Abstract

Remote sensing satellites have demonstrated to be a helpful instrument. Indeed, satellite images have been successfully exploited to deal with several applications including environmental monitoring and prevention of natural disasters. In the last years, the increasing of the availability of very high spatial resolution (VHR) remote sensing images resulted in new potentially relevant applications related to land cover control and environmental management. In particular, optical sensors may suffer from the presence of clouds and/or of shadows. This involves the problem of missing data, which may result in an important problem especially in the case of VHR images. In this thesis, new methodologies of detection and reconstruction of missing data region in VHR images are proposed and applied on areas contaminated by the presence of clouds and/or shadows. In particular, the proposed methodological contributions include: i) a multiresolution inpainting strategy to reconstruct cloud-contaminated images; ii) a new combination of radiometric information and spatial position information in two specific kernels to perform a better reconstruction of cloud-contaminated regions by adopting a support vector regression (SVR) method; iii) the exploitation of compressive sensing theory adopting three different strategies (orthogonal matching pursuit, basis pursuit and a genetic algorithm solution) for the reconstruction of cloud-contaminated images; iv) a complete processing chain which exploits a support vector machine (SVM) classification and morphological filters for the detection and a linear regression for the reconstruction of specific shadow areas; and v) several evaluation criteria capable to assess the reconstructability of shadow areas. All of them are specifically developed to work with VHR images. Experimental results conducted on real data are reported in order to show and confirm the validity of all the proposed methods. They all suggest that, despite the complexity of the problems, it is possible to recover in a good way missing areas obscured by clouds or shadows.