The operation of the Three Gorges Reservoir(TGR) generated a high amplitude of hydro-fluctuation belt(HFB). The preservation and restoration of the HFB had become a major scientific issue after water storage. The classification of bank slopes is the basis for carrying out the protection and restoration of HFB. Taking four typical drinking water sources of the TGR as the research objects, firstly, based on GF-2 remote sensing images covering the study area, and on the basis of radiometric calibration, orthoscopic correction, atmospheric correction, etc., combined with the samples of different bank slope types in the HFB obtained by UAV shooting and visual interpretation, and an object-oriented method for identifying bank slope types in the HFBa was constructed. Secondly, combined with random forest, support vector machine and neural network methods, the classification of bank slope types of typical water sources was carried out, and the classification effect of different machine learning methods was compared to realize the accurate identification of bank slope types in the HFB of typical water sources. Finally, the influence of pixel oriented and object oriented strategies on the classification accuracy of the bank slope in the fall zone was analyzed. The results show that the classification of bank slopes based on multiresolution segmentation-object-oriented classification is a convenient, cost-effective method, and has high accuracy. It can be used for classification of bank slope types in the large-scale HFB of the TGR. This method can solve the problems of internal spectral heterogeneity and increased homogeneity between objects in high-resolution remote sensing images, effectively improving the accuracy of slope classification. The study was of great significance in promoting ecological protection, restoration, and management of the HFB in the TGR, and maintaining important ecological security barriers in the Yangtze River Basin.