DeepLab-AFOS: A marine oil spill detection method based on advantageous features from polarimetric SAR

Area of Science:

• Remote Sensing
• Marine Ecology
• Environmental Monitoring

Background:

• Marine oil spills pose significant threats to ecosystems and economies.
• Accurate detection and extraction are crucial for effective response.
• Existing methods using only SAR backscattering struggle to differentiate oil spills from look-alikes.

Purpose of the Study:

• To address the challenge of distinguishing oil spills from look-alikes in marine environments.
• To develop an improved method for marine oil spill detection and extraction using polarimetric SAR data.

Main Methods:

• Constructed an advantageous features dual-polarimetric SAR dataset (AF-OS) integrating covariance matrices, polarimetric parameters, and textural descriptors.
• Proposed a polarimetric SAR oil spill detection method named DeepLab-AFOS.
• Conducted comparative experiments to evaluate feature combinations and model performance.

Main Results:

• The AF-OS dataset features significantly enhance oil spill extraction accuracy.
• DeepLab-AFOS achieved superior performance over other models, with 97.2% accuracy, 95.8% F1 score, and 92.06% MIoU.
• Applied DeepLab-AFOS to a real-world event in the Gulf of Mexico, showing accurate identification and differentiation from look-alikes.

Conclusions:

• The AF-OS dataset and DeepLab-AFOS method offer effective solutions for accurate marine oil spill extraction.
• The integrated multidimensional features are key to overcoming the limitations of traditional SAR backscattering analysis.
• This approach provides a robust tool for marine oil spill monitoring and emergency response.