| The extraction of typical line objects (road and coastline) is widely used not only in military reconnaissance but also in civilian use. With the development of remote sensing technology, the information from a single sensor can not fully reflect the characteristics of the target. In order to gain more information, the idea of fusion is be introduced into the extraction of line objects. Therefore, the collaborative use of the information of spatial and spectral characteristics can improve the accuracy of recognition. In this dissertation, panchromatic images, multi-spectral images and SAR images for the data source, the issue of line objects extraction in multi-source remote sensing images is researched and analyzed mainly by the use of new-type wavelet fusion(dual-tree complex wavelets(DT-CWT) and nonsubsampled contourlet transform(NSCT).Firstly, on the basis of research on wavelet, the new-typical wavelets theories is deeply studied, especially in the nature of DT-CWT and NSCT. The nature are analyzed from the multi-direction, shift invariant and anisotropy. A image fusion method based on the combination of DT-CWT and NSCT is presented. This method can make line features more obvious in the fused image, which is in favor of the following process of target extraction.Secondly, this dissertation researches the method of extracting road and coastline object which are based on fusion. In response to the features of different sources, the different pretreatment for the extraction is used. A road extraction method based on the Hough transform and mathematical morphology is presented. This method extracts roads through the " Extraction - Fusion- Aftertreatment " three steps with the primitive roads. This method can increase the precision for the extraction of road. For the coastline, an improved method for classification is used to extract coastline from the fused image. This method can get over shortcomings of classification, make line features more obvious in the image. |
PDF Full Text Request