Please use this identifier to cite or link to this item:
標題: Study of Classification with SPOT Images-Feng-Shan Upper Stream Watershed as an Example
作者: Hung, Haw-ren
關鍵字: Neural Network classification
Supervised Classification
Unsupervised Classification
出版社: 水土保持學系
摘要: There were many developed cases in the Feng-Shan upper stream watershed during 1988∼1994. With the four stage of 1988(predevelop),1992(developing),1994 and 1999(developed)SPOT images, this study analyzed the land use changes. The conclusion are as following: 1.According to the change detection with SPOT images on land use of Feng-Shan upper stream watershed, it appears that the agricultural use is tend to plant high economical crop. After the traffic transportation was constructed and the immigration has growing up, the Lung-Tan county of the study area is urbanized significantly. 2. After the four images: 1.primary image, 2.the image after topographic effect was corrected,3.the image after the thematic maps were cut off,4.the image after topographic effect and the thematic maps were cut off were classified by maximum likehood method of Supervised classification, we can see that: from the best to the worst accuracy is 4., 2., 3., 1. 3.The result of classification is that Unsupervised classification is the worst, Supervised is better than Unsupervised, and Neural Network classification is the best. Particularly, the more the complex pixcels are included in image, the better effect of Neural Network classification will be.
鳳山溪上游集水區內於1988~1994六年間歷經許多大型坡地開發案。本研究旨在利用1988(開發前)、1992(開發中)、1994及1999年(開發後)四期影像做為大型坡地開發前、中、後之地形、地貌改變之材料,進行土地利用變遷分析。 本研究經以非監督式與監督式分類法分別對四期經過不同先期處理之十六張影像進行分類,並將其中分類結果最好的四張影像再以類神經網路法進行分類,最後以航空照片進行地真核對,比較其準確度後,獲得重要結論如下: 1.以SPOT衛星影像監測鳳山溪上游集水區土地利用變遷之面積變化,顯示本試區農業利用有朝向種植高經濟作物之趨勢。而在交通開發後由於人口陸續移入,試區內之龍潭地區已呈現明顯之都市化現象。 2.以監督式分類法中之最大相似法對四期之四種經過處理之影像先行分類再行比較,發現:經地形效應及主題圖切除後之影像因去除陰影不確定性之干擾並切除大面積已確定之建地區塊,以致分類準確度最高;經地形效應修正後之影像居次;經主題圖切除後之影像又次之;以原始影像逕行監督式分類之準確度則為最低。 3.由類神經網路分類法、監督式分類法及非監督式分類法三法分類之結果相互比較,得知:類神經網路分類法因具備人類思考模式並能判釋混合像元,以致表現最佳;監督式分類法因掌握地文資訊,但無法判釋混合像元,所以居次;非監督式分類法因純粹以像元光譜特性聚集分類,結果最差。尤其含有混合像元愈多之影像,類神經網路分類法愈能表現其優點。
Appears in Collections:水土保持學系



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.