都市蔓延趨勢與物件導向結合衛星影像分析之研究-以台北都會區為例

Detecting Urban Sprawl by Employing the Object-Oriented Approach on Satellite Image – Using Taipei Metropolitan Area as An Example

彭治平(Chih-Ping Peng)

指導教授 ： 黃怡碩

朝陽科技大學/理工學院/營建工程系/博士(2017年)

摘要

一個國家經濟持續成長及永續發展的關鍵，在於國內重大公共建設的投資促進經濟發展外，更重要的是縮短城鄉都市空間的差距，改善國民生活品質。一個指標性城市常常會影響國家發展的趨勢，台北都會區是一個緊密發展的城市，人口總數占全國人數30%，人口密度為台灣都會區中最大，且為政治、經濟活動中心，建構台北都會區的空間發展型態相對重要。都市計畫是以人口數為規劃依據，凡都市交通運輸規劃、土地使用規劃、公共設施規劃等，以都市計畫內容所述為原則。
本研究採用1984-2014年都市變遷，以「區」為空間單元尺度，運用都市蔓延指數(SI)判斷各城市蔓延狀態，並配合非都市計畫區土地蔓延指標來檢討，可了解台北都會區蔓延趨勢，但是無法確認蔓延範圍。唯有藉衛星遙測技術則可解決此一問題。以多時期衛星影像，依需要特性做出分類成果，應用於監測都市土地覆蓋物變遷，是一個快速、有效的方法，也是個可靠的輔助工具。
常態化差異植生指標(Normalized Difference Vegetation Index ,NDVI) 可記錄在一個特定時間，綠色植物的分佈。數值高代表有良好綠色植物環境地區，數值低代表有建築物、道路及其他人造構造物的地區。當都市人口遷移產生遷移時，周邊城市自然植被分布將減少，道路、建築物將取而代之，NDVI值會趨近於「0」；依據NDVI 和倂結果，形成三類物件：NDVI 較小的物件(為建築物或人造物所形成)、NDVI 較中等的物件及NDVI 較大的物件(為植物生長狀況良好)。藉著比較不同時期的物件，尤其是NDVI 較小的物件與NDVI 較大的物件，可作為都市變遷偵測的依據。故本研究以NDVI 為分類依據，藉著Landsat分析紅外線熱波段與全光譜影像資訊，採用1984、1993、2004及2014年衛星影像作為分析的主要對象，以總變異數(total variation)為基礎的能量方程式，能轉化成起始層集函數與時間的偏微分方程式，利用迭代的方式與事前設定的水平集值作為劃分NDVI 圖的依據，在劃分的過程中NDVI值差不多的地區將會合併，隨著時間的增加，能量會逐漸達到穩定，一個複雜的NDVI圖就可以簡化成為預設的類別。總變異數計算可以保有原圖的內容而不失去原有組成特性與邊界資料，配合有線差分法，將可以提供一種快速且穩定的最佳分割法。
從分割結果發現，發現台北都會區以原台北行政區域為基準，往西的方向蔓延，以新北市淡水區、三峽區、樹林區；台北市以內湖區較為顯著。與依人口密度數探討蔓延趨勢相吻合。

關鍵字

水平集 ； 影像分割 ； 蔓延指標 ； 都市蔓延 ； NDVI

並列摘要

Aside from investing in national public constructions to promote economic development, minimizing the urban-rural gap to increase living quality is the key to a prosperous nation and sustainable financial growth. In every country, there is a city or cities could influence a country’s development significantly. Taipei Metropolitan Area has the highest population density in Taiwan. Taipei Metropolitan Area alone holds up to thirty percent of the whole country’s population. It is not only the capital of Taiwan but also the center core for business and government. Therefore, efficient use of space in Taipei Metropolitan Area is critical. Urban planning is based on the population of the city; it serves as a guideline for planning out public mass transportation, land utilization, and general constructions.
This paper adopts the urban changes from the year of 1984 to 2014 and uses kilometers as the unit of distance. It utilizes urban Sprawl Index (SI) to determine the sprawling level of different cities. Along with the comparisons of non-urban sprawl indices, the study was able to graph the Taipei Metropolitan area sprawl trend but unable to pinpoint the sprawling area. Only through remote satellite can be solved the issue. With the segmented satellite images, classifying the image with the specified characteristics and apply the results to monitor the changes of land covers is just a few clicks away. It is an effective, efficient, and a reliable tool.
The normalized difference vegetation index, NDVI, is used to record the distributions of vegetation surrounding urban areas within a particular period. The class with low NDVI values corresponds to those regions containing roads, buildings, and other artificial construction, whereas the region with high NDVI values denotes areas that contain vegetation in good health. When the population in a city spreads, the green area around the city decreased and replaced with pavement and buildings to accommodate the increase in population. Thus, the NDVI will be close to zero. By grouping area objects based on the NDVI, the land is categorized into three group: small NDVI objects (developed area with buildings and road); medium NDVI objects and large NDVI (area with green vegetation). By comparing the object of different phases, especially small NDVI objects and large NDVI objects, changes in the urban environment can be located and quantified. Hence, this paper utilizes NDVI classification as well as Landsat to analyze red band of the whole light spectrum in the data in the attempt to discuss and explain the difference among the satellite images of 1984, 1994, 2004 and 2014. Using an energy equation based on total variation, it resolves to a partial differential equation with time and initial level set function. Using iteration and predetermined settings to divide NDVI images, areas with similar NDVI values will be grouped. With time, the energy stabilizes, a complex NDVI image can be simplified into a predetermined class. Total variation calculation can maintain the original data of the image and not lose the edge information. Combined with the finite-difference method, it can create an optimal, fast, and stable divide method.
From the results of the segmentation, the study concluded that the sprawl of Taipei Metropolitan Area originated from the administrative region and spread west. The significant regions of New Taipei City are Tamsui District, Sanxia District, Shulin District and the significant regions of Taipei City is Neihu District. The urban sprawl trend matches the population density spread.

並列關鍵字

Image segmentation ； Sprawl Index ； Urban Sprawl ； Level set ； NDVI

參考文獻

[52]薛立敏、曾喜鵬，「區域人口遷入台北都會區地點選擇之實證研究」，住宅學報，第十一卷第二期，2002。

[53]姜壽浩，遙測與地理資訊系統的整合與應用，講義，2015。

[54]張國楨，「以多時期與PCA+NDVI法改善地物分類之正確性與完整性」，地理研究，第57期，2012。

[01]Artis, D. A., and Carnahan, W. H., “Survey of Emissivity Variability in Thermography of Urban Areas,” Remote Sensing of Environment, Vol. 12, pp. 313-329, 1982.

[02]Bruzzone, L., and Serpico, S. B., “An Iterative Technique for the Detection of Land-cover Transitions in Multitemporal Remote-sensing Images,” IEEE Transactions on Geoscience and Remote Sensing, Vol. 35, No. 4, pp. 858-867, 1997.