As satellite imaging technology develops rapidly, high spatial resolution remote sensing data in China and other countries are becoming more abundant, which plays an important role in the fields of national economic construction, resource and environmental protection, etc. Due to improvement of spatial resolution, the details of surface features are clearly distinguishable in high-resolution remote sensing images; however, it also brings highly heterogeneous challenges to automatic recognition of surface features. It is a key difficult point in remote sensing image processing to extract location, boundary, and type of surface features automatically and accurately in high-resolution remote sensing images. In view of the high spatial heterogeneity of surface features in high-resolution remote sensing images, Object-based image analysis (OBIA for short) paradigm has become the main means used for extracting conventional high-resolution remote sensing information [1]. Based on its basic technical framework, it obtains segmented objects through segmentation of remote sensing image, and then uses machine learning or artificial intelligence technology to classify or recognize segmented objects to obtain attribute information such as surface feature type. Remote sensing image segmentation aims to divide the image into homogeneous regions based on spatial continuity, and its core idea lies in achieving spatial aggregation or division based on the similarity or discontinuity between adjacent pixel representations. Generally speaking, remote sensing image segmentation only refers to geometric partition, and the segmented region does not contain surface feature type information, so, in the object-oriented analysis framework, it needs to classify or recognize the segmented objects to obtain surface feature type information.