Cloud cover is the main factor affecting the quality of remote sensing image. Cloud detection for remote sensing images is one of the principal problems that must be solved in remote sensing data restoration processing. On the basis of extensive investigation of existing articles, the research status of cloud detection is analyzed, and then a classification and comprehensive overview of cloud detection methods is presented, the cloud detection methods for several kinds of commonly used satellite data are also given. By comparing the cloud detection methods, the existing problems and development trend of cloud detection method are discussed.
The high precision matching of satellite images has been a problem of much concern. In this paper, the panchromatic images in the same track of GF-1 satellite PMS sensor are treated as investigated subjects. Based on the time-series statistical results of the offset property of the conjugate points, a self-adaptive image matching method is proposed considering mosaic imaging characteristics. In this method, the self-adaptive algorithm is added in the traditional image matching process. It can realize the self-adaptive selection of the target search window, search range and search direction using the iterative computation. The experiment results prove that the proposed method can achieve high precision self-adaptive matching of the GF-1 panchromatic images in the same track, which is useful for other similar satellites in future.
An improved RANSAC algorithm was proposed for point cloud segmentation and geometric primitives extraction of buildings with multiple facets and complex roof structures, including two innovations. Firstly, the “split-segment” strategy combined with regional growth concept is proposed to improve the segment result and efficiency of classic RANSAC algorithm; Secondly, an improved RANSAC algorithm with variant consensus set threshold is presented. By automatically adjusting the consensus set threshold value, geometric primitives with scale difference are likely to meet the validity test, thus avoiding the over-segmentation and under- segmentation problems of classic RANSAC algorithm with fixed consensus set threshold.
Many mainstream segmentation algorithms for high resolution remote sensing image (HRI)rarely consider the segmentation quality in their region merging process. In order to solve this problem, this paper proposed a strategy to optimize heuristics with the purpose of enhancing segmentation accuracy of HRI captured over agricultural areas. Intra- and inter- region homogeneity models were firstly proposed, with the former constructed upon within-region spectral variance, and the latter considering edge strength extracted from multi-spectral and vegetation information. The criterion of the proposed heuristics was then constructed by combining the intra- and inter- region homogeneity. The new criterion enables the merging process to take into account the segmentation quality, thus constraining over- and under- segmentation errors effectively. Two scenes of HRI acquired over agricultural areas were utilized for validation experiment, and the performance of the proposed method was compared with other two newly proposed methods. By analyzing the quantitative evaluation of the segmentation results, it is found that the proposed method can remarkably improve the segmentation accuracy of HRI in agricultural landscape.