Supervised Classification Of Polarimetric Interferometric SAR Images

Forests play an important role in carbon storage and carbon cycle in the global ecosystem.Effectively identifying forest stand types and extracting forest parameters are important tasks for remote sensing observation.Optical remote sensing has been widely used in forest resource monitoring and forest parameter acquisition,which can effectively obtain forest parameters and monitor information.However,in climate-changing,cloudy,rainy and foggy areas,it is difficult to obtain full coverage forest information in the short term,and the phenomenon of"same matter and different spectrum" will also appear.Synthetic aperture radar in this regard is a useful complement to optical remote sensing.It is not affected by time,space,climate and other conditions,and can work around the clock and around the clock.Therefore,the application of SAR data to the forest type identification and extraction shows unprecedented advantages.Based on the current status of forest resource monitoring and stand type identification research,this paper uses polarization information and interference information of fully polarized SAR data to classify forest types in Jiangle Forest Farm,Jiangle City,Fujian Province:(1)14 types of poles are used.The target decomposition method is used to process RANDARSAT-2 data to obtain 47 polarization features.The 47 polarization features are combined into a feature set to classify remote sensing images.(2)The 47 polarization features use random forest's Method to evaluate the importance,then sort the importance in ascending order,remove the less important features one by one from low to high,and calculate the confusion matrix respectively to obtain a set of feature sets with the highest overall classification accuracy;(3)at the obtained pole Four types of interference coherence features are added to the normalized feature set to form a new feature set.The remote sensing images are re-classified,the classification accuracy is compared,and the classification results are analyzed.The research results show that:(1)appropriately reducing the classification features can reduce the dimensions of the attribute features,improve the classification speed of the classifier,and improve the classification accuracy;(2)the decomposition components generated by the polarization target decomposition based on the ground object scattering mechanism recognition can achieve remote sensing The classification of images with higher accuracy,and the main diagonal elements of the[T]matrix obtained from the non-coherent target decomposition based on the coherent matrix have a significant effect on the classification effect;(3)The addition of interference coherence can more effectively divide different geographical classes The classification accuracy for different stand types is improved.