Article Preview
Top1. Introduction
Clouds which exist in optical remote sensing images with high possibility can degrade limiting to their applicability for earth observation. The Enhanced Thematic Mapper Plus (ETM+) land scenes are reported to be about 35% cloud covered by Yongjun Zhang et.al (2019) globally. Ground cover information is degraded by thin clouds or even completely occluded by thick clouds, which remarkably limits further analysis and applications of such images. In particular, the effect of clouds varies according to the thickness. Thin clouds allow part of underlying objects being observed, which are often ambiguous and could be fairly subtle to formulate and solve such cloud associated problems. On the other hand, thick clouds allow no groundcover information being observed, thus solutions are required urgently to overcome such a challenging problem.
However, because of the significant influence of atmospheric density and cloud layer change on remote sensing processes, most of the remotely sensed images encounter different levels of cloud contamination. The attenuation and even loss of some image information caused by cloud not only reduces the quality and utilization of remote sensing data dramatically but also causes the difficulty of the analysis and application of remote sensing images. In order to improve the usability of remote sensing images (Hemalatha et. al., 2017), it is indispensably essential to conduct cloud detection and removal before any task-specific remote sensing analysis.
In recent years, a large number of cloud detection methods have been proposed. For moderate-spatial resolution and low-spectral-resolution sensors like Landsat, many automated cloud detection algorithms have been developed based on a single Landsat image. Y.Shen et.al (2016) proposed an assessment using the measurement that is traditionally done in remote sensing studies is impossible because of the spatiotemporal variability of clouds. An alternative approach is to find a reference image from the same remote sensor, and the image is cloud free. In addition, the image should be of the closest acquisition or near anniversary dates such that the temporal/seasonal variation is minimized.
Since the algorithm is applied to the entire study area, we not only need to assess the algorithm's ability to remove clouds. S. Qiu et.al (2017) proposed Clouds and cloud shadows are a pervasive, dynamic, and unavoidable issue in Landsat images, and their accurate detection is the fundamental basis for analyzing LTS. Many cloud and/or cloud shadow detection algorithms have been proposed in the literature. For cloud detection, most approaches are based on a single-date Landsat image, which rely on physical-rules or machine-learning techniques. Fei Wen et.al (2017) proposed that the inevitable existence of clouds and their shadows in optical remote sensing images, certain ground-cover information is degraded. A two-pass robust principal component analysis (RPCA) framework for cloud removal in the satellite image sequence was used. First, a plain RPCA is applied for initial cloud region detection, followed by a straightforward morphological operation to ensure that the cloud region is completely detected. Subsequently, a discriminative RPCA algorithm is proposed to assign aggressive penalizing weights to the detected cloud pixels to facilitate cloud removal and scene restoration.
The main contributions in this paper are: The proposed concept is formulated using Simple Linear Iterative Clustering (SLIC) for clustering the similar superpixels by Radhakrishna Achanta et.al (2011) and form a Column Stack. The cloud detection and removing is proposed by Fei Wen et.al (2018) which can be formulated using Group sparsity constrained Robust Principal Component Analysis (GRPCA) and Discriminative Robust Principal Component Analysis (DRPCA). The input image sequence of the same area obtained at different times can be misaligned. First, simple linear iterative clustering (SLIC) superpixel segmentation and arranging each image to a column of a matrix are conducted as preprocessing. Then, Group-sparsity constrained RPCA (GRPCA) combined with geometrical transformation proposed by Yongjun Zhang et.al (2019) is applied to detect cloud and shadow regions initially and also generate a well aligned image sequence.