Please use this identifier to cite or link to this item:
標題: ERS-1 SAR與Landsat TM組合影像應用於埔里與南投地區之地覆分類
Land Cover Classification with Conbination of ERS-1 SAR and Landsat TM Data over Pu-Li and Nan-Tou Areas
作者: 蘇婉霞
Su, Wan-Hsia
關鍵字: Remote Sensing
出版社: 森林學系
摘要: 在遙測領域中,雷達與傳統光學式感測器各擅勝場。對地表資源應用而言,雷達及 可見光和紅外光相輔相成之效,正變得益形重要。本研究係探討ERS-1 SAR對埔里 山區與南投平地地覆之辨識能力,而最終目的則在瞭解ERS-1 SAR與Landsat TM的組 合影像可否產生協力增效作用,從而明顯改善兩試區地覆分類準確度。首先將各類 地覆對SAR之反射進行統計分析與初步判釋,建立地覆對雷達之反射特性表。然後 再以最大概似分類法分別對組合影像及TM影像分類且評估分類準確度。結果顯示在 山區與平地的組合影像之分類準確度未因SAR C波段之加入而有改善,反而略有下降 。就山區而言,ERS-1 SAR採高俯角設計導致嚴重地形效應,亦即地形坡度主導SAR 之反射,使地覆的類別內變異量大於類別間變異量,導致ERS-1 SAR對山區地覆辨識 困難。就平地而言,ERS-1 SAR無法有效分辨地覆乃因它係單頻、單偏振雷達且C波 段不是地覆辨識之最適頻率。因此,ERS-1 SAR對兩試區地覆辨識能力不佳。雖然如 此,若能針對試區選擇適用之多頻、多偏振之SAR波段組合並慎選俯角使地形效應最 小化,則在與被動遙測資料結合後,極為可能產生協力增效作用,改善分類準確度 。後續研究應探討如何針對試區確認最適用之雷達波段組合。
In remote sensing, each of the radar and the optical sensor has its own advantages over the other. Clearly the synergism of radar data and visible and reflected infrared for land resource applications is becoming of greater practical importance. The objective of this study was to investigate how ERS-1 SAR could effectively discriminate land cover types in Pu-Li mountain area and Nan- Tou flat land area and eventually to determine if the synergism of ERS-1 SAR and TM data could be achieved to improve significantly the classification accuracies of land cover types in these two areas. Statistical analysis and preliminary interpretation of the radar backscatter from different land cover types was conducted to establish a look-up table for surface features and their signatures on the SAR image. Multispectral classifications using the maximum likelihood classifier and accuracy assessment were performed both on the composite images and on the TM images. The results showed that the classification accuracies of the composite images over Pu-Li and Nan-Tou areas were decreased slightly with the addition of SAR C band. In the mountain area, radar backscatter was dominated by topographic slope due to ERS-1 SAR(s high depression angle (small local incidence angle). The within-category variance of each land cover type became greater than among-category variance due to the topographic effects introduced by the high depression angle. This made ERS-1 SAR C band difficult to discriminate land cover types in Pu-Li area. In the flat land area, ERS-1 SAR could not discriminate land cover types effectively, since it operates on a single frequency and on a single polarization mode and this frequency is not optimal for land cover classification. Therefore, ERS-1 SAR has poor discrimination ability for land cover types in these two areas. Nevertheless, there may have a high possibility to improve land cover mapping with the synergistic use of SAR and visible/infrared data if multi-frequency, multi-polarization SAR data with an appropriate depression angle that can minimize topographic effects are used. The follow-up studies will be to investigate how to determine an optimal SAR band combination.
Appears in Collections:森林學系



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