Abstract: Remote sensing monitoring is an important means to carry out dynamic monitoring of regional soil and water loss. By classifying and comparing remote sensing images in different periods of the same region, the dynamic monitoring of soil and water loss factors such as land-use could be realized. In traditional method, the artificial visual drawing was usually used to obtain the land-use classification, which is time-consuming and inefficient. Taking the remote sensing images in different periods of the same region as the research object, and based on eCognition, a method using multi-scale segmentation and object oriented classification is introduced. Engineering practice has proved that this method can obtain image classification results quickly, and has higher accuracy and efficiency.