研究目的
To propose an automatic multi-temporal method for cloud detection without prior knowledge of the reference image, by fitting robustly the pixels of multi-temporal images contaminated by clouds to show the inherent gradual change of the landscape with time instants.
研究成果
The proposed P-norm based regression model effectively detects clouds in multi-temporal optical remote sensing images by creating reference images that account for the landscape's gradual change over time. It outperforms existing methods like Fmask and MSCD in detecting cloud shadows and thin clouds, though it has limitations with extremely thin or low clouds.
研究不足
The method cannot detect cloud components whose pixel values are within the underlying landscape scale, such as extremely thin or low clouds with small intensity variations.
1：Experimental Design and Method Selection:
The method involves fitting robustly the pixels of multi-temporal images contaminated by clouds using a P-norm based regression model to create reference images showing the landscape's gradual change over time. Cloud detection is then performed by thresholding the difference between target and reference images.
2：Sample Selection and Data Sources:
Landsat-8 OLI dataset images near Norfolk, Virginia, USA, from 11 April 2013 to 14 April 2014, were used. Sub-images of 768×1024 pixels were extracted for analysis.
3：List of Experimental Equipment and Materials:
Landsat-8 OLI images, computational tools for image processing and analysis.
4：Experimental Procedures and Operational Workflow:
Images were co-registered, sub-images were cropped for analysis, and the P-norm based noise reduction algorithm was applied. Cloud detection was performed by comparing target images to reference images derived from the regression model.
5：Data Analysis Methods:
The difference between target and reference images was analyzed to detect clouds as outliers, with a threshold set to identify 98% of a Gaussian distribution as inliers.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
