Water Body Identification from Satellite Images Using a Hybrid Evolutionary Algorithm-Optimized U-Net Framework.

Area of Science:

• Remote Sensing
• Environmental Science
• Computer Vision

Background:

• Automated water body identification from satellite imagery is crucial for environmental monitoring and water resource management.
• Current deep learning methods often require manual hyperparameter tuning, limiting their automation and robustness.
• Complex, multi-scale scenarios present challenges for existing segmentation techniques.

Purpose of the Study:

• To develop a fully automated segmentation framework for water body identification.
• To overcome the limitations of manual hyperparameter tuning in deep learning models.
• To enhance the robustness and automation of remote sensing image analysis.

Main Methods:

• Integration of an enhanced U-Net model with a hybrid evolutionary optimization strategy.
• Development of a fully automated framework requiring no human intervention.
• Application of the framework to public Kaggle and Sentinel-2 datasets.

Main Results:

• Achieved 96.79% Pixel Accuracy and 94.75 F1-Score, outperforming baseline models by over 10%.
• Effectively addressed class imbalance issues inherent in remote sensing data.
• Demonstrated enhanced feature representation capabilities without manual tuning.

Conclusions:

• The proposed framework offers a viable and efficient solution for fully automated remote sensing image analysis.
• Significant potential for applications in large-scale water resource monitoring and disaster management.
• Advances the field of automated environmental monitoring through improved deep learning techniques.