Flood Detection in Gaofen-3 SAR Images via Fully Convolutional Networks

Area of Science:

• Earth Observation
• Remote Sensing
• Artificial Intelligence

Background:

• Effective flood monitoring is crucial for emergency response and rescue operations.
• Accurate and timely detection of flooded areas is a primary challenge in disaster management.
• Synthetic Aperture Radar (SAR) imagery offers consistent data acquisition for flood mapping.

Purpose of the Study:

• To develop a fast and novel flood detection method for emergency monitoring.
• To apply the method to Gaofen-3 SAR images for practical flood mapping.
• To improve accuracy and reduce computational time compared to existing techniques.

Main Methods:

• Utilized a fully convolutional network (FCN), a variant of the VGG16 architecture.
• Fine-tuned the FCN model specifically for flood detection requirements.
• Employed Gaofen-3 SAR images as the data source for flood mapping.

Main Results:

• Achieved higher accuracy in flood detection with reduced training time and fewer samples.
• Demonstrated robust and accurate flood mapping results.
• Significantly decreased detection time compared to state-of-the-art methods.

Conclusions:

• The proposed fine-tuned FCN method provides an efficient and accurate solution for emergency flood detection.
• This approach enhances the capability for rapid flood mapping using SAR imagery.
• The algorithm offers a valuable tool for improving disaster response and rescue efforts.