Classification Of Upland Crops With Multitemporal Polarimetric SAR Data

Dryland crops include important food crops(corn,wheat,etc.)and typical economic crops(cotton,peanut,etc.),rapid access to Dryland Crop area information can provide important data support for crop yield estimation and food security.In the critical stage of crop growth in North China,frequent cloud and rain often have a great impact on the quality and quantity of optical image acquisition,thus reducing the accuracy and timeliness of crop area monitoring.Because it is not affected by cloud,rain and haze,does not rely on sunlight imaging,and has the advantages of all-weather monitoring,synthetic aperture radar(SAR)is widely used in crop recognition research.Different from optical data,full polarimetric SAR data contains scattering matrix,geometric structure details and dielectric constant information of the target.It is very sensitive to the geometry and height of the surface vegetation scatterer,which can make up for the shortcomings of optical remote sensing,and has unique advantages in crop recognition and monitoring.At present,crop classification based on SAR data still has some shortcomings,such as low classification accuracy,lack of classification machine rationality and single research object.In view of the current actual business needs and the deficiencies of current research,this article mainly conducted the following research:(1)Based on the multi-temporal RADARSAT-2 data,the classification accuracy of two typical dryland crops in Shenzhou District,Hebei Province is compared with support vector machine and random forest classification methods.Combined with the random forest classifier,the scattering features obtained by different decomposition methods are sorted in importance,and the classification time,classification algorithm and scattering features that are most suitable for extracting dryland crops in the area are found,which provides rapid extraction of dryland crop planting area and spatial distribution.Reference;(2)The scattering mechanism of cotton and maize and its reasons were analyzed qualitatively and quantitatively,and the scattering mechanism of crops was determined from the polarization response graph,scattering power,average scattering Angle and entropy,so as to conduct the classification research of upland crops based on the scattering mechanism.The main conclusions are as follows:(1)The classification accuracy of SVM is slightly better than that of random forest(the classification accuracy of random forest is about 2% lower than that of SVM),but SVM needs more classification time,so the random forest classifier is more suitable for the dry land crop area extraction of SAR data;(2)The average scattering angle ?,Shannon entropy,Shannon entropy intensity component,eigenvalue 1,eigenvalue 3,MCSM surface scattering component and T matrix main diagonal element 1 were used on July 14(Maize Jointing Stage and cotton bud stage).Combined with Shannon entropy,Shannon entropy intensity component,eigenvalue 2,T matrix main diagonal element 2 on September 24(early maturity stage of corn and middle heading stage of cotton),using random forest classifier,we can get high classification accuracy(overall classification accuracy is 90.2239%,kappa number is 0.8422).It not only has a high classification accuracy,but also shortens the classification time,reduces the number of images involved in classification,and provides a reference value for SAR data used in remote sensing monitoring of crop area;(3)Only using radarsat-2 image of August 7(late heading stage of corn,early boll stage of cotton),39 features obtained after decomposition are put into random forest classifier,which can make the classification accuracy of two crops reach the highest under single time phase(corn production accuracy is 98.63%,user accuracy is 78.08%;cotton production accuracy is 63.67%,user accuracy is 76.04%;overall classification accuracy is 78.98%,kappa coefficient is 0.6688);(4)Among the four typical scattering types,the whole growth cycle of corn and cotton is dominated by body scattering and surface scattering,and the power of secondary scattering and spiral scattering is always in a low state.Among them,corn shows the body scattering mainly characterized by vertical dipole,while cotton shows the body scattering mainly characterized by horizontal dipole;(5)LAI and plant height have great influence on the scattering mechanism of crops.In the early stage of crop growth,when the plant height and LAI are both in a small state,the two crops are dominated by surface scattering.With the growth of crops,the plant height gradually increases,the leaves gradually flourish,the LAI value becomes higher and higher,and the scattering mechanism of crops gradually becomes complex.The dominant scattering type changes from surface scattering to body scattering,and the body scattering and spiral scattering power of cotton are both in the growth cycle.The work of body scattering and helix scattering greater than that of corn.(6)According to the difference of dihedral Angle scattering,volume scattering and surface scattering power of different ground objects and their variation trend,the overall classification accuracy of the three polarization features selected for the four typical ground objects in the study area can reach 94.47%,and the Kappa coefficient is 0.9143,which can meet the practical application requirements and provide reference for the classification of dryland crops.