Enhancing Cloud Detection in Polar Regions Using Combined Spectral and Textural Features for Landsat 8/9 OLI Imagery

Remote sensing is a cost-effective and efficient method for studying polar regions. However, cloud cover in optical remote sensing images can diminish the data integrity for certain snow/ice applications. Existing cloud detection algorithms are primarily designed for mid and low latitude images, whi...

Full description

Saved in:
Bibliographic Details
Main Authors: Xinran Shen, Teng Li, Chong Liu, Shaoyin Wang, Lei Zheng, Qi Liang, Xiao Cheng, Jiaqi Yao
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11048513/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Remote sensing is a cost-effective and efficient method for studying polar regions. However, cloud cover in optical remote sensing images can diminish the data integrity for certain snow/ice applications. Existing cloud detection algorithms are primarily designed for mid and low latitude images, which poses significant challenges in polar regions due to the similar spectral characteristics of clouds and snow/ice surface. This study proposes a new cloud detection algorithm for Landsat 8/9 OLI/TIRS images, which combines spectral and texture features to more accurately differentiate clouds from snow. Initially, the algorithm employs function of mask (Fmask) for preliminary cloud detection, followed by a block processing strategy that integrates the gray-level co-occurrence matrix to extract local texture features for secondary discrimination, helping to eliminate snow misidentified as clouds. In addition, the algorithm utilizes short-wave infrared (1.57&#x2013;1.65 <inline-formula><tex-math notation="LaTeX">$\mu$</tex-math></inline-formula>m) and cirrus bands (1.36&#x2013;1.38 <inline-formula><tex-math notation="LaTeX">$\mu$</tex-math></inline-formula>m) to extract cirrus clouds and employs morphological closing operations to fill gaps in the cloud mask. The algorithm maintains an average accuracy of approximately 97% for different types of clouds. Tested on the Landsat 8 cloud cover assessment validation data, a cloud mask dataset verified by experts, this algorithm achieves an average detection accuracy of 93%, signifying improvements of 37% and 45% over the automatic cloud cover assessment and Fmask algorithms, respectively. Texture-based methods effectively reduce snow-cloud misclassification but may inadvertently misclassify texturally similar features, highlighting the need for improved discrimination in classification algorithms. In summary, this novel method significantly enhances the efficiency and precision of cloud detection in polar optical remote sensing images. Consequently, it improves the accuracy of other quantitative remote sensing investigations, such as atmosphere correction, albedo estimation, and so on.
ISSN:1939-1404
2151-1535