SAR画像を利用した分光特性の類似性に起因する土地被覆誤分類の修正

DOI Web Site 被引用文献1件 参考文献26件

書誌事項

タイトル別名
  • Correction of Landcover Mis-classification Caused by the Similarity of Spectral Characteristics Using SAR Image.
  • Correction of landcover mis-classificat

この論文をさがす

抄録

In the landcover classification using optical multi-spectral scanning data some mis-classifications of landcover categories occur due to the similarity of spectral characteristics. A method has been developed to correct the mis-classification of Open Land category into High Density category classified from Satellite Pour l'Observation de la Terre High Resolution Vidicon (SPOT-HRV) image by using roughness information derived from Japan Earth Resources Satellite 1 Synthetic Aperture Radar (JERS-1 SAR) image. Twenty four core clusters were made based on a clustering algorithm as supervised data. After labeling them to 8 landcover categories, HRV image was classified by Nearest Neighbor Classification (NNC) and Maximum Likelihood Classification (MLC) methods. The occurrence of mis-classification is detected by comparing the classified image with color aerial photograph and investigated using error matrix.<BR>In the correction procedure, the pixel position of the High Density category of HRV is transformed into a position of SAR image using Affine transformation. The pixel positions of both images are in the Universal Transverse Mercator (UTM) coordinate system. The SAR pixels surrounding this transformed pixel position play an important part in the correction method and is termed SAR neighbor pixels.<BR>Each digital value of SAR neighbor pixels is logarithmically transformed to classify the pixel to be rough or flat. When the logarithmically transformed value of the SAR neighbor pixels is greater than 59, the pixel is decided to be urbanized pixel. The proportion of the numbers of SAR pixels determined to be urbanized within SAR neighbor pixels is calculated and defined as an urbanized ratio. The mis-classification of HRV image is estimated based on the urbanized ratio. The HRV pixels classified to High Density category are changed to Open Land when the urbanized ratio is less than 0.05.<BR>By comparing the landcover images before and after correction with the information interpreted from aerial photograph, it is seen that the category accuracy of Open Land category classified by NNC and MLC were increased to 92.1% from 66.9% and to 94.6% from 74.5%, respectively. Approximately 80% of Open Land pixels mis-classified into High Density of HRV can be corrected. The results show that the method proposed is useful for corrcting the mis-classification between High Density and Open Land caused by similarity of spectral characteristics.

収録刊行物

被引用文献 (1)*注記

もっと見る

参考文献 (26)*注記

もっと見る

詳細情報 詳細情報について

問題の指摘

ページトップへ