A Dual Attention Encoding Network Using Gradient Profile Loss for Oil Spill Detection Based on SAR Images

Area of Science:

• Environmental Science
• Remote Sensing
• Artificial Intelligence

Background:

• Marine oil spills pose significant environmental threats, necessitating improved monitoring techniques.
• Synthetic Aperture Radar (SAR) is crucial for marine surveillance, but accurately identifying oil spills in SAR images is challenging due to noise and blurred boundaries.
• Deep learning offers potential for automated oil spill detection, but requires specialized architectures to handle SAR image complexities.

Purpose of the Study:

• To develop and evaluate a deep learning model for accurate marine oil spill detection and identification from SAR imagery.
• To address the challenges of noise, blurred boundaries, and uneven intensity in SAR images for oil spill monitoring.
• To improve the accuracy of oil spill boundary recognition using a novel loss function.

Main Methods:

• Proposed a Dual Attention Encoding Network (DAENet) with a U-shaped encoder-decoder architecture.
• Integrated a dual attention module to fuse local and global features across different scales.
• Employed a gradient profile (GP) loss function to enhance boundary recognition accuracy.
• Trained and evaluated the network using the Deep-SAR oil spill (SOS) dataset and GaoFen-3 satellite data.

Main Results:

• DAENet achieved a mean Intersection over Union (mIoU) of 86.1% and an F1-score of 90.2% on the SOS dataset.
• On the GaoFen-3 dataset, DAENet demonstrated superior performance with an mIoU of 92.3% and an F1-score of 95.1%.
• The proposed method significantly improved detection and identification accuracy compared to existing approaches.

Conclusions:

• The DAENet provides a feasible and effective solution for marine oil spill monitoring using SAR imagery.
• The dual attention mechanism and GP loss function are key contributors to the network's high performance.
• This research advances the capability for timely and accurate detection of marine pollution events.