Abstract: It is difficult to distinguish water from shadow (especially the shadows of tall buildings) and dark ground objects in high resolution remote sensing images, especially in GF-2 remote sensing images. This study analyzes the spectral features of typical terrains of the GF-2 remote sensing images through a lot of experiments. A new comprehensive water index method (NCWI) is proposed to enhance water body region information; and then, the improved method of maximum between-class variance (OSTU) combining with the chicken swarm optimization algorithm (CSO) are used to quickly and adaptively determine the optimal segmentation threshold to obtain the final water body region. To demonstrate the effectiveness of the proposed algorithm, the method of NDWI algorithm, the multi-band spectrum-photometric algorithm and the principal component analysis synthesis algorithms are used for comparison in water-body extraction. The confusion matrix and the field sampling are applied as the statistical metric to quantitatively evaluate the performance of the algorithms mentioned above. The verification results indicate that the new method can be used to extract quickly and effectively extract water body information, and the accuracy reached 97.82%, 97.44%, 92.13%, 96.94% respectively.

Key words: GF-2 image, water extraction, new comprehensive water index, OSTU, shadows of tall buildings, chicken swarm optimization