Dual-Stream Difference Modeling with Deep-Guided Multiscale Fusion for Mangrove Change Detection

Area of Science:

• Environmental Science
• Remote Sensing
• Coastal Ecology

Background:

• Accurate mangrove change detection is crucial for coastal ecosystem monitoring.
• Tidal disturbances and unstable land-water boundaries pose significant challenges for existing methods.
• Deep learning models struggle to differentiate true changes from tide-induced spectral variations.

Purpose of the Study:

• To develop a novel deep learning model for accurate mangrove change detection.
• To address the limitations of existing methods in handling tidal interferences and boundary complexities.
• To improve the balance between semantic consistency and boundary accuracy in change detection.

Main Methods:

• Proposed DSDGMNet (Dual-Stream Difference Modeling and Deep-Guided Multiscale Fusion).
• Dual-stream difference-driven strategy to reduce tidal interference and enhance sensitivity to structural changes.
• Deep-guided multiscale fusion module for integrating global context with fine boundary details.

Main Results:

• DSDGMNet achieved an F1-score of 71.36% on the GBCNR dataset, outperforming SNUNet (68.87%) and ChangeFormer (66.39%).
• On the WHU-CD dataset, DSDGMNet achieved an F1-score of 91.38%, surpassing DDLNet (89.85%) and ChangeFormer (88.82%).
• Demonstrated superior performance in mangrove change detection within complex intertidal zones.

Conclusions:

• DSDGMNet effectively addresses the challenges of mangrove change detection in intertidal environments.
• The proposed method shows significant improvements over existing deep learning approaches.
• Highlights the potential for enhanced coastal ecosystem monitoring through advanced remote sensing techniques.