RURAL SETTLEMENTS SEGMENTATION BASED ON DEEP LEARNING U-NET USING REMOTE SENSING IMAGES

Abstract. Accurate and efficient extraction of rural settlements from high-resolution remote sensing imagery is of paramount importance for rural government management. Unplanned rural settlements are quite common. Understanding the spatial characteristic of these rural settlements is of great importance as it offers indispensable information for land management and decision-making. In this setting, the U-net architecture is proposed in this study for rural settlements differentiation by image segmentation on high-resolution satellite images of rural settlements in Zagora province, Draa-Tafilalet region, Morocco. To predict pixels in remote sensing images representing rural settlements in this province. Image segmentation is conducted using different encoders in the U-net architecture, and the results are compared. Experimental results demonstrate that the proposed method effectively mapped and discriminated rural settlements areas with an overall accuracy of 98%, achieving comparable and improved performance over other traditional rural extraction methods.