Multi-source Remote Sensing Data Fusion Approach And Its Application

With the quick development of the spatial information science and the computer, remote sensing is regarded as a complicated scientific technology including multi-source, multi-channels, multi-polarizations, many spatial plans and many years' observation of the earth. Abundance remote sensing data of the same surface or event on the earth can be supplied to the people at the same time by different remote sensors now. These data are consistency and complementarity, but also redundancy. How to make use of these abundance data and extract the information we need from them is one of the problems always puzzling the remote sensing researchers.More and more attention is paid to the multi-source remote sensing data fusion now because it can synthesize advantage of different data by some data fusion algorithms or models and produce more accurate information about the objects we are interest in. But most of the existed data fusion algorithms are based on the fusion of different optical remote sensing data and only less are created for the fusion of different SAR data or SAR data and optical data.Besides the limits of the fusion algorithms, the application of the multi-source remote sensing data mostly focuses on the surface with natural objects but not surface with man-made objects, such as urban areas. But their development and management demand the help of the multi-source remote sensing data fusion very much. With the well analysis of the advantage and limits of present multi-source remote sensing data fusion, this thesis deals with the fusion of multi-source remote sensing data. The algorithms we developed are based on the characteristics of the SAR and optical data and the fusion are carried out with different algorithms. Their application in the objects recognition and extraction in urban areas gives good evaluation and validation for their promising attributes. The major contributes of the thesis are as follows.(1) According to the statistical characteristic of the fully developed speckle in SAR images, a Laplacian pyramid transformation fusion algorithm based on the local conditional information of the data is proposed which provides high quality data fusion results when applied to the fusion of different polarization SAR images. As an application of the fusion data, ideal strong and weak backscattering classes are extracted automatically from it with the Kittler-Illingworth algorithm based on theGeneralized Gaussian model.(2) A hybrid algorithm of the BP-ANN/GA (Back Propagation Artificial Neural Network and Genetic Algorithm) is developed to optimize the initial weights and make fast convergence of the BP-ANN. This algorithm is applied to the classification of urban terrain surfaces with fused data of infrared and SAR images which well demonstrates the classification speed and accuracy of it.(3) In order to automatically detect the main road network in dense urban area from SAR images, fusion classification of the infrared and SAR images is combined with the edge detection of SAR image with "logic and" method. The classification result can effectively reduces the noise in the edge detection, and also well removes possible confusion of no-road objects, i.e. linearly-featured rivers. With a further connection and extension of the road candidates, the main road network is detected.(4) A fast marching level set method is presented for the fusion of the multi-spectral images and SAR image by defining the speed function with the different features of them not only with their gradient information as usual. The fusion between multi-spectral images and SAR images with different spatial resolution and polarization is proved to be great helpful for the combination of the advantages of them and provides good result for the half-automatic road extraction in dense urban area.(5) To overcome the disadvantage of the conventional Dempster-Shafer data fusion algorithm in assigning the conflicts in evidence combination, a Dempster-Shafer data fusion method considering both the certainty of the evidence and the average support of the evidence to different subsets in the assignment framework is proposed. As an example, it is applied to the fusion of multi-temporal SAR images for the automatic change detection of the urban areas and gives a better detection than the conditional Dempster-Shafer method.