Extracting rivers from high-resolution satellite images has many important implications. At present, most methods are devoted to extracting rivers from the spectral characteristics or texture analysis of rivers. But for the image which is with the phenomenon of the same object with different spectra or different objects with the same spectra, or has serious noise, or is hard to determine the scale of texture analysis, the method based on water spectrum analysis or texture analysis is not very suitable. The rivers in high-resolution satellite images are generally irregular in structure, and it is more likely that the rivers have different spectral features and texture features due to various reasons. However, in some satellite images, rivers may have approximately uniform width over a wide range. In view of such a situation, a river extraction method combining stroke width transform and geometric feature set is proposed innovatively. Firstly, the Canny edge detector is used to extract the edge of the image, and the edge map is used as the input of the stroke width transform algorithm to obtain the stroke width map. Then, the connected pixels are grouped by using the connected component algorithm, and next, the connected components obtained after the grouping are filtered according to the geometric feature set, and finally the remaining connected components experience the process for filling holes. Experiments using the GF-1 satellite images show that the method can suppress the noise well while extracting the target river. At the same time, compared with the Multiplicative Duda operator and the region growing algorithm, the proposed method has obvious advantages in the aspects of extraction effect and algorithm stability.