耕地"非粮化"遥感提取研究OA

In order to improve the extraction efficiency and accuracy of"non-grain"map spots of cultivated land,Fengnan District of Tangshan City was taken as an example.Based on Google Earth Engine(GEE)platform,cultivated land data,crop data sets,crop phenological characteristics of 2019-2022 and crop time series detector were used to collect crop sample points.Sentinel-2 and Sentinel-1 long time series remote sensing data were used to construct crop classification feature set,and random forest algorithm was used to classify crops and extract non-grain crop spots.The research findings were as follows:in crop classification,by analyzing the difference of NDVI and MNDWI time series of different types of crops,verifying random sample points,extracting sample sets for classification,we finally obtained crop classification results with high precision.During the study period,the overall accuracy of crop classification results was greater than 91%,Kappa coefficient was greater than 0.87,and the relative error between the extracted non-grain crop area and statistical data in each year was less than 1.5%,indicating high extraction accuracy of non-grain crops.During the study period,non-grain crops were widely planted in western towns of Fengnan District and accounted for a relatively high proportion.For the historical years in which field sample points cannot be obtained,the research method has good applicability in extracting the planting range of non-grain crops,and the extraction accuracy is high,which can provide method reference for the supervision of cultivated land"non-grain".