Research Of Crop Classification And Biomass Retrieval Based On Synthetic Aperture Radar Remote Sensing Images

In recent years,with the rapid growth of the number of optical and radar satellites launched in the world,the use of satellite remote sensing technology to obtain the classification and growth information of crops has gradually become one of the research hotspots in the field of remote sensing,which has been widely used in the field of crop census,monitoring of crop growth,crop yield estimation and so on.The working conditions of the polarized synthetic aperture radar(SAR)are not limited by cloud and rain weather and the amount of light.On the basis of summarizing the domestic and foreign research achievements of synthetic aperture radar(SAR)crop classification and biomass inversion,this paper makes an in-depth exploration of crop related research in changchun,jilin province.The specific research work and innovative achievements are as follows:(1)Compared with single scene data,multi-temporal SAR data considers the temporal variability of surface vegetation and can provide more scattering information.In this study,c-band GF-3 data combined with multi-temporal sentinel-1 SAR data were used as the data source.Longjia town,changchun city,northeast China,was selected as the experimental area,and the typical crops(soybean,rice,maize)in this area were classified.The scattering Angle,entropy and anisotropy(H/α/A)polarization characteristics were extracted by combining GF-3 and sentinel-1 SAR data.The results show that combining the backward scattering coefficient of GF-3 and sentinel-1 and the polarization characteristic information,the optimal classification results can be obtained.The overall classification accuracy reached 86.95%,and the kappa coefficient was 0.833.It is fully proved that multi-source and multi-temporal SAR has obvious advantages in identifying crops.(2)Considering that the optical satellite data has unique spectral remote sensing information,which can supplement the radar scattering information to a certain extent,this paper studies the crop classification combining SAR and optical data.On the basis of c-band sentinel-1 and GF-3 SAR data,red,green,blue and near-red bands of sentinel-2 optical satellites were added as data sources,and the convolutional neural network(CNN)and Visual Geometry Group(VGG)were used as classification methods and the results were compared.The results show that,based on optical and SAR data,the overall classification accuracy of CNN neural network classification is90.62%,kappa coefficient is 0.867.The overall classification accuracy of VGG neural network was 91.66%,and the kappa coefficient was 0.874.It is proved that the combination of the two is more suitable for the extraction of crop category information,which can effectively improve the accuracy of crop classification.(3)T Polarimetric SAR has a significant correlation with the accumulation of maize biomass,which can be used to invert maize biomass in a large range.The traditional water cloud model(WCM)relies on high-precision soil water content data,which has many limitations in practical use.In this study,a semi-empirical calibration of the water cloud model was performed to retrieve biomass.By adding the surface roughness parameters generated by the SAR polarization difference simulation,the expression of soil scattering is improved,and the relationship between maize biomass and radar backscatter coefficient is more accurately expressed.The change detection approach is used to simulate soil moisture content,which reduces measurement errors and input parameters,and finally obtains the weighted fusion result of dual polarization inversion.The results show that,regardless of VV polarization or VH polarization,the calibrated water cloud model inversion is better than the traditional WCM.For the semi-empirical calibrated water cloud model,the root-mean-square error(RMSE)and the coefficient of determination(R~2)are 1.642kg/m~2 and 0.803,respectively,which has higher accuracy.In this paper,synthetic aperture radar and optical data are used as data sources,and the spectral and radar polarization information are combined to complete the classification of some crops in Northeast China.A semi-empirical calibration of the WCM was used to invert the biomass of the extracted maize during the whole growth period,which provided a reference for future crop remote sensing research in northeastern China.