Crop Remote Sensing Classification Based On Multi-source Data In Complex Cropping Structure Area

Crop remote sensing classification is the basis of crop yield estimation and planting area estimation,and it is also an important part of remote sensing technology in agricultural application.The accurate estimation of accurate classification of various crops can be achieved on the crop planting area,spatial distribution and layout,and provides the main input parameters for the crop yield estimation model.At present,the crop classification using optical data technology has been relatively mature,but the classification of the same period of crop growth monitoring optical data still exist in the same spectrum,and because the effects are vulnerable to weather,resulting in the critical period of crop growth often unable to obtain remote sensing data available.Based on this situation,synthetic aperture radar(Synthetic Aperture,Radar,SAR)data become available to the data source,the SAR data can be all-weather,all day long time work;having strong penetrating ability of vegetation,clouds,which can get the surface vegetation information,and can reflect the internal structure and other relevant information,but as a measurement system of the coherent signals of SAR often disturbed by speckle noise,which affects the classification accuracy of ground objects.In view of the SAR data and optical data differences in crop classification that reflects the scene,3 optical data using the crop growth period and 3 Sentinel-1A radar data,taking the typical black soil region of Heilongjiang Province as the study area,combined with the agricultural insurance company insured plots of data to select the best the extraction in the study area,the scope of arable land as a mask for classification;crop classification related research separately on the SAR data and optical data fusion band combination,using the maximum likelihood method,and through the comparative analysis of the real surface covered plots of the data analyzed,and the analysis results;scattering characteristics in different periods of different crops,and validation of different crop classification accuracy.In this paper,the main results are as follows:(1)Sentinel-1A range image processing(GRD).GRD data is a unique data format Sentinel-1A data,the use of ENVI,Toolbox and other software to complete the radiometric calibration of the data,noise filtering,terrain correction and other processing steps.The appropriate filtering method is selected to improve the classification accuracy,which provides a standard data source for the extraction and utilization of crop information from image data.(2)combining with SAR data and optical data can improve crop classification accuracy.The maximum likelihood classification method using multi temporal SAR data and single phase combined optical band data after the total classification accuracy than using single phase optical data to improve the classification accuracy of about 3~13%,increased by 8~18%than the use of radar data classification accuracy alone.The combination of different crop backscattering analysis results can be obtained in different periods of maximum likelihood classification results and backscattering analysis results are basically consistent with the fusion;PCA transform,Gram-Schmidt transform,NNDiffuse transform after the total classification accuracy compared to the optical data gathering slightly improve the classification accuracy and classification accuracy,using the optical data set and Sentinel-1A data in September 13th NNDiffuse transform fusion ratio of the optical data set is increased by 4%.(3)SAR data will significantly enhance crop characteristics.SAR data can be used to improve the accuracy of remote sensing classification causes crop is abundant structural information contained in the data can improve the optical spectrum information,significantly increased the difference between different crops,the field boundary obviously,enhances the separability between crop to crop,enhances the recognition ability of analysis.