多云多雨区耕地非农化识别的SAR协同监测应用研究OA

Monitoring cropland non-agriculturalization is a critical technical approach for ensuring the supply of food and safeguarding food security.Addressing the challenges of acquiring complete optical imagery in regions characterized by frequent cloud cover and rainfall,as well as the limitations of traditional single-source Synthetic Aperture Radar(SAR)imagery-based methods for identifying cropland non-agriculturalization,which suffer from low accuracy,we propose a novel method that integrates multi-source,multi-scale temporal SAR features for the identification of cropland non-agriculturalization patches in arable land.This methodology commences with the extraction of arable land change patches using the temporal coherence map derived from Sentinel-1 SAR imagery.Subsequently,it employs a variety of statistical features from high-resolution COSMO-SkyMed SAR imagery,in conjunction with the XGBoost model,to accurately classify suspected change patches.The efficacy of this method was validated through engineering verification experiments and quantitative qualitative analyses conducted across four distinct districts and counties in Guangdong Province.The identification accuracy in these four experimental areas reached a maximum of 87.95％,with an average precision of 79.04％.These results demonstrate that the proposed research method is effective in identifying cropland non-agriculturalization patches in regions with frequent cloud cover and rainfall,thus facilitating timely monitoring and mitigation of the non-agriculturalization phenomenon in arable land.