Scholarlink

<p>Clouds and their accompanying shadows pose inevitable challenges for optical satellite remote sensing images. Therefore, the identification of clouds and cloud shadows in satellite images is a crucial preprocessing step. This has become a significant research focus in recent years, particularly due to its importance in time series analysis. Detecting clouds and shadows accurately is a complex task, given the presence of other structures with similar spectral characteristics. While thick clouds are relatively easier to detect, thin clouds present a more challenging scenario. The proposed model, referred to as CSDNet, employs an encoder-decoder network architecture with a bottleneck attention module to enhance the accuracy of cloud and shadow detection. This paper introduces an automatic cloud and shadow detection network, CSDNet, designed to identify clouds and their associated shadows automatically. In this context, the encoder captures hierarchical features from the input satellite images, and the decoder generates pixel-wise segmentation maps. The incorporation of a bottleneck attention module enhances the model's ability to focus on relevant regions and features, particularly those associated with clouds and shadows. The encoder network captures the most prominent features associated with clouds and shadows, while the symmetric decoder employs fractionally strided convolution to rescale the output back to the original input image size. The proposed model aims to improve the robustness and efficiency of cloud and shadow detection compared to existing methods. Experimental results on Landsat satellite dataset showcase the effectiveness of CSDNet, achieving a dice coefficient of 94.89% for cloud detection and 91.22% for shadow detection. Additionally, the method attains a precision of 95.70% in cloud detection and 93.19% in shadow detection. These results highlight the accuracy of CSDNet, outperforming several traditional methods and establishing its efficacy in automatic cloud and shadow detection in satellite imagery.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CSDNet: automatic cloud and shadow detection from satellite images based on encoder decoder network

  • 期刊论文

  • 发表日期：

    2025年4月17日

查看全文
  • S. R. Surya,
  • M. Abdul Rahiman

摘要

Clouds and their accompanying shadows pose inevitable challenges for optical satellite remote sensing images. Therefore, the identification of clouds and cloud shadows in satellite images is a crucial preprocessing step. This has become a significant research focus in recent years, particularly due to its importance in time series analysis. Detecting clouds and shadows accurately is a complex task, given the presence of other structures with similar spectral characteristics. While thick clouds are relatively easier to detect, thin clouds present a more challenging scenario. The proposed model, referred to as CSDNet, employs an encoder-decoder network architecture with a bottleneck attention module to enhance the accuracy of cloud and shadow detection. This paper introduces an automatic cloud and shadow detection network, CSDNet, designed to identify clouds and their associated shadows automatically. In this context, the encoder captures hierarchical features from the input satellite images, and the decoder generates pixel-wise segmentation maps. The incorporation of a bottleneck attention module enhances the model's ability to focus on relevant regions and features, particularly those associated with clouds and shadows. The encoder network captures the most prominent features associated with clouds and shadows, while the symmetric decoder employs fractionally strided convolution to rescale the output back to the original input image size. The proposed model aims to improve the robustness and efficiency of cloud and shadow detection compared to existing methods. Experimental results on Landsat satellite dataset showcase the effectiveness of CSDNet, achieving a dice coefficient of 94.89% for cloud detection and 91.22% for shadow detection. Additionally, the method attains a precision of 95.70% in cloud detection and 93.19% in shadow detection. These results highlight the accuracy of CSDNet, outperforming several traditional methods and establishing its efficacy in automatic cloud and shadow detection in satellite imagery.