A new urban river network extraction method and spatial scale analysis

Abstract

In view of the confusing problem of urban river network water and building shadows in hyperspectral images, we analyzed typical shadow and water spectrum in AISA hyperspectral image. On the basis of Normalized Difference Vegetation Index (NDVI), the 588 nm height factor was introduced to constitute an anti-shadow water extraction method (ASWEM). Compared with NDVI extraction results, this method can effectively suppress shadows, especially those cast in buildings, improve water extraction accuracy and reduce water body commission error. The commission error is reduced from 45% to 10.4%, and Kappa coefficient is increased from 0.664 to 0.863. The change of spatial scale has a significant impact on the water extraction results. The lower the image resolution, the more serious the water leakage is, and some small rivers will not be able to extract. However, due to the influence of the mixed pixels, the spectral characteristics of the shadows are weakened to some extent, and the commission error is reduced. As the resolution decreases further, the number and mixing of mixed pixels increases, and the commission error increases.