Terraces mapping by using deep learning approach from remote sensing images and digital elevation models

Abstract

AbstractTerraces are striking artificial landforms on slopes and are widely distributed in the world. Terraces are vital to soil and water conservation and agricultural production. However, the automatic extraction of terraces entails certain drawbacks, such as low accuracy and poor generalization ability. This study proposes a novel approach to automatically extract terraces from remote sensing images and digital elevation models (DEMs) with high precision. First, terrace samples with annotated images are collected to train the model. Then, three sample areas with varying field conditions in the Loess Plateau are selected as the experimental data to extract the terraces. DEMs are used to eliminate the noise. Subsequently, the visual interpretation results are used to evaluate the accuracy of the extraction results. Furthermore, the proposed approach is compared with the spectral angle mapper approach. Results indicate the advantages of adopting the proposed approach, which is flexible and applicable to complex terrace conditions.