Research On Building Extraction Method Based On PolSAR Incoherent Decomposition

With the rapid development of the world economy and the construction of intelligent cities,the information extraction of building objects has important theoretical significance and practical value for the application of land use survey,urban planning and urban change monitoring.PolSAR(polarimetric synthetic Aperture Radar)plays an important role in the field of urban feature extraction because of its characteristics of multi-parameter,multi-channel,multi-polarization and more complete information recording.Extraction and monitoring have become an important research topic in remote sensing image interpretation.Firstly,this paper systematically analyzes the research status of building extraction based on PolSAR image at home and abroad,and proceeds from the principle of PolSAR imaging.The polarization characteristics and polarization parameters of images are extracted and analyzed by scattering matrix second-order statistical matrix and various polarization decomposition methods.Meanwhile the texture features of images are extracted and analyzed based on gray level co-occurrence matrix.In this paper,the evaluation parameters are constructed to quantitatively analyze the ability of each feature to identify the building,and the preliminary feature selection is carried out according to the correlation of each feature,and the feature set of building extraction is constructed.After detailed analysis of each feature,this paper discusses the influence of feature fusion based on principal component transform and weighted feature fusion based on construction-based evaluation parameters on the result of building extraction.The results show that the weighted feature fusion based on evaluation parameters ensures the accuracy and greatly improves the operation efficiency.At the same time,this paper analyzes the effect of K-nearest neighbor algorithm,stochastic forest algorithm,naive Bayes algorithm,linear discriminant analysis algorithm and SVM algorithm in building recognition.SVM algorithm is the best in building extraction based on each feature and feature combination.On this basis,a building extraction algorithm based on compound kernel function and high-level feature is proposed,and the complex kernel function and high-level semantic feature are explained and experimented in detail.The results show that the composite kernel function is more effective than RBF and MLP in building extraction based on single feature or feature set,and the precision of building extraction is greatly improved by the addition of high-level features.Finally,the paper summarizes the main research work and innovation,and puts forward the further research plan.