Study On Terrain Classification Of Polarimetric SAR Images Based On Multi-feature Fusion

With the deep research of the polarimetric SAR (Synthetic Aperture Radar)image classification, many supervised and unsupervised classification methods have been proposed one after another in recent years. The early polarimetric SAR image classification algorithm is based on its statistical characteristics. After that, the inherent characteristic of the physical scattering mechanism also acts as an additional advantage for providing category judgment information. However, the defect of the latter is that there is no one-to-one correspondence between the features and the scattering mechanism, which, to some extent, will lead to the terrain classification fuzzy. Many literatures have showed that the comprehensive image texture features and polarimetric information can improve the classification accuracy. Based on the background above, this paper starts from the analysis and fusion of polarimetric characteristics and texture characteristics, with the development of new effective algorithm as the main work objective, to do research on the pixel-based classification methods. The main work and achievements are summarized as follows:Analysis of the texture, polarization and other feature extraction methods were carried out, and many characteristics were obtained, from the collection of which multiple features were chosen to conduct serial fusion according to some selection methods and act as the classification algorithm input. Experiments show that, the result of classification obtained by using the after-fusion characteristics is superior to that of combination classification by using single feature or few features.The MSCM (Multi-band Supervised Classification Model Based on the statistical characteristics) classification method was put forward. In this method multiple features were firstly used to structure multiband data, and then different classes of features were clustered in different feature space regions by means of Euclidean distance. Experiments show that, the model can effectively classify the polarimetric SAR data, and has the characteristics of high calculation efficiency.Based on the research and analysis of multiple features fuzzy pattern recognition, the polarized SAR image classification algorithm based on fuzzy theory was realized. In this algorithm after-fusion multidimensional feature vector was used as an input, then the grade of membership of single feature object was determined according to the descending ridge distribution, and finally the category of the object was judged on the principle of maximum degree of membership. Experiments show that, as for the data in the paper, this algorithm can more significantly improve the classification accuracy compared with the commonly used H/A/alpha-Wishart classification algorithm at present.The polarimetric SAR image classification method based on SOM network was achieved. In this method after-fusion feature vector was used as an input, and different responses were made to different input patterns. Experiments show that, this method has high classification accuracy but with low computational efficiency.