基于多源卫星时序数据的当季冬小麦识别技术OA

Crop intercropping and fragmented planting patterns are major factors limiting the accuracy of crop identification using remote sensing.Multi-source satellite time-series data can effectively distinguish target crops from other vegetation by capturing their unique growth characteristics during specific phenological stages.In this study,the spatial distribution of winter wheat was extracted by integrating the EVI derived from Sentinel-2 across various growth stages,along with NDBI,SAVI,FY-3D EVI time series data,and VV,VH and VH/VV time series data from Sentinel-1.Principal component analysis and the random forest algorithm were employed for feature selection and classification.The results showed that the EVI trends of winter wheat during the emergence to green-up stages differed significantly from those of other vegetation.Similarly,VV and VH/VV backscatter features showed clear distinctions after the green-up stage.The overall classification accuracies using time-series data from sowing to wintering,heading and maturity stages were 95.58%,98.41%,and 98.65%,respectively.Data from the sowing-heading period achieved higher accuracy for field roads and boundaries compared to pre-wintering data.The addition of the FY-3D dataset improved the overall identification accuracy by 1.71pp to 4.10pp across different growth stages,while the inclusion of Sentinel-1 data increased accuracy by 0.21pp to 1.66pp.