利用Curvelt Transform進行農村變遷偵測

Employing the Curvelet Transform to Study the Spatial Change of the Rural Areas in Taiwan

吳協政(Hsieh-Cheng Wu)

指導教授 ： 黃怡碩 ； 林商裕

朝陽科技大學/理工學院/營建工程系/碩士(2013年)

摘要

自民國60年代起，農地具有成本低廉及易取得之優點，因此許多工廠、住宅和交通建設逐漸深入台灣農村。民國90年代農業發展條例修訂後，農地開放自由買賣，探討農村變遷的偵測和統計則形成重要的研究課題。從衛星影像圖擷取空間特徵的方法，是地理資訊研究中值得探討的問題。遙感探測是一種遠距離不直接接觸物體而取得其訊息的探測技術，藉由衛星影像分析顯示出地形空間和資訊結果，以利用成本低、人力少達到較高的資料收集範圍。本研究選取的區域為台南市竹門社區及南投縣南港社區，其衛星拍攝年份為2005年及2010年，將兩張同一地區不同時間之衛星影像圖利用類神經網路分類法與Curvelt Transform方法分析比較。類神經網路分類法會因為選取類別間的像素數值差異小而導致分類有問題，Curvelt Transform方法則可避免此問題，將兩張衛星影像圖轉換成相近像數的數值，藉由相減和濾除植披相近的區域，得到竹門社區植披變好的區域面積為14.5944公頃，變差的區域面積為13.5436公頃；南港社區植披變好的區域面積為15.2748公頃，變差的區域面積為36.5196公頃。

關鍵字

Curvelt Transform ； 類神經網路 ； 遙感探測

並列摘要

Since the 1970’s, the advantages of agricultural land are that they are inexpensive and easy to obtain. Therefore, numerous factories, houses and transport routes are gradually constructed in rural Taiwan. After the revision of the Agricultural Development Act in the 2000’s, agricultural land is open for free trading. Investigating the changes in the detection and statistics of the rural areas has now become an important topic of research. From the method of spatial feature extraction of satellite images, the study of geographic information is an issue worth exploring. Remote sensing is a long-distance, non-physical contact form of detection technology where information can be obtained. Through satellite image analysis, the results of the topographical space and information are shown. Then by taking advantage of low cost and less manpower, a higher data collection range can be attained. This study has selected the areas: Zhumen Community in Tainan City and the Nangang Community in Nantou County. The satellite images taken were from 2005 and 2010. Two satellite images of the same area at different times were compared using Artificial Neural Network Classification and the Curvelt Transform methods. The Artificial Neural Network Classification method will cause classification problems because of the small differences in the category of pixel values during selection. However, the Curvelt Transform method can avoid this problem. Two satellite images are transformed into similar numerical values, then by subtracting and filtering out areas with similar vegetation, the Zhumen Community area with a better vegetation coverage of 14.5944 hectares is obtained, and the deteriorated area has an area of 13.5426 hectares. The Nangang community vegetation with better vegetation has 15.2748 hectares, and the deteriorated area is 36.5196 hectares.

並列關鍵字

Curvelt Transform ； Artificial Neural Network ； Remote sensing

參考文獻

15. 陳繼潘、徐守道、陳世旺，「應用非監督性類神經網路於SPOT衛星 影像分類之研究」，航測及遙測學刊，第二卷，第一期，第1-12 61頁，1997。

22. 謝漢欽、鄭祈全，福山地區SPOT多期影像植生綠度分析，林業試驗所研究報告期刊，第10卷，第1期，p105-p120，1995。

3. 劉守恆，「衛星影像於崩塌地自動分類組合之研究」，碩士論文，國立成功大學地球科學研究所，台南，2002。

4. 黃俊偉，「大範圍地區土地使用分類之研究」，碩士論文，國立中央大學土木工程研究所，桃園，2001。

6. 林家慶，「應用福衛二號時間序列影像監測曾文水庫集水區崩塌地之時空變化」，碩士論文，國立成功大學地球科學系，台南，2007。