Study And Application Of Dual-polarization SAR Image Classification

Papers on ALOS PALSAR dual-polarization SAR data characteristics and terrain scattering mechanism research, uses the target decomposition method that extract the polarization target decomposition of dual-polarization SAR image classification data processing, in order to improve the polarization SAR image classification precision. Compared with optical data and traditional radar data, polarimetric SAR data not only contains amplitude information, but also have phase information, so we can obtain more polarization information of the ground feature. Polarimetric SAR image record the Back-scattering information of every object under the different polarization combinations states of the resolution units.Because the polarimetric radar has the characteristics that, not under the influence of cloud layer day and night, can penetrate vegetation and the shallow surface, multi-band and multi-polarity, high resolution active image, so polarimetric SAR radar has distinctive advantages in many hands, like target detection and recognition, urban planning and changing, detection of the crop growth, interpretation of geological structure and construction, monitoring and mapping the geological hazard and metal exploration under tectonic belts and so on.Because of the heterogeneity of the object geometrical characteristics, polarimetric SAR receiving echoes has a complex scattering process, in the analysis of polarimetric SAR imaging mechanism, some parameters typifying object property must be extracted from these complex scattering echo, target decomposition method thus emerge as the times require. This article focus on two hands:revealing the scattering mechanism of object characterized by dual polarimetric data and promoting the classification accuracy of dual polarimetric SAR images, by further studied the classification method of dual polarimetric SAR images, achieved the following results:1. Combined with the data characteristic of dual polarimetric SAR, taking statistical property of speckle in images and noise model as its theoretical foundation, the ALOS PALSAR dual polarimetric data was deal with multi look processing firstly. And then the multi looked image was de-noised and contrastive analysed using the filter algorithms of Boxcar filter, Lee-sigma filter and enhanced Lee filter. Experiments have shown that the radiometric resolution of polarimetric SAR images can be raised by multi look processing; and each filter algorithm has the despeckling function. In addition, enhanced Lee filter algorithm is a high-performance and superior method that can strongly suppress speckle and keep image resolution and polarization information.2. Dual polarimetric SAR image classification based on statistical property. On account of the statistical property of polarimetric SAR data, the dual polarimetric SAR image classification was studied with ML and SVM classification algorithm. Experiments have shown that, compared with ML classification algorithm, SVM classification algorithm improved the accuracy of classification results, which validated that the selection of classifier directly impacted the accuracy quality of polarimetric SAR image classification.3. Polarimetric characteristic parameter extraction. By Cloud target decomposed to the coherent matrix of dual polarimetric SAR data, extracted four characteristic parameters reflecting target scattering mechanism, and analyzed different types of scattering mechanisms and physical meanings corresponding to the four parameters, provided available characteristic parameter set to the following polarimetric SAR image classification based on target decomposition.4. Research of dual polarimetric SAR image classification algorithm based on Cloud target decomposition. As the characteristic parameters got after target decomposition have relatively definite physical meanings, there are two advantages that high efficiency and strong feasibility applying target decomposition technology to the research of polarimetric SAR image classification. Combining the characteristic parameters and high-performance SVM classifier, we achieved the classification algorithm of object by dealing with dual polarimetric SAR image. The result have shown that, the accuracy of dual polarimetric SAR image classification based on target decomposition can reach85.41%, so get better classifying effect.