Vinsamlegast notið þetta auðkenni þegar þið vitnið til verksins eða tengið í það: http://hdl.handle.net/1946/15664
Satellites provide very valuable data about the Earth, e.g., for environmental monitoring, weather forecasting, map-making and military intelligence. But satellites are expensive, both to build and operate. Therefore it is important to make the best use of the data obtained from available satellites, e.g., by combining the output from different sensors.
A good example of this is the fusion of multispectral satellite images of low spatial and high spectral resolution with panchromatic images of high spatial and low spectral resolution. This kind of image fusion is called pansharpening and it is the topic of the thesis.
The thesis is comprised of three parts. In the first part, an observational model for the pansharpening process is derived. A new pansharpening method is developed that involves solving an ill-posed inverse problem dictated by the observational model. The solution is based on a convex optimization problem that is regularized by total variation. The performance of the method is evaluated using quantitative quality metrics for pansharpening using two well-known datasets and the results are compared to existing state-of-the-art methods. The proposed method is shown to give excellent results.
In the second part of the thesis, the solution of the inverse problem for pansharpening using sparsity regularization is investigated. These methods exploit the sparsity of the coefficients of multi-scale overcomplete transforms. Methods based on the two main paradigms of sparsity optimization, i.e., the analysis and synthesis formulations, are derived and the two approaches are then compared in a number of experiments.
In the final part, the classification of pansharpened remote sensing imagery is addressed using two kinds of unsupervised classifiers. Images produced using several pansharpening methods are classified and the results compared and analyzed with the complementary nature of the spectral and spatial quality of the pansharpened images in mind. Furthermore, it is investigated how the use of techniques based on mathematical morphology can be used to increase the classification accuracy.