Global climate warming and human activities have caused large areas of permafrost degradation and thermal erosion gully in the Tibetan Plateau,seriously affecting the engineering construction and the ecological environment in permafrost regions. In this study,high resolution unmanned aerial vehicle (UAV) images and object - oriented classification approaches were applied to extracting the thermal erosion gullies in Eboling Mountain of Heihe River. Five kinds of object - oriented supervised learning algorithms,namely nearest neighbor,K - nearest neighbor,decision tree,support vector machine (SVM) ,and random forest,were analyzed for the capability and accuracy of the extraction of thermal erosion gullies in detail. The field GPS data were used for evaluating the classification accuracy. The results show that,in the object - oriented image analysis,the segmentation scale parameters have little effect on the extraction of thermal erosion gullies,wheres classification features have a greater impact,so it is important to select the appropriate classification features. The overall accuracies of the five machine learning methods are all over 90%,among which the Kappa coefficient of the SVM is higher than the other four classification methods. This means that SVM is more suitable for the thermal erosion gullies boundary extraction of UAV images in this study. The combination of high resolution UAV images and object - oriented classification methods has broad application prospects in the extraction of the thermal erosion gullies.