Land Use And Land Cover Change Detection By Multi-Polarimetric SAR Images

Land Use and Land Cover Change detection is one of the important applications of multi-polarization SAR images. However, SAR images have complex electromagnetic properties and are seriously affected by speckle noise, which make the application of SAR images particularly difficult. In this dissertation, some major research has been carried out, and new algorithms are presented considering the characteristics of the multi-polarization SAR images and the problems existing in the current change detection algorithms. To focus on multi-polarization SAR image change detection, the polarimetric image fusion, which is one of the key technologies of change detection, was researched first and then the change detection using multi-polarization SAR was implemented at the pixel level and the object level, respectively.The major tasks and contribution of this dissertation are:First, characteristics of polarimetric SAR images are discussed and analyzed. To understand the characteristics of multi-polarization SAR images, the differences between SAR images and optical images were compared fully from three aspects: geometric properties, radiation properties and resolution characteristics.Second, speckle noise usually affects the interpretation and application of SAR images severely. In this dissertation, typical SAR image speckle noise reduction methods have been analyzed, and the performance of these methods has been reviewed by specific experiments. Since no current de-noising methods can perform well in both noise suppression and detail maintenance, this study does not take the image de-noising preprocessing, but reduces speckle noise by the change detection algorithm itself. This has been proven to be a good solution on noise suppression in this dissertation.Third, a set of NSCT based polarimetric SAR image fusion approaches are presented, including image fusion using different polarimetric SAR images in the same band, and image fusion using different polarimetric SAR images in two different bands. These methods perform better than the wavelet approach in terms of integrating and enhancing the details of the original image and improving the ratio of signal to noise.Fourth, a novel Change Detection Algorithm based on Multi-scale Feature Level Image Fusion in the NSCT Domain was proposed. In this approach, the noise impacts were reduced only by NSCT multi-scale decomposition and the usage of the low-frequency sub-band coefficients. Moreover, the multi-scale feature level fusion strategies in this change detection algorithm can not only improve image quality, but also reduce the noise further.Fifth, a novel object-oriented multi-polarization SAR image change detection approach based on single-phase image segmentation was presented. In this method, an image can be segmented with the same polygons of the other image objects, even in the case of the lack of GIS auxiliary data. Moreover, the image objects, which the change parcels extracted during the first process, were segmented again and the changes were further detected. In this way, the change areas were extracted accurately.Sixth, different experiments on polarimetric SAR image fusion and change detection were carried out using airborne or spaceborne multi-polarization SAR images. By these experiments, some good experimental results have been obtained and a series of valuable conclusions have been drawn.