Red Tide Detection Method Based on Improved U-Net Model-Taking GOCI Data in East China Sea as an Example.

Area of Science:

• Marine Biology
• Remote Sensing
• Artificial Intelligence

Background:

• Eutrophication in China's coastal waters causes frequent red tides, damaging fisheries and aquatic resources.
• Accurate red tide detection and prediction are crucial for marine ecosystem management.
• Existing methods struggle with detecting red tide edges that have complex boundaries.

Purpose of the Study:

• To develop an improved deep learning model for accurate red tide detection, particularly focusing on complex boundaries.
• To enhance the separability between red tides and seawater using spectral and spatial features.
• To improve the overall accuracy and reliability of red tide detection using optical remote sensing data.

Main Methods:

• An improved U-Net model incorporating NDVI for enhanced red tide characteristic information.
• Integration of the ECA channel attention mechanism to optimize spectral band weighting.
• Introduction of a shallow feature extraction module with Atrous Spatial Pyramid Convolution (ASPC) for multi-scale feature fusion.

Main Results:

• The proposed method achieved high detection accuracy (95.90%), precision (97.15%), recall (91.53%), and F1-score (0.94).
• Successfully detected red tide edges with complex boundaries, outperforming existing methods.
• Demonstrated good applicability and effectiveness in different regional red tide detection experiments.

Conclusions:

• The improved U-Net model effectively addresses the challenge of complex red tide boundaries in detection.
• The integration of NDVI, ECA attention, and ASPC significantly enhances red tide feature extraction and detection accuracy.
• This approach offers a robust solution for large-scale, high-precision red tide monitoring in marine environments.