Intelligent Extraction of Terracing Using the ASPP ArrU-Net Deep Learning Model for Soil and Water Conservation on the Loess Plateau

Abstract

The prevention and control of soil erosion through soil and water conservation measures is crucial. It is imperative to accurately and quickly extract information on these measures in order to understand how their configuration affects the runoff and sediment yield process. In this investigation, intelligent interpretation algorithms and deep learning semantic segmentation models pertinent to remote sensing imagery were examined and scrutinized. Our objective was to enhance interpretation accuracy and automation by employing an advanced deep learning-based semantic segmentation model for the astute interpretation of high-resolution remote sensing images. Subsequently, an intelligent interpretation algorithm model tailored was developed for terracing measures in high-resolution remote sensing imagery. Focusing on Fenxi County in Shanxi Province as the experimental target, in this research we conducted a comparative analysis between our proposed model and alternative models. The outcomes demonstrated that our refined algorithm model exhibited superior precision. Additionally, in this research we assessed the model’s generalization capability by utilizing Wafangdian City in Liaoning Province as another experimental target and performed a comparative analysis with human interpretation. The findings revealed that our model possesses enhanced generalization ability and can substantially augment interpretation efficiency.