It is a crucial step in ice flood monitoring to retrieve river channels from remote sensing images. However, it is often the case that one image cannot cover the whole flooded channels, and several images are usually of different scale and deformation. Consequently they cannot be put together directly to form the whole flood channel. In order to get information of the whole ice flood sections, river channels fusion must be used. Features of river channels in remote sensing images of different scale are analyzed in this paper at first, and then the transformation model of point coordinates in different images is established. Parameters in the mathematical model are solved using the genetic algorithm, and river channels in different images are fused together. The genetic algorithm is stable and converges fast according to the processing result. This algorithm also can be adapted to other features fusion in remote sensing images.