Extraction of Settlements in Plains and Mountainous Areas Based on High-Resolution Remote Sensing Data and SAM Segmentation Algorithm

The implementation of the revitalization strategy has led to significant changes with the backdrop of rapid socioeconomic development in the spatial distribution of settlements. The precise and effective identification of settlements has become increasingly vital. Remote sensing research on settlements extraction has predominantly centered on flat plains and areas undergoing rapid economic growth. This study based on multi-source remote sensing images from google earth engine (GEE) to explore parcel segmentation using the segment anything model (SAM) and the Canny segmentation algorithm across various terrains. Random forest (RF), classification and regression trees (CART), and support vector machine (SVM) techniques were used for settlements extraction. Results are as follows: (1) the extraction accuracy of settlements in mountainous areas is more stable compared to plains when using different segmentation and classification methods. However, in the plains, the extraction accuracy of settlements using the SAM based RF (overall accuracy of 0.98) method surpasses that of mountainous areas (overall accuracy of 0.95); (2) The area and landscape evolution of settlements in mountainous areas are greater than those in plains. This research provides a solid foundation for the effective implementation of rural revitalization strategies, efficient resource management, and sustainable environmental protection.