Extraction method of abandoned farmland based on cloud computing platform
Quickly and accurately identifying the situation of abandoned farmland in China is important for increasing grain production,and monitoring abandoned farmland via remote sensing is an important means. The farmland in the southwestern mountainous areas is fragmented,and vegetation in abandoned farmland grows rapidly,making it difficult to distinguish abandoned farmland from normal farmland. Traditional methods based on remote sensing to extract abandoned farmland are time-consuming and have computational limitations. To address these issues,this paper utilized the Google Earth Engine (GEE) cloud platform and Sentinel-2 remote sensing images. It employed two machine learning methods,random forest (RF) and support vector machine (SVM) to extract farmland in Yibin City,Sichuan Province. The normalized difference vegetation index (NDVI) threshold segmentation model was used to segment abandoned farmland and draw a distribution map of abandoned farmland in Yibin City from 2018 to 2021. During this period,the abandoned farmland area accounted for 6.37％,5.15％,4.31％,and 3.02％ of the total farmland,respectively,with an overall calibration accuracy of 83.87％. The application of machine learning methods based on cloud computing platforms,combined with NDVI threshold segmentation models,made it possible to extract abandoned farmland on a large scale with fewer field survey samples.
万方数据
维普
