Woodland Identification Based On The Combination Of Interferometric And Polarimetric Information Of SAR Data

Woodland has a close relationship with people’s daily life. It plays an important role in air cleaning, water conservation and ecological balance adjustment. Therefore the real-time monitoring of woodland information becomes increasingly significant. Currently, with the breakthrough and development of hardware technology of airborne SAR sensor, more and more new types of airborne SAR data becomes the preferred data source in feature extraction and land cover classification. However, due to the impact of the speckle noise inherent in all of the SAR images, most classification and segmentation methods for optical images cannot provide a good classification result when applied to SAR data. Thus, there is an urgent need to develop an appropriate woodland extraction method for these massive multi-band, multi-polarized airborne SAR data. Under the support of the863program named object-oriented high-credible SAR processing system, this paper focused mainly on the comprehensive utilization of polarimetric and interferometric information for high precision woodland information, the major innovative contents are as follows:1) An improved classification method for full-polarimetric SAR images was carried out in this paper. Firstly, using the H-A-α decomposition method to extract polarized characteristics of P-band full-polarization SAR image. And based on these characteristics, applied H-α/Wishart and H-A-α/Wishart unsupervised classification methods to extract the woodland information. The results showed that no matter which one of the unsupervised method is easily confusing actual land cover features and cause excessive classification. In consideration of the inadequacies of above algorithms, We proposed an improved algorithm:applied a supervised classification based on the complex Wishart distribution after the H-α/Wishart unsupervised classification process. Then using this improved method in P-band polarimetric SAR data classification. The result showed that this improved algorithm can effectively avoid the excessive classification and obtained a better result than two of the unsupervised classification methods.2) For the first time, proposed the idea of using the interferometric elevation map of X-band to optimize the result of woodland identification and realized the effective integration of the polarimetric scattering information and the interferometric information. We used the threshold segmentation technique to extract the woodland information and applied the support vector machine (SVM) classifier directly on the segmentation result. The classification result of X-band interferometric elevation map showed that the internal and edge of the woodland area are all continuous. Then applied the decision-level fusion of the P-band classification outcome with the X-band segmentation result using the Dempster-Shafer evidence theory which obtained the high-precision a nd relia ble information of the woodland region.3) Combined with the research content and requirements of the subject, we compiled C++program. Realized multi-source information decisio n-level fusion based on the D-S evidence theory and make up for the lack of decision-level fusion in most commercial softwares.