Research On Polarimetric SAR Image Classification Using Multifeatures Combination Of Best Choice

Polarimetric SAR can provide polarization information by measuring four polarization modes,as a result,targets can be described by these features completely.Therefore,the interpretation of polarimetric SAR image is always one of the hot spots in the field of polarimetric SAR.This paper proposes multifeatures combination of best choice of polarimetric SAR image classification method,solving the problem of polarimetric SAR image classification accuracy using single feature and feature selection in the process of classification to reduce data redundancy.And Because the polarimetric SAR images contain not only the polarization features,but also the information of texture and color,adding texture feature and color feature to image classification being a important means and guiding of the direction of polarimetric SAR image classification.These features are extracted from the experimental data:S matrix,C matrix and T matrix,Pauli,SDH,Huynen,Cloude,Van Zyl,Freeman,and SSCM,texture and color.In the process of selecting the features of the classification of polarimetric SAR image,the feature selection is realized by setting up the feature selection criteria and defining the feature selection parameters.The feature selection parameters for each feature is calculated respectively,in order to achieve the purpose of improving the accuracy and reducing the data redundancy.Features are selected and combined by using the multifeatures combination of best choice method of optimal selection.And then by using the single feature and different multiple features respectively,using random forest,SVM,minimum risk Bayes and KNN,this paper does the experiment to classify polarimetric SAR image of experimental area,and analyze the accuracy of the experimental results.As following are the main research conclusions:(1)Using single feature to classify polarimetric SAR image,the accuracy is low,especially measurement dataon four classifiers.The classification accuracy of the measuring data is the lowest,and The classification accuracy of target decomposition has improved,but ignores details;(2)The method of multifeatures combination of best choice has a better effect on classification accuracy and work efficiency.The method can improve the overall classification accuracy of single by 64.11%.Compared with the classification accuracy of all the features,can improve by the highest 8.97%,and running time is shortened by 34 seconds;(3)The method of multifeatures combination of best choicecan improve the classification accuracy of all kinds of ground objects,but in the same classification algorithm,the improvement is different.