Marine Oil Pollution Monitoring Based on a Morphological Attention U-Net Using SAR Images.

Area of Science:

• Remote Sensing
• Marine Pollution Monitoring
• Artificial Intelligence

Background:

• Oil spills pose significant threats to marine ecosystems and require accurate detection methods.
• Synthetic Aperture Radar (SAR) is a crucial tool for oil spill monitoring due to its all-weather, day-and-night imaging capabilities.
• Existing deep learning models often struggle with fragmented or incomplete detection of oil spill areas.

Purpose of the Study:

• To develop an improved deep learning model for more complete and accurate oil spill detection using SAR images.
• To enhance the detection of oil spill areas by reducing fragmentation and holes in the results.
• To improve the generalization ability of the model in detecting oil pollution incidents.

Main Methods:

• Proposed an improved full-scale aggregated MobileUNet (FA-MobileUNet) model.
• Modified the convolutional block attention module (CBAM) with morphological concepts to create a morphological attention module (MAM).
• Applied label smoothing during training to address dataset category imbalance and improve generalization.

Main Results:

• The FA-MobileUNet model achieved a mean intersection over union (mIoU) of 84.55%.
• This represents a 17.15% improvement in detection performance compared to the original U-Net model.
• Experimental validation using oil pollution incidents in Taiwan demonstrated consistent detection extents with official reports.

Conclusions:

• The proposed FA-MobileUNet model with MAM effectively reduces fragments and holes, providing more complete oil spill area detection.
• The model's improved accuracy and generalization capability enhance its suitability for marine oil pollution monitoring.
• The study validates the model's effectiveness in real-world scenarios, offering a valuable tool for environmental protection agencies.