基于特征增强的冬小麦冻害精细空间分布遥感提取方法OA北大核心CSTPCD

How to obtain the large-scale fine frost damage spatial distribution of crop timely and accurately is an urgent problem in agricultural field.Based on the characteristics of winter wheat growth changes after frost damage,we proposed a winter wheat frost damage fine spatial distribution extraction method(WWFDFSDEM)to extract fine spatial distribution from high resolution remote sensing imagery.Selected two high-resolution remote sensing images before and after frost damage as data source,based on the characteristics of normal winter wheat and frost damage winter wheat,the three channels of red,near-red and green and NDVI were selected as the basic features;we made full use of the spatial correlation of pixel-level features to enhance the details of features.On the basis of cross entropy,the loss function was established by adding the feature difference factor within each class and between the classes to enhance the distinguishing ability of the feature.Gaoqing County,Zibo City,Shandong Province was selected as the research area,Gaofen-2 remote sensing image was used as the data source,and Decision Tree,SegNet,RefineNet,ErfNet and UNet were used as the comparison models to carry out the comparison experiment.The accuracy(94.5%),precision(90.8%)and recall(91.3%)of the results generated by WWFDFSDEM were superior to the comparison methods,which proved the effectiveness of the method in extracting the fine spatial distribution of frost damage,it can meet the needs of extracting fine spatial distribution of frost damage in agricultural production management,agricultural insurance and other fields.