An approach of the multi-scale texture classification for urban land cover /use using high-spatial resolution satellite imagery was proposed in this paper, in which the decision tree classifier was employed. The comparison with the band to be extraction was performed for three images. The grey-level co-occurrence matrix was adopted to calculate texture values of twenty windows. The J-M distance was used to optimize the texture scales for the eight classes of land cover /use. It was founded that maximum J-M distance appears in the window 15×15 for broadleaf-evergreen, conifer, 27×27 for grass land, 47×47 for bare soil, 67×67 for building and water, respectively. The experimental results showed that overall accuracy with multi-scale texture was 81.7% for eight urban types. The comparison with both the single scale texture and original spectrum showed that the overall accuracy of multi-scale texture was higher than ~6% of the single scale texture and ~11% of the original spectrum respectively. The results also indicate that multi-scale texture method is more accurate and reasonable with real world, and can reduce the "salt-and-pepper" effect. This is achieved by the proposed method, in which the classification with optimization the texture scales is of the most critical value for mapping urban land cover/use using high spatial resolution satellite image.

Date Published: 26 July 2007
PDF: 10 pages
Proc. SPIE 6752, Geoinformatics 2007: Remotely Sensed Data and Information, 67523G (26 July 2007); doi: 10.1117/12.761232

Published in SPIE Proceedings Vol. 6752:
Geoinformatics 2007: Remotely Sensed Data and Information
Weimin Ju; Shuhe Zhao, Editor(s)