Research Of Cloud Removal Algorithm For Landsat Satellite Image

Landsat satellite images contain abundant spectral information, and have been widely used in industrial and agricultural production, resource exploration, environment disaster evaluation and so on. Due to the impact of various climate factors, Landsat satellite images are often covered by clouds. Cloud cover creates "blind area" in the Landsat satellite images, resulting in the lack of remote data and reducing the utilization of Landsat satellite images, which brings difficulties to the image post-processing. Cloud removal of the Landsat satellite images can obtain clear images, which can effectively improve the utilization and accuracy of interpretation of the Landsat satellite images and enhance the effectiveness and availability of the Landsat satellite data.According to the distributions of thin cloud and thick cloud in the Landsat satellite images, this paper proposes several thin cloud and thick cloud removal algorithms. The main works and conclusions are as follows:1. Proposed a thin cloud removal algorithm for Landsat satellite images based on image fusion by using dual tree complex wavelet transform. Firstly, the multi-temporal Landsat satellite images are decomposed by dual tree complex wavelet transform. Then the low frequency fusion coefficients are obtained by combining selecting and weighting regional energy. High frequency fusion coefficients are obtained by using selection strategy based on regional contourlet contrast. Finally, the clear cloud-free images are obtained by reconstructing the fusion coefficients of high frequency and low frequency.2. Proposed a cloud removal algorithm for Landsat satellite images based on multi-dimensional output support vector regression. Support vector value contourlet transform are constructed by the combination of multi-dimensional output support vector regression and directional filter bank, which is used to decompose the Landsat satellite images at multi-scale, multi-direction and multi-resolution. On this basis, the object information(the high frequency details information) covered by thin clouds is grabbed to realize the thin cloud removal. For the thick cloud region, object information in the cloud cover region is predicted by using the multi-source and multi-temporal Landsat satellite images in the same area and the multi-dimensional output support vector regression learning method, which solves the radiation differences among the multi-source images and the overlapping problem of cloud. The cloud-free Landsat satellite images are obtained.3. Proposed a thick cloud removal algorithm for Landsat satellite images by using the similar pixel replacement method. For the Landsat satellite images containing cloud and seasonal changing for object information, the similar pixels and the similar feature region corresponding to the cloud covered region of multi-temporal images are determined by using characteristic parameters which is composed by gray feature, fractal geometry and the sum and difference histograms. The global function based on temporal Markov random field (MRF) is constructed to simulate the most appropriate similar pixels and fill the missing pixels in the target image to achieve thick cloud removal for the Landsat satellite images.