森林資源調查往往需要投入大量的人力與物力,利用遙測技術節省調查成本,已成為森林資源調查研究之重要議題。ADS40數值多光譜航照影像,除了具有高空間解析力外,亦提供多光譜資訊及高輻射解析力,其應用性相當廣泛。本研究以ADS40數值多光譜航照影像為研究材料,參照第四次森林資源調查之土地利用分類。為消除高解析力影像在進行分類時常產生之椒鹽效應,同時縮減影像運算之資料量,本研究採用物件導向分析法處理影像資訊,於影像切割結果中發現,切割尺度400之影像切割成果較佳;而比較逐像元影像與物件影像之分類成果顯示,以物件影像之分類成果較佳且透過影像物件化處理能有效消除椒鹽效應。本研究另將各幅影像資訊進行影像疊合,並利用最大概似法進行分類以比較各組疊合影像之分類準確度,其中以依照Duncan’s統計分析法所選取之多尺度疊合影像之精確度較高為54.82%,Kappa值0.4997。本研究另採用知識庫分類法,並以分類精度之Kappa值比較兩種分類法則對於ADS40數值多光譜航照影像之適用性。影像分類結果顯示,以知識庫分類法總體精確度78.20%,Kappa值0.7597之分類成果優於最大概似法總體精確度。
Traditional forest inventory have to put in a lot of manpower and material resources to address weaknesses at the lowest cost, and new forest inventory methods based on remote sensing that have been an important issue. ADS40 airborne multispectral images are high-resolution aerial photographs that have been widely applied in forest inventory. In this study, we used the ADS40 airborne multispectral images to classify the land-use types of the Fourth Survey Forest Resource project. The use of High-Resolution Remote-Sensing (HRRS) imagery of classification is liable to cause heavy Salt-and-Pepper Noises; therefore, I chose object-oriented classification methods for ADS40 images to reduce material quantity of imageodesy. The results showed that the image segmentation scale 400 type was better than other image segmentation scales; hence image classification results of accuracy were used to compare pixel-base and object-base image classifications. The analysis determined that object-base image classification was better than pixel-base image classification. Each image information type was collected and used as the image of the superimposed images, and Maximum Likelihood Classification was used to compare different superimposed images of overall classification accuracy. The image group was elected of using Duncan statistical analysis that the result has 54.82% of overall classification accuracy, 0.4997 of overall kappa, and Kappa values were used to compare the two classification methods of Maximum Likelihood and the Knowledge-Based Classification to applicability of ADS40 images. In conclusion, Knowledge-Based Classification showed that the result had 78.20% of overall classification accuracy and 0.7597 of overall kappa statistics better than results of maximum likelihood classification.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。