透過您的圖書館登入
IP:3.145.54.199
  • 學位論文

應用支援向量機法於多重解析度影像水稻紋理判釋研究

Application of Support Vector Machine to Interpretation of Rice Texture in Multi-resolution Image

指導教授 : 張國楨
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


水稻為我國主要糧食來源,目前水稻田清查仍是以傳統人力進行現地調查,較不符合時間及經濟效益,現今已有許多航遙測影像可供使用,影像判釋不再侷限於單一影像,本研究旨在探討不同影像間水稻坵塊判釋成效,若航空影像遭遇遮蔽時,他種衛星影像是否能夠替換,亦在探討半自動化模組是否快速正確輸出判釋成果。 本研究使用DMC數位航空影像、Planet Scope衛星影像及Sentinel-2衛星影像,採用支援向量機法進行物件式導向及像元式水稻坵塊判釋分類,由研究成果顯示,物件式導向分類整體精度以DMC數位航空影像的96%為最佳,其次為Planet Scope衛星影像的93.3%,最差為Sentinel-2衛星影像的54%,像元式分類整體精度以Planet Scope衛星影像的91.1%為最佳,其次為DMC數位航空影像的86%,最差為Sentinel-2衛星影像的64%。 3種影像匯入半自動化模組實際運行(含手動設定路徑及相關參數)平均時間,由實驗成果顯示利用半自動化模組與人工操作相比,物件式導向運行時間可節省約8倍左右的時間,像元式運行時間則是大約節省3倍左右的時間。 後續在實務工作上DMC數位航空影像遭遇影像遮蔽問題時,可考慮選用Planet Scope衛星影像,分類方法皆可使用物件式導向及像元式進行分類,Sentinel-2衛星影像則不考慮使用,另可搭配半自動化模組輔助人員進行影像判釋作業,對於整體影像判釋流程可節省時間及減少人為錯誤。

並列摘要


Rice is one of the most important crops in Taiwan. However, the investigations of rice fields are executed manually in an inefficient way. Nowadays, there are various kinds of remote sensing technologies that can be applied on field investigation. This research discusses the results of paddy field identification between different images and describes if aerial photograph can be replaced by other satellite images when the block occurred. Additionally, the results of semi-automatic module are discussed in this research. DMC digital aerial images, Planet Scope satellite images and Sentinel-2 satellite images are used in support vector machine (SVM) for identifying paddy field in object-oriented and pixel-based classifications. According to the result, the precision of object-oriented classification of DMC images is up to 96% followed by Planet Scope satellite images 93.3%. Sentinel-2 satellite images 54% is the third. On the other hand, the best precision of pixel-based classification is 91.1% from Planet Scope satellite images. The second best is 86% from DMC images which followed by 64% from Sentinel-2 satellite images. The experimental result shows that the process of semi-automatic module applied on object-oriented classification is 8 times faster than manual operation while pixel-based classification is 3 times faster (including manual parameter setting time). To solve the blocking issue in DMC digital aerial images, Planet Scope satellite images is the recommended substitute in both object-oriented and pixel-based classifications. However, Sentinel-2 satellite images are not options. Finally, the classification process with semi-automatic module can improve the quality of result and save time considerably.

參考文獻


一、中文碩士論文
王主一(2012)。運用物件導向方法進行崩塌地自動判釋。明新科技大學土木工程與環境資源管理研究所碩士論文。
王怡舜(2018)。多重解析度影像應用於水稻坵塊萃取適用性研究。國立臺灣師範大學地理學系空間資訊碩士在職專班論文。
林埏祥(2012)。以物件導向分析方法進行土地利用/覆蓋分類及其精度評估。明新科技大學土木工程與環境資源管理研究所碩士論文。
莊尚樺(2015)。應用SVM及紋理資訊於UAV影像水稻田判釋之研究。國立中興大學土木工程學系碩士學位論文。

延伸閱讀