Coastal Wetland Classification with GF-3 Polarimetric SAR Imagery by Using Object-Oriented Random Forest Algorithm

Area of Science:

• Remote Sensing
• Environmental Monitoring
• Geospatial Analysis

Background:

• Optical imagery is limited by cloud cover for coastal wetland monitoring.
• Synthetic Aperture Radar (SAR) offers a weather-independent alternative.
• Polarimetric SAR (PolSAR) enhances land cover classification by detecting diverse backscattering mechanisms.

Purpose of the Study:

• To classify land cover in coastal wetlands using Gaofen-3 (GF-3) polarimetric SAR imagery.
• To develop and evaluate an object-oriented random forest algorithm for this classification task.
• To identify optimal polarimetric SAR features for improved wetland classification accuracy.

Main Methods:

• Extraction of 16 commonly used SAR features from GF-3 polarimetric SAR imagery.
• Feature selection using random forest (RF) and sequential backward selection (SBS) to identify optimal parameters.
• Application of an object-oriented RF algorithm for land cover classification in Yancheng Coastal Wetlands.

Main Results:

• Key features for wetland classification included Shannon entropy, Span, orientation randomness, and Yamaguchi decomposition-derived scattering mechanisms (volume, double, surface, helix).
• The object-oriented RF approach with optimal features achieved an overall classification accuracy of 92%.
• Successful differentiation of various land cover types within the Yancheng Coastal Wetlands was demonstrated.

Conclusions:

• The proposed object-oriented random forest algorithm effectively classifies coastal wetland land cover using GF-3 polarimetric SAR data.
• Optimal feature selection significantly enhances classification accuracy.
• This methodology provides a robust solution for coastal wetland monitoring where cloud cover is a concern.