A Deep Learning Method for Near-Real-Time Cloud and Cloud Shadow Segmentation from Gaofen-1 Images

Area of Science:

• Earth Observation
• Artificial Intelligence
• Image Processing

Background:

• Gaofen-1 satellite provides crucial near-real-time Earth observation data for mapping and environmental monitoring.
• Accurate cloud and cloud shadow segmentation is essential for automated processing of satellite imagery.
• Existing methods require validation for effective application in near-real-time scenarios.

Purpose of the Study:

• To propose a robust deep learning-based multiscale segmentation method for cloud and cloud shadow detection.
• To improve the efficiency and accuracy of segmenting clouds and cloud shadows in Gaofen-1 satellite images.
• To validate the proposed method against existing state-of-the-art techniques.

Main Methods:

• Utilized a deep learning approach incorporating residual convolutional layers for spectral-spatial feature extraction.
• Developed a novel loss function for precise extraction of cloud and cloud shadow footprints.
• Implemented a multiscale segmentation strategy for comprehensive image analysis.

Main Results:

• The proposed method achieved improved accuracy in cloud and cloud shadow segmentation compared to existing methods.
• Demonstrated efficient computational cost, suitable for near-real-time processing.
• Experimental results validated the effectiveness on Gaofen-1 satellite imagery.

Conclusions:

• The developed deep learning method offers a more accurate and computationally efficient solution for cloud and cloud shadow segmentation.
• This advancement supports enhanced near-real-time processing of Earth observation data from missions like Gaofen-1.
• The findings contribute to more reliable environmental monitoring and geographical mapping applications.