量小者之高程位置為萃取依據。方法之二稱為高振幅法：以IMF中，超過振幅門檻值之位置，定為特徵點高程位置。研究中並結合Surfer 8.0中之交叉驗證法，萃取地形特徵點並與兩種特徵萃取方法進行比對分析。
由實驗結果顯示，三種特徵萃取方法經美國攝影測量及遙感探測學會之地圖精度標準（ASPRS, American Society for Photogrammetry Remote Sensing ）驗證後，皆符合精度要求。當壓縮率介於5％至25％時，三種特徵萃取方法中以高振幅法之高程精度較佳。然而，當壓縮率大於25％時，交叉驗證法之精度將會高於高振幅法。

關鍵字

希伯特黃轉換 ； 經驗模態分解法 ； 特徵萃取 ； 多重解析度分析 ； 數值地形模型

並列摘要

Digital Terrain Model is stored in digital form to describe surface feature and the spatial distribution of Numerical collection. In recent years, to obtain high-resolution DTM technology has gradually matured, makes the stereo visual simulation of three-dimensional space can be show vivid to the user. However, in the practical application , when displaying a wide range and the low complexity of digital terrain model, High-resolution and high accuracy DTM data is not only to increase storage space but also reduce its processing efficiency. Therefore, for the viewing distance, selecting the appropriate resolution of the digital terrain model to display, it is the concept of Multi-Resolution Analysis. This research topic is also very important recently.
This study applied Empirical Mode Decomposition of Hilbert-Huang Transform, DTM can be decomposed into a number of IMF components. This study proposed two kinds of feature extraction method. We obtained DTM feature position from IMF components. It can be as a basis for Multi-resolution Analysis. One of feature extraction method is extraction point at low trend difference method: It is obtain the main trend surface from IMF, after subtracting the original elevation to select the smaller difference of the elevation poistion. Another of feature extraction method is extraction point at high amplitude: according to IMF, the location of more than amplitude threshold is set to feature point. This study combined with cross validation method to compare with the two kinds of extraction methods.
From the experimental results, the three feature extraction methods verified by the map accuracy standards of American Society for Photogrammetry Remote Sensing that are in compliance with accuracy requirements.
When the compression rate is between 5-25%, in the three kinds of feature extraction methods, the best method is extraction point at high amplitude. However, when the compression rate is more than 25%, the accuracy of cross validation will better than the extraction point at high amplitude method.

並列關鍵字

Empirical Mode Decomposition ； Feature extraction ； Multi-Resolution Analysis ； Digital Terrain Model

參考文獻

[37] 曾志豪，應用HHT於軌道結構分析之研究，中原大學土木工程學系，碩士論文，2006

[1] A.Linderhed, “2-D empirical mode decomposition – in the spirit of image compression ” , in Wavelet and Independent Component Analysis Applications IXI , vol. 4738 , pp.1-8 , April 2000.

[3] C. Y. Lo , L. C. Chen , “Canopy Extraction Using Airborne Laser Scanning Data in Forestry Areas ” The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences , Vol. XXXVII , Part B3b ,2008.

[5] H.Hariharan, et al, “Image Fusion and Enhancement via Empirical Mode Decomposition”, Journal of Pattern Recognition Research, vol.12, no.36, pp.18-32, 2006.

[6] J. C. Nunes , Y. Bouaoune , E. Delechelle , O. Niang , and P. Bunel , “Image analysis by bidimensional empirical mode decomposition ” , Image and Vision Computing , vol.21 , pp. 1019-1026 , 2003.