Abstract: Due to the difficulties to exactly detect the similar rivers, a novel river detection algorithm was proposed based on multi-feature fusion of the remote sensing image. Firstly, the local entropy features, the texture features and corner information features of the simple images were extracted as the feature vectors of the support vector machine (SVM) in order to obtain the decision function. Then the decision function was employed to perform the coarse detection of rivers in tested images. Finally, in order to obtain the complete rivers, the geodesic active contour model was used, which was combined with the results of the coarse detection of rivers. Experimental results demonstrated the proposed algorithm has good performance in the location and connection of rivers, and the rivers can be detected accurately without the background interference.